Looking back to move forward: - UNSWorks

Looking back to move forward:Retrospective automated analysis of

prostate radiotherapy trials data

Author:Dale Roach

Supervisors:A/Prof Lois Holloway

Dr Michael Jameson

A/Prof Jason Dowling

Prof Martin Ebert

A thesis in fulfilment of the requirementsfor the degree of Doctor of Philosophy in the

Faculty of MedicineUniversity of New South Wales

July 2020

Faculty Web Site URL Here (include http://)

http://www.university.com

Thesis/Dissertation Sheet

Surname/Family Name : Roach

Given Name/s : Dale Anthony

Abbreviation for degree as give in the University calendar : PhD Faculty : Faculty of Medicine

School : SWS Clinical School

Thesis Title : Looking back to move forward: Retrospective automated analysis of prostate radiotherapy trials data

Abstract 350 words maximum: (PLEASE TYPE)

Radiotherapy clinical trials incorporate quality assurance (QA) protocols to improve the efficacy of treatment. Uncertainties in contouring and treatment planning propagate throughout the radiotherapy workflow. Automated methods can reduce these uncertainties, however analysis techniques providing dosimetric insight should be incorporated to ensure assessments of quality are clinically relevant. This thesis investigated and developed advanced analysis techniques for assessing variability in contouring and treatment planning for prostate radiotherapy. An initial study investigated correlations between contouring metrics and simulated treatment outcome for prostate radiotherapy. Measurements of contouring variability between manual and gold standard Clinical Target Volume (CTV) contours were correlated with variations in dosimetry from treatment plans developed using these contours. It was found that volumetric contouring metrics correlated with outcome, while commonly utilised overlap and boundary contouring metrics did not. Inter-observer contouring variability for multiple male pelvic structures was then assessed utilising these metrics. Variability between observers was small for structures routinely contoured within the clinic. However, substantial variation existed for structures of emerging interest in radiotherapy toxicity studies. An atlas was developed for automated contouring, with CTV retrospectively contoured upon the RADAR, CHHiP, and RT01 clinical trial datasets. Vector mappings identified spatial regions where contouring variability significantly increased the risk of treatment failure. These regions primarily occurred near soft-tissue boundaries, such as the bladder and rectum, whereupon it was shown that contouring variability of CTV does impact patient outcome for prostate radiotherapy. Finally, an assessment of automated treatment techniques shared between centres was undertaken. Automated treatment techniques were robust enough to be adapted to meet other centre’s protocols, while generating treatment plans that satisfied local clinical protocols. To improve treatment within clinical trials uncertainties associated with each step of the radiotherapy workflow must be understood. QA protocols incorporating the techniques developed within this thesis, combined with the emerging automated techniques investigated, would improve the statistical significance of future clinical trials while improving the efficacy of patient treatment.

Declaration relating to disposition of project thesis/dissertation I hereby grant to the University of New South Wales or its agents a non-exclusive licence to archive and to make available (including to members of the public) my thesis or dissertation in whole or in part in the University libraries in all forms of media, now or here after known. I acknowledge that I retain all intellectual property rights which subsist in my thesis or dissertation, such as copyright and patent rights, subject to applicable law. I also retain the right to use all or part of my thesis or dissertation in future works (such as articles or books). …………………………………………………………… Signature

……….……………………...…….… Date

The University recognises that there may be exceptional circumstances requiring restrictions on copying or conditions on use. Requests for restriction for a period of up to 2 years can be made when submitting the final copies of your thesis to the UNSW Library. Requests for a longer period of restriction may be considered in exceptional circumstances and require the approval of the Dean of Graduate Research.

ORIGINALITY STATEMENT ‘I hereby declare that this submission is my own work and to the best of my knowledge it contains no materials previously published or written by another person, or substantial proportions of material which have been accepted for the award of any other degree or diploma at UNSW or any other educational institution, except where due acknowledgement is made in the thesis. Any contribution made to the research by others, with whom I have worked at UNSW or elsewhere, is explicitly acknowledged in the thesis. I also declare that the intellectual content of this thesis is the product of my own work, except to the extent that assistance from others in the project's design and conception or in style, presentation and linguistic expression is acknowledged.’ Signed …………………………………………….............. Date ……………………………………………..............

COPYRIGHT STATEMENT ‘I hereby grant the University of New South Wales or its agents a non-exclusive licence to archive and to make available (including to members of the public) my thesis or dissertation in whole or part in the University libraries in all forms of media, now or here after known. I acknowledge that I retain all intellectual property rights which subsist in my thesis or dissertation, such as copyright and patent rights, subject to applicable law. I also retain the right to use all or part of my thesis or dissertation in future works (such as articles or books).’ ‘For any substantial portions of copyright material used in this thesis, written permission for use has been obtained, or the copyright material is removed from the final public version of the thesis.’ Signed ……………………………………………...........................

Date ……………………………………………..............................

AUTHENTICITY STATEMENT ‘I certify that the Library deposit digital copy is a direct equivalent of the final officially approved version of my thesis.’

Signed ……………………………………………...........................

Date ……………………………………………..............................

i

INCLUSION OF PUBLICATIONS STATEMENT

UNSW is supportive of candidates publishing their research results during their candidature as detailed in the UNSW Thesis Examination Procedure. Publications can be used in their thesis in lieu of a Chapter if: • The student contributed greater than 50% of the content in the publication and is the

“primary author”, ie. the student was responsible primarily for the planning, execution and

preparation of the work for publication • The student has approval to include the publication in their thesis in lieu of a Chapter from

their supervisor and Postgraduate Coordinator. • The publication is not subject to any obligations or contractual agreements with a third

party that would constrain its inclusion in the thesis Please indicate whether this thesis contains published material or not.

☐ This thesis contains no publications, either published or submitted for publication (if this box is checked, you may delete all the material on page 2)

☐

Some of the work described in this thesis has been published and it has been documented in the relevant Chapters with acknowledgement (if this box is checked, you may delete all the material on page 2)

☒ This thesis has publications (either published or submitted for publication) incorporated into it in lieu of a chapter and the details are presented below

CANDIDATE’S DECLARATION

I declare that:

• I have complied with the Thesis Examination Procedure

• where I have used a publication in lieu of a Chapter, the listed publication(s) below meet(s) the requirements to be included in the thesis.

Name Dale Roach

Signature Date (dd/mm/yy)

Postgraduate Coordinator’s Declaration (to be filled in where publications are used in lieu of Chapters) I declare that:

• the information below is accurate • where listed publication(s) have been used in lieu of Chapter(s), their use complies

with the Thesis Examination Procedure • the minimum requirements for the format of the thesis have been met.

PGC’s Name

PGC’s Signature Date (dd/mm/yy)

ii

For each publication incorporated into the thesis in lieu of a Chapter, provide all of the requested details and signatures required Details of publication #1: Full title: Correlations between contouring similarity metrics and simulated treatment outcome for prostate radiotherapy Authors: Dale Roach, Michael Jameson, Jason Dowling, Martin Ebert, Peter Greer, Angel Kennedy, Sandie Watt, Lois Holloway Journal or book name: Physics in Medicine & Biology Volume/page numbers: Volume 63/035001 Date accepted/ published: 4 January 2018 Status Published ✓ Accepted and In

press In progress

(submitted)

The Candidate’s Contribution to the Work The candidate developed treatment plans for all patients under the tutelage of Sandie Watt. All contouring similarity and radiobiological metrics were measured by the candidate utilising software developed by Jason Dowling and Lois Holloway respectively. Correlations and the manuscript were completed by the candidate, with feedback provided by co-authors. Location of the work in the thesis and/or how the work is incorporated in the thesis: In lieu of Chapter 3 Primary Supervisor’s Declaration I declare that: • the information above is accurate • this has been discussed with the PGC and it is agreed that this publication can be

included in this thesis in lieu of a Chapter • All of the co-authors of the publication have reviewed the above information and have

agreed to its veracity by signing a ‘Co-Author Authorisation’ form. Supervisor’s name Prof. Lois Holloway

Supervisor’s signature Date (dd/mm/yy)

iii

Details of publication #2: Full title: Multi-observer contouring of male pelvic anatomy: Highly variable agreement across conventional and emerging structures of interest Authors: Dale Roach, Lois Holloway, Michael Jameson, Jason Dowling, Angel Kennedy, Peter Greer, Michele Krawiec, Robba Rai, Jim Denham, Jeremiah De Leon, Karen Lim, Megan Berry, Rohen White, Sean Bydder, Hendrick Tan, Jeremy Croker, Alycea McGrath, John Matthews, Robert J Smeenk, Martin Ebert Journal or book name: Journal of Medical Imaging and Radiation Oncology Volume/page numbers: Volume 63/264-271 Date accepted/ published:27 November 2018 Status Published ✓ Accepted and In

press In progress

(submitted)

The Candidate’s Contribution to the Work The candidate assisted in the development of the contouring protocol, as well as participating by contouring the structures required for the study. The candidate compiled and cleaned all data as required, and performed the assessment of contouring variations between observers. The manuscript was completed by the candidate, with feedback provided by co-authors. Location of the work in the thesis and/or how the work is incorporated in the thesis: In lieu of Chapter 4 Primary Supervisor’s Declaration I declare that: • the information above is accurate • this has been discussed with the PGC and it is agreed that this publication can be




iv

Details of publication #3: Full title: Adapting automated treatment planning configurations across international centres for prostate radiotherapy Authors: Dale Roach, Geert Wortel, Cesar Ochoa, Henrik Jensen, Eugene Damen, Philip Vial, Tomas Janssen, Christian Rønn Hansen Journal or book name: Physics and Imaging in Radiation Oncology Volume/page numbers: Volume 10/7-13 Date accepted/ published: 14 April 2019 Status Published ✓ Accepted and In

press In progress

(submitted)

The Candidate’s Contribution to the Work The candidate participated in the initial AutoPlanning configuration development phase, developing planning techniques that met the other centre’s protocol. These planning techniques were applied to the validation datasets, with the candidate responsible for all planning at Liverpool. All datasets were analysed by the candidate, with assistance from code written by Christian Hansen. The manuscript was completed by the candidate, with feedback provided by co-authors. Location of the work in the thesis and/or how the work is incorporated in the thesis: In lieu of Chapter 5 Primary Supervisor’s Declaration I declare that: • the information above is accurate • this has been discussed with the PGC and it is agreed that this publication can be




Acknowledgements

First and foremost, I want to extend my utmost gratitude to my supervisors A/Prof Lois

Holloway and Dr Michael Jameson. Throughout my candidacy they have consistently

provided the support and drive I required to complete my thesis. It has been an amazing

journey, and I could not have hoped for more wonderful supervisors to guide me along

the way. To Lois, thank you for your kindness and belief in me. To Michael, thank you

for your tutelage and compassion. I can never understate the level of appreciation I feel

towards you both.

To my supervisors A/Prof Jason Dowling and Prof Martin Ebert, I could always rely on

you both for being patient and considerate mentors during my candidacy. Jason, thank you

for your assistance in so many aspects of my work, along with the passionate discussions

about our respective football clubs! Martin, your enthusiasm for the project ensured I

grew passionate about the work I was doing. Thank you both again for being wonderful

supervisors.

To all past and present staff and students at the Ingham Institute that I shared my

journey with, thank-you for sharing such a wonderfully memorable chapter of my life.

The friendships I made have ensured that the journey was enjoyable, filled with laughter

over many coffee breaks. While there are too many to thank individually, mention must

go to Rob Finnegan and Jarryd Buckley. Here’s to the many enlightening discussions and

shared experiences throughout our journeys presenting our respective work around the

globe.

To the staff at Liverpool and Macarthur Cancer Therapy Centres, thank you for the

support that was readily available throughout my candidacy, as well as the tolerance and

assistance afforded me when juggling thesis writing with starting my work as a registrar.

Thank you to Melanie Rennie, Sandie Watt, and Rohan Gray for teaching me the art

of treatment planning. Thank you to Shivani Kumar and Vikneswary Batumalai for the

collaborative work, as well as helping me feel part of the clinical environment.

To my fellow registrars Joshua Hiatt and Iliana Peters, thank you for all the assistance

you both graciously offered while I finalised my candidacy. Sincere thanks also go out to

viii

ix

Vinod Nelson and Tania Ervin, whose support in ensuring this thesis was completed in

a timely fashion was tremendously appreciated. Finally, a large thanks goes out to Amy

Walker, who I have now shared the journey from research to registrar with for nearly five

years. The laughter and kindness will never be forgotten, and I look forward to usurping

you at your desk.

The all the collaborative groups I have been involved with, thank you for the amazing

opportunities I have been presented over these past few years. To Angel Kennedy, Marco

Marcello, and all the other staff and students at Sir Charles Gairdner Hospital, thank you

for making me feel so welcome during my stay. Thank you also to Phil Vial, Christian

Rønn Hansen, and Geert Wortel for involving me in your research. It was an amazing

opportunity to be directly involved with so many centres around the world during my

candidacy.

To my mum and dad, thank you for all your love and support throughout the entire

journey. I know you were both nervous when I said I would be at university for four more

years, but your faith in what I could accomplish was never in doubt. I hope I have made

you both proud. Finally, to my dearest Navami. Never could I have imagined I would find

someone so caring and supportive to stand by me during my journey. The love between

us will never wane, and I dedicate this thesis to the future that we will make together.

List of Publications1. Roach, D., Jameson, M. G., Dowling, J. A., Ebert, M. A., Greer, P. B., Kennedy, A.

M., ... & Holloway, L. C. (2018). Correlations between contouring similarity metrics

and simulated treatment outcome for prostate radiotherapy. Physics in Medicine &

Biology, 63(3), 035001.

2. Kumar, S., Holloway, L., Roach, D., Pogson, E., Veera, J., Batumalai, V., ... &

Moses, D. (2018). The impact of a radiologist‐led workshop on MRI target volume

delineation for radiotherapy. Journal of medical radiation sciences, 65(4), 300-310.

3. Roach, D., Holloway, L. C., Jameson, M. G., Dowling, J. A., Kennedy, A., Greer,

P. B., ... & Ebert, M. A. (2019). Multi‐observer contouring of male pelvic anatomy:

Highly variable agreement across conventional and emerging structures of interest.

Journal of medical imaging and radiation oncology, 63(2), 264-271.

4. Roach, D., Wortel, G., Ochoa, C., Jensen, H. R., Damen, E., Vial, P., ... &

Hansen, C. R. (2019). Adapting automated treatment planning configurations across

international centres for prostate radiotherapy. Physics and Imaging in Radiation

Oncology, 10, 7-13.

5. Kennedy, A., Dowling, J., Greer, P. B., Holloway, L., Jameson, M. G., Roach, D.,

... & Ebert, M. A. (2019). Similarity clustering‐based atlas selection for pelvic CT

image segmentation. Medical physics, 46(5), 2243-2250.

6. Roach, D., Denham, J. W., Joseph, D. J., Gulliford, S. L., Dearnaley, D. P.,

Sydes, M. R., ... & Ebert, M. A. (2020). Spatial analysis of target volume contours

within three international prostate radiotherapy trials - Clinical impact of contouring

variations on patient outcome. Submitted, Under Review.

List of Publications xi

8. Batumalai, V., Burke, S., Roach, D., Lim, K., Dinsdale, G., Jameson, M., ... &

Vinod, S. (2020). Impact of dosimetric differences between computed tomography

(CT) and magnetic resonance imaging (MRI) derived target volumes for external

beam cervical cancer radiotherapy. Submitted, Under Review.

9. Marcello, M., Denham, J.W., Kennedy, A., Haworth, A., Steigler, A., ... , Roach,

D., ... & Ebert, M.A. (2020). Reduced dose posterior to prostate predicts for

increased treatment failure in pooled voxel-wise analysis of 3 randomised phase 3

trials. Submitted, Under Review.


D., ... & Ebert, M.A. (2020). Association of spatial radiotherapy dose distribu-

tion and urinary toxicity across three randomised phase 3 trials. Submitted, Under

Review.


D., ... & Ebert, M.A. (2020). Dose hotspots in central rectum and low-intermediate

doses to perirectal fat space correlate with rectal bleeding and tenesmus respectively

in pooled voxel-wise analysis of 3 randomised phase 3 trials. Submitted, Under

Review.

List of Abbreviations

3D-CRT Three-Dimensional Conformal RadiotherapyADT Androgen Deprivation TherapyAP AutoPlanningaRVD Absolute Relative Volume DifferenceAS Androgen SuppressionASR Age Standardised RateCHHiP Conventional versus Hypofractionated High-dose Intensity-modulated

radiotherapy for Prostate cancerCI Conformity IndexCNN Convolutional Neural NetworkCOV Coefficient of VariationCSIRO Commonwealth Scientific and Industrial Research OrganisationCT Computer TomographyCTV Clinical Target VolumeDCP Distance-to-Agreement Cut-pointDICOM Digital Imaging and Communications in MedicineDIR Deformable Image RegistrationDMPO Direct Machine Parameter OptimisationDNA Deoxynucleic AcidDRE Digital Rectal ExaminationDSC Dice Similarity CoefficientDtA Distance-to-AgreementDVH Dose-Volume HistogramDX Dose Volume LevelEAS External Anal SphincterEBRT External Beam RadiotherapyEORTC European Organisation for Research and Treatment of CancerEPID Electronic Portal Imaging DeviceEUD Equivalent Uniform DoseG0,1,2 Gap cell cycle phasesGTV Gross Tumour VolumeGy GrayHD/HAUS Hausdorff DistanceHDR-BT High Dose-Rate BrachytherapyHI Homogeneity IndexHIFU High-Intensity Focussed UltrasoundHR Hazard RatioIAS Internal Anal SphincterICC Intraclass Correlation CoefficientICRU International Commission on Radiation Units and MeasurementsIGRT Image Guided Radiotherapy

List of Abbreviations xiii

IMRT Intensity Modulated RadiotherapyIsoX Isodose VolumeISUP International Society of Urological PathologyLAM Levator Ani MuscleLDR-BT Low Dose-Rate BrachytherapyLINAC Linear AcceleratorM Mitosis cell cycle phaseMASD Mean Absolute Surface DistanceMeV Mega-electronVoltMLC Multileaf CollimatorMRC Medical Research CouncilMRI Magnetic Resonance ImagingMU Monitor UnitMV MegavoltageNCCN National Comprehensive Cancer NetworkNHMRC National Health and Medical Research CouncilNHS National Health ServiceNTCP Normal Tissue Complication ProbabilityNVB Neurovascular BundleOAR Organs-at-RiskPET Positron Emission TomographyPRM Puborectalis MusclePRO Patient Report OutcomesPRS Peri-rectal SpacePRV Planning Organ-at-Risk VolumePSA Prostate-Specific AntigenPTV Planning Target VolumePV Protocol ViolationsQA Quality AssuranceQARC Quality Assurance Review CentreRADAR Randomised Androgen Deprivation and RadiotherapyRO Radiation OncologistROI Region of InterestRTOG Radiation Therapy Oncology GroupS Synthesis cell cycle phaseSBRT Stereotactic Body RadiotherapySD Standard DeviationSTAPLE Simultaneous Truth and Performance Level EstimationSUAS Setup Accuracy StudySV Seminal VesiclesTCP Tumour Control ProbabilityTMC Trial Management CommitteeTNM Tumour, Nodal, MetastasisTPS Treatment Planning SystemTROG Trans-Tasman Radiation Oncology GroupTRUS Transrectal UltrasoundVMAT Volumetric Modulated Arc TherapyVOLSIM Volume Similarity

List of Figures

1.1 Early variations in quality will be encapsulated by all further steps of theradiotherapy workflow, affecting the overall quality of the clinical trial. Thefour key topics addressed by this thesis focus on the early stages of theradiotherapy workflow, with latter topics directly tied to improving clinicaltrial quality assurance. How each topic relates to the radiotherapy workflowis represented by the shading of the corresponding circles. . . . . . . . . . . 3

2.1 Incidence (solid line) and mortality (dashed line) rates of prostate cancerin Australian men. Image adapted from [1]. . . . . . . . . . . . . . . . . . . 9

2.2 Male pelvic anatomy. Image adapted from [2]. . . . . . . . . . . . . . . . . . 102.3 Zonal anatomy of A) a young, and B) an older male prostate. Image adapted

from [3]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112.4 Gleason Grading schematics, where a higher grade is assigned to cancer

cells possessing less structural similarity with healthy prostate cells. Imageadapted from [4]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.5 Evolution of utilisation of treatment technique, stratified by risk assessment.Image adapted from [5]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.6 Murine melanoma cell survival curve. Multiple independent trials are plot-ted (boxes), along with the geometric mean values (triangle) to generatea cell survival curve. The solid curve corresponds to the linear quadraticmodel [6]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.7 Linear Quadratic Model components. Each individual component of themodel are shown in dashed blue, with the final survival curve shown in red[6]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.8 The therapeutic Ratio. If the dose required for 50% TCP (Dt) is less thanthe dose required for 50% NTCP (Dn), then the therapeutic ratio will begreater than one, and treatment will overall be beneficial to the patient (A).Conversely, a therapeutic ratio less than one corresponds to treatment withgreater complications to normal tissue than tumour control (B) [6]. . . . . . 26

2.9 Early versus late responding tissues. At low doses early responding tissueshave a smaller surviving fraction. When the dose is increased the survivallines cross, resulting in a significantly reduced surviving fraction for lateresponding tissues at increased doses [7]. . . . . . . . . . . . . . . . . . . . . 27

xiv

List of Figures xv

2.10 Hypothetical cellular survival curve undergoing fractionated treatment. Ateach point of fractionation, a new shoulder region appears following re-pair of sub-lethal damage. Consequently, the dose required to produce anequivalent effect (for example, 2% cell survival) is substantially larger thanif fractionation weren’t utilised [8]. . . . . . . . . . . . . . . . . . . . . . . . 28

2.11 Cell repair following differing fractionation regimes. The red curve corre-sponds to survival if no fractionation was utilised (i.e., a large single dose).Blue and green curves correspond to treatments of differing dose per frac-tion. As the dose per fraction is further reduced, the equivalent survivalcurves (dashed lines) will approach the limit where only lethal damage froma single track is observed (yellow line) [6]. . . . . . . . . . . . . . . . . . . . 29

2.12 Example DVH for a prostate cancer patient undergoing VMAT treatment.Curves corresponding to dose distributions for CTV (red), PTV (blue),bladder (yellow), and rectum (green) are shown. Treatment plan charac-teristics such as CTV dose uniformity and reduced high doses within therectum can be quickly and concisely observed. . . . . . . . . . . . . . . . . . 37

2.13 Comparison between 3D-CRT (left) and IMRT (right) prostate cancer treat-ment plans. The 95% isodose level (i.e. the region receiving at least 95%of the prescription dose) is shaded red. IMRT produces a more conformaltreatment plan, with faster dose drop-off outside the target volume [9]. . . . 39

2.14 Relapse free survival for patients with and without treatment protocol vi-olations (PV). Factors affecting relapse-free survival included inadequatecontouring, insufficient dose to the target volume during planning, dose-rate being too slow for treatment, and encountered technical difficulties[10]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

2.15 Locoregional control amongst patients whose plans were initially compliant(yellow), made compliant following QARC assessment (blue), non-compliantbut assessed as having no major impact on tumour control (purple), andnon-compliant but assessed as having a major impact on tumour control(red). The only statistically significant difference arose between the firstthree and the final grouping [11]. . . . . . . . . . . . . . . . . . . . . . . . . 47

2.16 Inter-observer contouring variation at the prostatic apex by five observers[12]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

2.17 Cumulative incidence of prostate cancer mortality. Significant reductionswere observed for patients received 18 months androgen deprivation therapyin conjunction with radiotherapy [13]. . . . . . . . . . . . . . . . . . . . . . 55

2.18 Examples of different atlases. From left to right; a single-atlas with a sin-gle template image, a multi-atlas with all template images utilised duringauto-contouring, a multi-atlas with a single best-fit template used for auto-contouring, and a multi-atlas with a subset of template images used forauto-contouring. Image reproduced from [14]. . . . . . . . . . . . . . . . . . 63

2.19 Steps required for multi-atlas development (above dashed line) and imple-mentation (below dashed line). Steps are ordered chronologically, withdashed blocks optional. Image adapted from [15]. . . . . . . . . . . . . . . . 64

2.20 Workflow for implementation of a multi-atlas, automatically contouringstructures on an incoming query image. Image is adapted from [16]. . . . . 65

List of Figures xvi

2.21 To deform the image on the left to match the image on the right, non-rigidregistration is utilised. Vector fields correspond to the individual transfor-mations applied to each voxel. Image adapted from [16]. . . . . . . . . . . . 66

3.1 Observer and STAPLE volume spread for each structure across all 35 pa-tients for trial 1. Notch box-plots show a significant difference in CTV andPTV median volumes by observer (B) compared to observers (A) and (C).Differences between observer median rectum volumes were found to be notstatistically significant. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

3.2 Spread in observer dice similarity coefficient (DSC) for CTV, PTV, bladder,and rectum across all patients for trial 1 . . . . . . . . . . . . . . . . . . . . 74

3.3 Patient 32 containing the poorest overlapping PTV contour (DSC = 0.7612)on T2 MR. Observer (A) (red), (B) (purple), and (C) (orange) contours areoutlined on (clockwise from top) transverse, coronal, and sagittal images.STAPLE volume is shaded light blue. . . . . . . . . . . . . . . . . . . . . . 75

3.4 Patient 1 PTV contours on T2 MR. All observer contours recorded DSC >0.9 with respect to the STAPLE volume shaded in light blue. However, dueto significant portions of Observer C’s (orange) PTV failing to include theSTAPLE volume, insufficient dose was delivered to the STAPLE PTV forthese treatment plans. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

3.5 Significant Spearman correlations for PTV (top image, p < 0.00035) andrectum (bottom image, p < 0.00026) for trial 1. Volume similarity, sensi-tivity, specificity and C-Factor significantly correlated with a range of ra-diobiological metrics for both structures. Most correlations identified wereweak, although TCPPoisson, minimum dose, and dose homogeneity showedmoderate correlations with volume similarity, sensitivity, and specificity forPTV. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

3.6 PTV volume similarity versus rectum NTCP, ρ = 0.33. . . . . . . . . . . . . 763.7 Variations in contoured volumes for each structure across the five patients

in trial 2. Gold standard volumes for each structure are shown as a blackcross. Due to the use of a majority vote, each gold standard volume is bydefinition smaller than the median volume for each structure. . . . . . . . . 77

3.8 Spread in observer DSC for CTV, PTV, bladder, and rectum across allpatients for trial 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

3.9 Patient 4 PTV contours on (clockwise from top) transverse, coronal, andsagittal images. Observer A (highlighted dark green) displayed the poorestoverlapping PTV with respect to the majority vote PTV (shaded light blue),with a DSC of 0.8167. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

3.10 Patient 4 dose distribution derived from observer A’s PTV contour (figure9). The majority vote PTV is outlined in light blue, while the 78 Gy, 50Gy, and 39 Gy isodose lines are shaded light green, red, and light blue re-spectively. This treatment plan resulted in zero tumour control probabilityfor the majority vote PTV. It can clearly be seen on the sagittal and coronalslices that significant portions of the majority vote PTV were under-dosedduring treatment planning. . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

List of Figures xvii

3.11 Significant Spearman correlations for PTV with p < 0.00035 for trial 2.Correlations for PTV were much stronger than those observed in trial 1(figure 5), ranging from moderate (sensitivity, C-Factor) to strong (volumesimilarity, specificity). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

4.1 Pairwise DSC for CT (blue) and MR (green) structures (see Table 1 foracronym definitions). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

4.2 Observer volume variations for CT and MR defined CTV. Significant dif-ferences between median CT and MR volume were observed for all patients(P < 0.05). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

4.3 (a) Sagittal view of patient 2, where variations in superior boundary forrectum contours between two observers (filled contours) resulted in pooroverlap of bowel bag contours (outlined). (b) Density distribution of all ob-server rectum contours for patient 2, ranging from 100% observer agreement(purple) to only a single observer (red). . . . . . . . . . . . . . . . . . . . . 104

4.4 Variations in contouring for left and right NVB for Patient 1. For manyobservers, no contour overlap was evident. . . . . . . . . . . . . . . . . . . . 104

4.5 Pairwise Volume Similarity for CT (blue) and MR (green) structures. Asno distinction between which observer contour would be treated as the goldstandard volume, absolute values of volume similarity are reported here. . . 112

4.6 Pairwise sensitivity scores for CT (blue) and MR (green) structures . . . . . 1154.7 Pairwise specificity scores for CT (blue) and MR (green) structures . . . . . 1184.8 Pairwise Hausdorff Distance scores for CT (blue) and MR (green) structures 1214.9 Pairwise Mean Absolute Surface Distance scores for CT (blue) and MR

(green) structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1244.10 Pairwise Euclidean Centroid Distance scores for CT (blue) and MR (green)

structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1274.11 Detail of genito-urinary structures . . . . . . . . . . . . . . . . . . . . . . . 161

5.1 Schematic of Distance-to-Agreement (DtA) vector construction betweenCTVAtlas (blue) and CTVManual (red). A vector propagating from thecentre-of-mass of CTVManual intersects CTVManual and CTVAtlas. DtAbetween CTVs along this vector is measured (red arrow), with positive DtAcorresponding to CTVManual extending further outwards than CTVAtlas.Vectors were sampled every 6°, in both azimuthal and elevation, providinga complete spatial mapping of contouring variation. . . . . . . . . . . . . . 169

5.2 Median DtA between CTVManual and CTVAtlas for RADAR (top-left),RT01 (top-right), and CHHiP (bottom-left) datasets following vector anal-ysis. Shaded areas within each figure correspond to spatial regions where nostatistically significant difference existed between CTVAtlas and CTVManual

(p > 0.05 following Wilcoxon Analysis). Positive DtA values (yellow shadedregions) correspond to CTVManual extending further compared to CTVAtlas.Negative DtA values (blue shaded regions) correspond to the converse. Spa-tial regions corresponding to median bladder and rectum position lyingwithin 5mm of CTVAtlas are shown in the bottom-right figure. . . . . . . . 171

5.3 Hazard Ratio mappings with corresponding significance levels for overallpatient survival across RADAR, RT01, and CHHiP trials. . . . . . . . . . . 172

List of Figures xviii

5.4 Hazard Ratio mappings with corresponding significance levels for local pro-gression across RADAR, RT01, and CHHiP trials. . . . . . . . . . . . . . . 173

5.5 Hazard Ratio mappings with corresponding significance levels for PSA pro-gression across RADAR, RT01, and CHHiP trials. . . . . . . . . . . . . . . 174

5.6 DtA Cut-points (DCP) across RADAR, RT01, and CHHiP patients, basedon recorded patient death. Each vector cut-point was calculated using max-rank sum statistics, whereby patients dichotomised by DCP result in themost significant difference in patient outcome calculated by Wilcoxon ranksum test. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

5.7 DtA Cut-points (DCP) across RADAR, RT01, and CHHiP patients, basedon local progression. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183

5.8 DtA Cut-points (DCP) across RADAR, RT01, and CHHiP patients, basedon PSA progression. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

5.9 Adjusted p-value mappings for recorded patient death across RADAR, RT01,and CHHiP patients. p-values were adjusted using the Free Step-Down Re-sampling method as described by Westfall and Young (Algorithm 2.8, pages66 – 67) [17]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

5.10 Adjusted p-value mappings for local progression across RADAR, RT01, andCHHiP patients. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186

5.11 Adjusted p-value mappings for PSA progression across RADAR, RT01, andCHHiP patients. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187

6.1 Schematic of the study design. Participating centres had a local prostateradiotherapy protocol and associated AP configuration. Each centre creatednew AP configurations for the other two protocols through modification oftheir local AP configuration, based on a training dataset of three patientsprovided by each centre (pre-contoured CT datasets). . . . . . . . . . . . . 194

6.2 Median PTV DVHs for all patients for AP Configurations A (blue), B (red),and C (green). Note that the scale begins at 60 Gy for clarity. Solidlines correspond to host-centre AP configuration, modified configurationsare dashed. Solid and dotted grey p-value curves, indicating significantdifferences between population median DVHs, are also illustrated. Thedashed black line shows p=0.05. (For interpretation of the references tocolour in this figure legend, the reader is referred to the web version of thisarticle.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

6.3 Median Rectum DVHs for all patients for AP Configurations A (blue), B(red), and C (green). Solid and dotted grey p-value curves again indicatesignificant differences between population median DVHs. (For interpreta-tion of the references to colour in this figure legend, the reader is referredto the web version of this article.) . . . . . . . . . . . . . . . . . . . . . . . . 196

6.4 Mean left femoral head DVHs for all patients for AP Configurations A(blue), B (red), and C (green). Note that scale begins at 60 Gy for clarity.Solid lines correspond to host-centre AP configuration, modified configura-tions are dashed. Solid and dotted grey p-value curves, indicating signifi-cant differences between population mean DVHs, are also illustrated. Thedashed black line shows p = 0.05. . . . . . . . . . . . . . . . . . . . . . . . . 215

List of Figures xix

6.5 Mean right femoral head DVHs for all patients for AP Configurations A(blue), B (red), and C (green). Note that scale begins at 60 Gy for clarity.Solid lines correspond to host-centre AP configuration, modified configura-tions are dashed. Solid and dotted grey p-value curves, indicating signifi-cant differences between population mean DVHs, are also illustrated. Thedashed black line shows p = 0.05. . . . . . . . . . . . . . . . . . . . . . . . . 216

6.6 Mean bladder DVHs for all patients for AP Configurations A (blue), B(red), and C (green). Note that scale begins at 60 Gy for clarity. Solidlines correspond to host-centre AP configuration, modified configurationsare dashed. Solid and dotted grey p-value curves, indicating significantdifferences between population mean DVHs, are also illustrated. The dashedblack line shows p = 0.05. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217

List of Tables

2.1 Gleason gradings and descriptions, defined at ISUP Consensus Conference2005 [18]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.2 Clinical TNM Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142.3 Clinical Staging of prostate cancer, where X corresponds to an unknown

PSA or Gleason . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.4 South Western Sydney Cancer Therapy CTV and PTV contouring guidelines. 342.5 EORTC defined QA levels for a clinical trial [19]. . . . . . . . . . . . . . . . 44

3.1 Contouring similarity and radiobiological metrics. . . . . . . . . . . . . . . . 723.2 Contouring similarity metric derivations. . . . . . . . . . . . . . . . . . . . . 733.3 Strength of Spearman’s correlation ρ. . . . . . . . . . . . . . . . . . . . . . . 733.4 Mean, standard deviation, and COV of structure volumes across trial 1

patient cohort. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 743.5 Mean differences between observer and STAPLE plan dosimetry. . . . . . . 753.6 Mean, standard deviation, and COV of observer structure volumes for trial

2 patients. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773.7 Mean differences between trial 2 observer and majority vote plan dosimetry. 783.8 ICCs and minimum required number of observers for study reliability. . . . 803.9 Department prostate treatment planning protocol . . . . . . . . . . . . . . . 843.10 Trial 1 DSC statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 863.11 Trail 1 95% Hausdorff Statistics . . . . . . . . . . . . . . . . . . . . . . . . . 863.12 Bladder contouring similarity and radiobiological metric correlations, Trial

1: STAPLE gold standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . 873.13 Rectum contouring similarity and radiobiological metric correlations, Trial

1: STAPLE gold standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . 873.14 CTV contouring similarity and radiobiological metric correlations, Trial 1:

STAPLE gold standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 883.15 PTV contouring similarity and radiobiological metric correlations, Trial 1:

STAPLE gold standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 883.16 Bladder contouring similarity and radiobiological metric correlations, Trial

1: observer gold standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . 903.17 Rectum contouring similarity and radiobiological metric correlations, Trial

1: observer gold standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . 903.18 CTV contouring similarity and radiobiological metric correlations, Trial 1:

observer gold standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

xx

List of Tables xxi

3.19 PTV contouring similarity and radiobiological metric correlations, Trial 1:observer gold standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

3.20 Trial 2 DSC statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 923.21 Trail 2 95% Hausdorff statistics . . . . . . . . . . . . . . . . . . . . . . . . . 933.22 Bladder contouring similarity and radiobiological metric correlations, Trial

2: majority vote gold standard . . . . . . . . . . . . . . . . . . . . . . . . . 943.23 Rectum contouring similarity and radiobiological metric correlations, Trial

2: majority vote gold standard . . . . . . . . . . . . . . . . . . . . . . . . . 943.24 CTV contouring similarity and radiobiological metric correlations, Trial 2:

majority vote gold standard. Note: correlations for iso90 weren’t possibledue to every observer patient having 100% coverage of the CTV at the 90%isodose level, exactly matching the majority vote plans for each patient. . . 95

3.25 PTV contouring similarity and radiobiological metric correlations, Trial 2:majority vote gold standard . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

3.26 Bladder contouring similarity and radiobiological metric correlations, Trial2: radiation oncologist and radiotherapist contours only . . . . . . . . . . . 96

3.27 Rectum contouring similarity and radiobiological metric correlations, Trial2: radiation oncologist and radiotherapist contours only . . . . . . . . . . . 96

3.28 CTV contouring similarity and radiobiological metric correlations, Trial 2:radiation oncologist and radiotherapist contours only. Note: correlations foriso90 weren’t possible due to every observer patient having 100% coverageof the CTV at the 90% isodose level, exactly matching the majority voteplans for each patient. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

3.29 PTV contouring similarity and radiobiological metric correlations, Trial 2:radiation oncologist and radiotherapist contours only . . . . . . . . . . . . . 97

4.1 Contoured structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1004.2 Mean, standard deviation (SD) and coefficient of variation (COV) of ob-

server contour volumes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1024.3 Contouring variation metrics . . . . . . . . . . . . . . . . . . . . . . . . . . 1084.4 DSC distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1104.5 Absolute Volume Similarity Distribution . . . . . . . . . . . . . . . . . . . . 1134.6 Sensitivity Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1164.7 Specificity Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1194.8 Hausdorff Distance Distribution . . . . . . . . . . . . . . . . . . . . . . . . . 1224.9 Mean Absolute Surface Distance Distribution . . . . . . . . . . . . . . . . . 1254.10 Euclidean Centroid Distance Distribution . . . . . . . . . . . . . . . . . . . 1284.11 Contouring experience levels of CT structures for all volunteers prior to

commencement of this investigation. Volunteers have been stratified asRadiation Oncologists (RO) or non-Radiation Oncologists (Non-RO). Me-dian experience level for each structure is bolded red, with the lower boundbolded if the median lies between two columns. . . . . . . . . . . . . . . . . 130

List of Tables xxii

4.12 Contouring experience levels of CT structures for all volunteers prior tocommencement of this investigation. Volunteers have been stratified asRadiation Oncologists (RO) or non-Radiation Oncologists (Non-RO). Me-dian experience level for each structure is bolded red, with the lower boundbolded if the median lies between two columns. . . . . . . . . . . . . . . . . 131

4.13 Wilcoxon rank sum testing of contoured volumes between Radiation Oncol-ogists (RO) and non-Radiation Oncologists (Non-RO). Median volumes foreach structure for each patient are given in cm3. P-values indicate differ-ences between median volumes, with p-values < 0.05 bolded red. However,Bonferroni corrections must be made due to the large number (85) of statis-tical tests undertaken. This requires p < 0.0006 for statistical significance,which no comparison reached. . . . . . . . . . . . . . . . . . . . . . . . . . . 132

4.14 CT Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1374.15 MR Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

5.1 Patient numbers, baseline variables, and recorded patient outcome acrosseach clinical trial. Patients were dichotomised based on disease-risk groupsand cancer stages. Baseline variable and patient outcome numbers are withrespect to patients analysed. . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

5.2 Hazard Ratios for patient death within bladder and rectum spatial regionsfor RADAR, RT01, and CHHiP patients. . . . . . . . . . . . . . . . . . . . 188

5.3 Hazard Ratios for local progression within bladder and rectum spatial re-gions for RADAR, RT01, and CHHiP patients. . . . . . . . . . . . . . . . . 189

5.4 Hazard Ratios for PSA progression within bladder and rectum spatial re-gions for RADAR, RT01, and CHHiP patients. . . . . . . . . . . . . . . . . 190

6.1 Target and high/constraint priority OAR prescriptions and evaluations.Complete lists of evaluation criteria can be found within Table 6.4 (sup-plementary material). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193

6.2 Target and high/constraint priority OAR mean and standard deviations(S.D.) for all protocols. Conformity and homogeneity indices are shown forall PTVs. Differences in metrics considered significant (p< 0.05) are bolded.Complete list of objectives can be found within Table 6.9 (supplementarymaterial). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194

6.3 Total deviations for each AP configuration. Specific deviations can be foundin Table 6.10 (supplementary material). . . . . . . . . . . . . . . . . . . . . 196

6.4 Target objectives, and OAR evaluations for all three protocols . . . . . . . . 1996.5 AP Technique A adapted across all three protocols . . . . . . . . . . . . . . 2016.6 AP Technique B adapted across all three protocols . . . . . . . . . . . . . . 2036.7 AP Technique C adapted across all three protocols . . . . . . . . . . . . . . 2056.8 Treatment planning system and beam setup configurations . . . . . . . . . . 2076.9 Patient demographics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2086.10 Target and OAR mean and standard deviations (S.D.) for each centre’s

protocols. Conformity and homogeneity indices are also shown for eachprotocol’s PTV. Differences in metrics considered significant (p < 0.05) arebolded. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209

List of Tables xxiii

6.11 Total recorded violations by each centre’s AP technique . . . . . . . . . . . 2126.12 Modified AP Technique B to meet distributed protocol B criteria . . . . . . 2136.13 Original and modified AP technique B for protocol B, with significant dif-

ferences between metrics bolded. . . . . . . . . . . . . . . . . . . . . . . . . 214

Contents

Acknowledgements viii

List of Publications x

List of Abbreviations xii

List of Figures xiv

List of Tables xx

Table of Contents xxiv

1 Introduction 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.2 Thesis Aims and Chapter Outlines . . . . . . . . . . . . . . . . . . . . . . . 51.3 The Journey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Literature Review 82.1 Prostate Cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.1.1 Epidemiology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.1.2 Anatomy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.1.3 Detection and Staging . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.1.3.1 Biopsies and Gleason Score . . . . . . . . . . . . . . . . . . 122.1.3.2 TNM Staging . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.2 Prostate Cancer Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.2.1 Overview of Treatment Techniques . . . . . . . . . . . . . . . . . . . 152.2.2 Radiotherapy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.2.2.1 External Beam Radiotherapy . . . . . . . . . . . . . . . . . 162.2.2.2 Brachytherapy . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.2.3 Non-Radiotherapy Treatments . . . . . . . . . . . . . . . . . . . . . 172.2.3.1 Active Surveillance . . . . . . . . . . . . . . . . . . . . . . . 172.2.3.2 Surgery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.2.3.3 Systemic Therapies . . . . . . . . . . . . . . . . . . . . . . 192.2.3.4 Focal Therapy . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.3 Principles of Radiotherapy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

xxiv

Table of Contents xxv

2.3.1 Cellular Damage and Death . . . . . . . . . . . . . . . . . . . . . . . 202.3.2 Radiobiological Models for Radiotherapy . . . . . . . . . . . . . . . 22

2.3.2.1 Linear-Quadratic Model . . . . . . . . . . . . . . . . . . . . 242.3.3 Therapeutic Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252.3.4 Fractionation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.3.4.1 Repair . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.3.4.2 Repopulation . . . . . . . . . . . . . . . . . . . . . . . . . . 292.3.4.3 Redistribution . . . . . . . . . . . . . . . . . . . . . . . . . 302.3.4.4 Reoxygenation . . . . . . . . . . . . . . . . . . . . . . . . . 302.3.4.5 Radioresistance . . . . . . . . . . . . . . . . . . . . . . . . 31

2.4 Radiotherapy Workflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.4.1 Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.4.2 Contouring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.4.2.1 Target Volumes . . . . . . . . . . . . . . . . . . . . . . . . 332.4.2.2 Organs-at-Risk . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.4.3 Treatment Planning . . . . . . . . . . . . . . . . . . . . . . . . . . . 352.4.3.1 3D-Conformal Radiotherapy . . . . . . . . . . . . . . . . . 372.4.3.2 Intensity Modulated Radiotherapy . . . . . . . . . . . . . . 382.4.3.3 Volumetric Modulated Arc Therapy . . . . . . . . . . . . . 402.4.3.4 Stereotactic Body Radiotherapy . . . . . . . . . . . . . . . 40

2.4.4 Treatment Delivery . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412.4.4.1 Image Guided Radiotherapy . . . . . . . . . . . . . . . . . 42

2.5 Quality Assurance in Radiotherapy . . . . . . . . . . . . . . . . . . . . . . . 432.5.1 Quality Assurance in Clinical Trials . . . . . . . . . . . . . . . . . . 432.5.2 Contouring Variability and Uncertainty . . . . . . . . . . . . . . . . 48

2.5.2.1 Origin of Contouring Uncertainties . . . . . . . . . . . . . . 492.5.2.2 Efforts to Reduce Contouring Uncertainties . . . . . . . . . 502.5.2.3 Clinical Impact of Contouring Variations . . . . . . . . . . 51

2.5.3 Treatment Planning Variability . . . . . . . . . . . . . . . . . . . . . 522.6 Clinical Trials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

2.6.1 RADAR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 542.6.1.1 Setup Accuracy Study (SUAS) . . . . . . . . . . . . . . . . 56

2.6.2 RT01 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 572.6.3 CHHiP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

2.7 Automated Contouring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 602.7.1 Atlas-Based Contouring . . . . . . . . . . . . . . . . . . . . . . . . . 612.7.2 Multi-Atlas Development . . . . . . . . . . . . . . . . . . . . . . . . 632.7.3 Atlas Workflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

2.7.3.1 Registration . . . . . . . . . . . . . . . . . . . . . . . . . . 652.7.3.2 Contour Propagation and Fusion . . . . . . . . . . . . . . . 67

2.7.4 Limitations of Atlases . . . . . . . . . . . . . . . . . . . . . . . . . . 68

3 Correlations between contouring similarity metrics and simulated treat-ment outcome for prostate radiotherapy 69

Table of Contents xxvi

3.1 Supplementary Material A . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

4 Multi-observer contouring of male pelvic anatomy: Highly variable agree-ment across conventional and emerging structures of interest 984.1 Supplementary Material A . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1074.2 Supplementary Material B . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

5 Spatial analysis of target volume contours within three internationalprostate radiotherapy trials – Clinical impact of contouring variabilityon patient outcome 1625.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164

5.1.1 Background and Purpose . . . . . . . . . . . . . . . . . . . . . . . . 1645.1.2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . 1645.1.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1645.1.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

5.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1655.3 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

5.3.1 Clinical Trial Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . 1665.3.2 Atlas-Based Contouring . . . . . . . . . . . . . . . . . . . . . . . . . 1685.3.3 Vector Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1695.3.4 Statistical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

5.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1715.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1765.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1805.7 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1805.8 Supplementary Material A . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

6 Adapting automated treatment planning configurations across interna-tional centres for prostate radiotherapy 1916.1 Supplementary Material A . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

7 Discussion and Conclusion 2187.1 General Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2197.2 Inter-observer contouring uncertainties for intact prostate radiotherapy . . . 2227.3 Contouring variability within clinical trial datasets . . . . . . . . . . . . . . 2247.4 Quality of automated treatment planning techniques between centres . . . . 2267.5 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2277.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230

Bibliography 232

CHAPTER 1

Introduction

1

Chapter 1. Introduction 2

1.1 Introduction

Prostate cancer is the leading cancer diagnosis in Australian men, with 18,291 new inci-

dents accounting for 26.2% of all cancer cases in 2014 [1]. This number was projected to

rise to 19,508 by 2019, corresponding to a 1 in 6 risk of Australian men developing prostate

cancer by the age of 85 [20]. Conversely, prostate cancer mortality risk is comparatively

low at an estimated 1 in 35 [20]. While this is attributable to the indolent nature of the

disease, it also showcases that treatment for prostate cancer is highly successful. Multiple

treatment techniques are available, with preferred treatment based on tumour staging,

patient age, and potential risks of treatment-induced toxicity [21].

Radiotherapy is one of the major treatment techniques that is utilised within the clinic.

Optimal usage recommendations state that 58% of all prostate cancer patients should

receive external beam radiotherapy (EBRT) at some stage during treatment [22]. To

ensure high quality radiotherapy is safely delivered to a patient, precision and accuracy of

the radiotherapy beam must be assessed. While continual developments in engineering and

technology have resulted in the ability to deliver incredibly precise treatment, accuracy is

inherently dependent on the manual processes involved during the radiotherapy workflow.

Specifically, variations in quality of contouring and treatment planning have been shown to

reduce the effectiveness of treatment for patients, while additionally reducing the efficacy

of clinical trials [11, 23].

Clinical trials utilise Quality Assurance (QA) protocols to reduce these uncertainties asso-

ciated with each step within the radiotherapy workflow. Notwithstanding modern adaptive

techniques, radiotherapy incorporates a comparatively linear workflow. Early variations in

quality will subsequently propagate throughout the radiotherapy workflow, compromising

both the effectiveness of treatment and the clinical trial. This is represented in Figure

1.1, whereby uncertainties within the early manual steps will be encapsulated by all fur-

ther stages of the radiotherapy workflow. It is therefore imperative that a comprehensive

assessment of these uncertainties is undertaken within each clinical trial to provide the

highest quality of treatment.


Figure 1.1: Early variations in quality will be encapsulated by all further stepsof the radiotherapy workflow, affecting the overall quality of the clinical trial. Thefour key topics addressed by this thesis focus on the early stages of the radiother-apy workflow, with latter topics directly tied to improving clinical trial qualityassurance. How each topic relates to the radiotherapy workflow is represented by

the shading of the corresponding circles.

Contouring, otherwise known as segmentation or delineation, is the outlining of target

volumes and organs-at-risk (OARs) on an image. As one of the early stages within the

radiotherapy workflow, errors and uncertainties associated with contouring will propagate

throughout the remaining radiotherapy processes. Consequently, contouring uncertainties

have been regarded as one of the largest sources of uncertainty within radiotherapy [24].

Assessments of contouring uncertainties have been documented for multiple sites [25],

however no consensus metric choice for analysis exists [26, 27]. Not only does this make

comparisons between studies difficult, with metrics often displaying no correlation with

one another [28], but the metric utilised may have little to no correlation with dosimetry

[29, 30]. Consequently, an assessment of correlations between contouring variation metrics

and simulated treatment outcome for prostate cancer radiotherapy was justified.

Automated contouring is well investigated within the literature, with commonly utilised

methods including shape models [31], deep learning [32], and atlas-based segmentation

[15, 16]. Atlas-based segmentation incorporates a priori clinical information in the form

of manual contours that make up the atlas. Consequently, the resultant quality of the


automatically generated contours will be dependent on the quality of the contours within

the atlas. A male pelvic atlas for intact prostate radiotherapy requires bladder, rectum,

femoral heads and bowel bag OAR contours, as well as the clinical target volume (CTV).

However, recent literature has investigated dose to neighbouring OARs such as penile bulb

[33], perirectal fat space [34], and various pelvic floor muscles [35] as potential sources of

toxicity during prostate radiotherapy.

A male pelvic atlas that incorporates these emerging OARs could therefore be utilised

to investigate large prostate radiotherapy clinical trial datasets. An assessment of inter-

observer contouring variations for these structures would need to be undertaken during

development of the atlas. When applied to the clinical trial dataset, variations between the

original manual contours and the atlas contours could be assessed. Contouring variations

in specific spatial regions could then be correlated with significant variations in patient

outcomes within the clinical trial.

Intensity Modulated Radiotherapy (IMRT) and Volumetric Modulated Arc Therapy (VMAT)

are modern radiotherapy treatment techniques commonly utilised by clinical trials. The

quality of the treatment plans generated by these techniques is inherently dependent on

the skill of the clinician responsible for developing the treatment plan [36, 37]. Automated

approaches to treatment planning provide tools to standardise the quality of these treat-

ment plans, while being developed at a fraction of the time [38]. Multiple studies have

assessed the Pinnacle3 AutoPlanning™ (AP) module for automated treatment planning

for a range of treatment sties [39–42]. Compared to manually derived treatment plans, AP

generated plans have been found to be non-inferior in quality, while providing significant

time savings for the clinic. However, AP configurations in these studies were developed

and validated within a single institution. The robustness and quality of AP configurations

adapted to new protocols, such as those distributed during a clinical trial, has thus not

been adequately investigated.

The overarching aim of this thesis was the development and utilisation of advanced analysis

techniques for assessing contouring and treatment planning quality. Analysis techniques

incorporated clinical trial outcome data during development, ensuring the results and


findings from this work could ultimately be translated into improving future clinical trial

quality assurance.

1.2 Thesis Aims and Chapter Outlines

The aims of the thesis, along with corresponding chapter outlines, are detailed below.

1. Assess inter-observer contouring uncertainties for target volumes and

organs-at-risk for intact prostate radiotherapy:

(a) Determine the contouring variation metrics that significantly correlate with

simulated treatment outcome for prostate radiotherapy (Chapter 3).

(b) Investigate the prevalence of inter-observer contouring variations for multiple

male pelvic structures on Computer Tomography (CT) and Magnetic Resonance

Imaging (MRI) (Chapter 4).

2. Retrospectively analyse contouring variability across multiple clinical trial

datasets, correlating contouring variation with recorded patient outcomes:

(a) Apply a validated male pelvic atlas to automatically contour patient datasets

within the RADAR, CHHiP, and RT01 clinical trials.

(b) Spatially analyse variations between the original trial and atlas CTV contours,

correlating contouring variations with multiple clinical trial outcomes (Chapter

5).

3. Evaluate the robustness and quality of automated treatment planning

techniques between international centres:

(a) Investigate the robustness and quality of automated treatment planning con-

figurations, with each participating centre’s configurations iteratively adapted

to generate treatment plans meeting the remaining centre’s protocols (Chapter

6).


Individual chapter outlines are broken down below:

Chapter 2 reviews the literature, covering the basics of prostate cancer and available

treatment methods. The principles of radiotherapy are presented, along with an overview

of the radiotherapy workflow. The notion of quality assurance is introduced, with an

emphasis on contouring and treatment planning variability. After discussing key prostate

radiotherapy clinical trials, the role and process of automation in radiotherapy is discussed.

Chapter 3 investigates multiple metrics that are commonly utilised to assess inter-

observer contouring variations for prostate radiotherapy. Correlations between these con-

touring metrics and simulated treatment outcome measures is determined. Significant

correlations are elucidated using two separate datasets.

Chapter 4 assesses the extent of inter-observer contouring variations from multiple ob-

servers for numerous male pelvic structures on CT and MRI datasets. Standard structures

that are routinely contoured for clinical radiotherapy, as well as multiple emerging struc-

tures of growing interest within clinical trials are investigated.

Chapter 5 details how the RADAR, CHHiP, and RT01 clinical trial datasets were ret-

rospectively contoured using the developed male pelvic atlas. Incorporation of vector

mappings allowed for a comprehensive spatial mapping of contouring variations between

original trial and automatically generated atlas CTVs. Spatial regions where contour-

ing variations correlated with multiple patient outcomes within the clinical trials were

identified

Chapter 6 compares automated treatment planning configurations from three interna-

tional cancer therapy centres, each utilising the Pinnacle3 treatment planning system.

Clinical prostate AP configurations were adapted by each centre to meet other centre’s

treatment planning protocols. Robustness of AP configurations is assessed using number

of protocol violations and various dosimetry metrics.

Chapter 7 summarises the preceding chapters, discussing the relevance of the presented

work when translated to improving the quality of radiotherapy within a clinical trial. An

outline of potential future work utilising the developed techniques and datasets is provided.


1.3 The Journey

Funding for this candidacy was provided by National Health and Medical Research Coun-

cil (NHMRC) Project Grant 1077788. The work carried out in this thesis was conducted

within the South Western Sydney Clinical School at Liverpool Hospital. Research was

principally undertaken at the Ingham Institute for Applied Medical Research, with con-

touring and treatment planning assistance provided by both the Institute and Liverpool

Hospital. Expertise and computational resources were provided by the Australian e-Health

Research Centre at CSIRO. Atlas development was undertaken at Sir Charles Gairdner

Hospital.

CHAPTER 2

Literature Review

8

Chapter 2. Literature Review 9

2.1 Prostate Cancer

2.1.1 Epidemiology

Cancer remains one of the largest burdens of disease globally, with an estimated 18.1

million new diagnoses and 9.6 million deaths worldwide in 2018 [43]. Prostate cancer

is one of the most prevalent forms of cancer, accounting for nearly a quarter of all non-

melanoma cancer diagnoses in men from developed nations [44]. Incidence rates of prostate

cancer are highest in developed countries, likely due to more thorough screening and

diagnostic practices [45]. Mortality rates, conversely, have been found to be higher in

developing countries [45]. Risk factors associated with prostate cancer include race, age,

family history, and diet [21, 46].

Approximately one in two Australian’s will develop cancer before the age of 85 [47], with

one in five Australian men developing prostate cancer [1, 48]. In 2012 the age standardised

rate (ASR) for prostate cancer incidence was approximately 112 out of every 100,000

Figure 2.1: Incidence (solid line) and mortality (dashed line) rates of prostatecancer in Australian men. Image adapted from [1].


Figure 2.2: Male pelvic anatomy. Image adapted from [2].

Australian men, whilst mortality ASR was 12.9 out of every 100,000 [49]. Figure 2.1

shows the changes in ASR rates over the last 30 years.

Prostate cancer treatment typically involves one or more of chemotherapy, surgery, hor-

monal therapy, or radiotherapy. Radiotherapy is delivered either externally using linear

accelerators (LINACs), or in vivo via brachytherapy, and is an essential component of

curative treatment for prostate cancer [50]. External radiotherapy should be utilised for

58% of all prostate cancer patients [22].

2.1.2 Anatomy

The prostate is a small, spherical gland within the male pelvis located anterior to the

rectum, and inferior to the bladder and seminal vesicles (Figure 2.2). The urethra passes

from the bladder neck and enters the prostate at the superior base as the prostatic ure-

thra, before exiting at the prostatic apex as the membranous urethra. Ejaculatory ducts

transporting sperm and seminal vesicle fluid enter the prostate at the posterior base, con-

necting to the prostatic urethra at the verumontanum. The prostate generates an alkaline

fluid that mixes with seminal vesicle fluid to generate semen [13].


Figure 2.3: Zonal anatomy of A) a young, and B) an older male prostate. Imageadapted from [3].

The prostate comprises two lobes subdivided into four zonal regions (Figure 2.3): the

peripheral, central, and transition zones, and the anterior fibromuscular stroma [51]. Can-

cers within the prostate are mostly adenocarcinomas located within the transition and

peripheral zones [3]. In young men the peripheral zone is the largest zonal region of the

prostate. However, over time hypertrophy of the transition zone will enlarge the entire

prostate, compressing the central zone during this process.

2.1.3 Detection and Staging

Prostate cancer screening incorporates Prostate-Specific Antigen (PSA) testing and Digital

Rectal Examinations (DREs) [52]. PSA testing involves taking a blood sample to detect

elevated PSA levels. Prostates secrete PSA, a protein that liquefies clotted semen after

ejaculation, from small internal glands [53]. Healthy men will have PSA levels of approxi-

mately 4 ng/mL of blood, although this value will be elevated if the patient has undergone

recent exercise or sexual activity. As most prostate cancers originate within the cells lining

the organ, elevated PSA levels can be indicative of prostate cancer. Additionally, rising

PSA levels could signify that the cancer is no longer localised [54].


Grading ISUP Consensus Definitions

1 • Circumscribed nodule of closely packed but separate, uniform,rounded to oval, medium-sized acini (larger glands than Grade 3

2 • Like Grade 1, fairly circumscribed, yet at the edge of the tumournodule there may be minimal infiltration• Glands are more loosely arranged, not quite as uniform as Grade1

3 • Discrete glandular units• Smaller glands than seen in Gleason Grade 1 or 2• Infiltrates in and amongst non-neoplastic prostate acini• Marked variation in size and shape• Smoothly circumscribed small cribriform nodules of tumour

4 • Fused microacinar glands• Ill-defined glands with poorly formed glandular lumina• Large cribriform glands• Cribriform glands with an irregular border• Hypernephromatoid

5 • No glandular differentiation, composed of solid sheets, cords, orsingle cells• Comedocarcinoma with central necrosis surrounded by papillary,cribriform, or solid masses

Table 2.1: Gleason gradings and descriptions, defined at ISUP Consensus Conference2005 [18].

Due to the prostate’s proximity to the rectum, a doctor can perform a DRE to determine

whether cancer is present in the prostate. A gloved finger is inserted in the anus and

pressed against the rectal wall to feel the prostate. If the doctor notices any abnormalities

in the prostate’s size and shape during a DRE, or if elevated PSA levels were recorded

during PSA screening, then the patient will be sent to be imaged or undertake a biopsy.

2.1.3.1 Biopsies and Gleason Score

Biopsies involve the extraction of tissue from the prostate, which are subsequently analysed

within a pathology laboratory. A transrectal ultrasound (TRUS) guides spring loaded

needles through the rectal wall and into the prostate [53]. Multiple core samples are

acquired throughout the entire prostate volume to investigate the presence of cancer in

multiple regions. Biopsies are assessed using the Gleason grading system, where a grading


Figure 2.4: Gleason Grading schematics, where a higher grade is assigned tocancer cells possessing less structural similarity with healthy prostate cells. Image

adapted from [4].

based on the patterns and magnitude of cancer growth will be assigned. These gradings

were introduced in 1966 [55], whereby five distinct patterns of growth have been identified

[4, 18]. Gleason grading definitions are listed in Table 2.1, with examples of these gradings

illustrated in Figure 2.4.

As cancers are heterogeneous structures, rather than homogeneous, a range of gradings

will usually be observed during the biopsy. The Gleason Score is the sum of the two

Gleason Gradings that make up the majority of the tumour [18]. If most of the tumour

(approximately 95%) is comprised of a single grade, then the Gleason Score will be double

this single grading. Consequently, the final Gleason Score assigned will lie between 2 and

10. However Gleason Scores of 4 or less are rarely diagnosed due to a lack of consensus

regarding clinical relevance and reproducibility [18].


Stage Definition

Tumour TX Primary tumour cannot be assessedStage T0 No evidence of primary tumour

T1 Clinically unapparent tumour neither palpable nor vis-ible by imaging

T1a Tumour incidental histologic finding in ≤ 5% of tissueresected

T1b Tumour incidental histologic finding in > 5% of tissueresected

T1c Tumour identified by needle biopsy (e.g., because ofelevated PSA)

T2 Tumour confined within prostateT2a Tumour involves one-half of one lobe or lessT2b Tumour involves more than one-half of one lobe, but

not both lobesT2c Tumour involves both lobesT3 Tumour extends through the prostate capsuleT3a Extracapsular extension (unilateral or bilateral)T3b Tumour invades seminal vesicle(s)T4 Tumour is fixed or invades adjacent structures other

than seminal vesicles, such as external sphincter, rec-tum, bladder, levator muscles, and/or pelvic wall

Lymph NX Regional lymph nodes were not assessedNodes N0 No regional lymph node metastasis

N1 Metastasis in regional lymph node(s)Distant M0 No distant metastasisMetastasis M1 Distant Metastasis

M1a Non-regional lymph node(s)M1b Bone(s)M1c Other site(s) with or without bone disease

Table 2.2: Clinical TNM Definitions

2.1.3.2 TNM Staging

Once diagnosed, prostate cancer is staged by the size, prevalence, and spread of the can-

cer throughout neighbouring tissues and organs. TNM staging for prostate cancer was

introduced in 1992 [56], with clinical and pathological definitions regularly updated [57].

Clinical TNM definitions for prostate cancer staging are listed in Table 2.2, whereby the

cancer will be assigned individual scores related to the tumour size (T), nodal involvement


Stage T N M PSA Gleason

I T1a-c N0 M0 < 10 ≤ 6T2a N0 M0 < 10 ≤ 6T1-T2a N0 M0 X X

II T1a-c N0 M0 < 20 7T1a-c N0 M0 ≥ 10, < 20 ≤ 6T2a N0 M0 ≥ 10, < 20 ≤ 6T2a N0 M0 < 20 7T2b N0 M0 < 20 ≤ 7T2b N0 M0 X X

IIB T2c N0 M0 ANY ANYT1-2 N0 M0 ≥ 20 ANY

III T3a-b N0 M0 ANY ANY

IV T4 N0 M0 ANY ANYANY N1 M0 ANY ANYANY ANY M1 ANY ANY

Table 2.3: Clinical Staging of prostate cancer, where X corresponds to an unknownPSA or Gleason

(N), and presence of distant metastasis (M).

Clinical staging of a patient involves an overall assessment of PSA levels, Gleason Score,

and TNM Staging (Table 2.3). Clinical decisions related to the optimal treatment method

for the patient are derived from this final clinical staging.

2.2 Prostate Cancer Treatment

2.2.1 Overview of Treatment Techniques

Treatment techniques for localised prostate cancer include active surveillance, radical

prostatectomy, androgen deprivation therapy, and radiotherapy. Often multiple tech-

niques are utilised throughout the management of a patient, such as neoadjuvant androgen

deprivation therapy combined with radiotherapy [58]. Clinical utilisation rates for these

techniques over the last 15 years are shown in Figure 2.5. The ideal treatment technique for


a patient will depend on multiple factors such as clinical staging, comorbidities, potential

for complications, and life expectancy [59, 60].

2.2.2 Radiotherapy

Following the discovery of X-rays by Roentgen in 1895, it was quickly realised that these

beams could be used in the treatment of skin lesions [61, 62]. Radiotherapy is the treatment

of cancer utilising ionising radiation to interact with molecules within the patient. This

radiation can be delivered either externally or in vivo via sources implanted within the

patient (brachytherapy). The molecular mechanisms governing radiation damage are the

same in both instances and are further described in section 2.3. Principles of Radiotherapy.

2.2.2.1 External Beam Radiotherapy

External Beam Radiotherapy (EBRT) is one of the most important treatment techniques,

with 48.3% of all cancer patients suitable for EBRT at some stage during treatment [63].

This value rises to 58% for prostate cancer patients [22]. EBRT is delivered using Linear

Accelerators (LINACs), which produce high energy beams of photons or electrons and

deliver them with sub-millimetre precision and accuracy. Clinical photon beams typically

Figure 2.5: Evolution of utilisation of treatment technique, stratified by riskassessment. Image adapted from [5].


operate at 6 and 18 MV, while electron beams can operate at (but not limited to) 6, 9,

12, 16 and 22 MeV [64]. Further information regarding the workflow for EBRT is found

in subsubsection 2.3.4.5. Radioresistance.

2.2.2.2 Brachytherapy

Brachytherapy delivers a localised dose to the target volume via insertion of radioactive

seeds within the patient. Brachytherapy treatment of prostate cancer uses iodine-121 or

palladium-103 radioisotopes for low dose-rate brachytherapy (LDR-BT), and iridium-192

for high dose-rate brachytherapy (HDR-BT) [65]. In LDR-BT the seeds are permanently

implanted in the patient for a set time, and is recommended for patients with low-risk

prostate cancer and an expected life expectancy exceeding 5 years [65, 66]. HDR-BT in-

volves temporary implanting of the radioactive seeds, and is typically used as a radioactive

boost in conjunction with EBRT for intermediate and high risk patients [65, 67].

Brachytherapy’s efficacy for patients with low-risk prostate cancer has been verified in

multiple studies investigating biochemical control and overall survival [68]. In long term

follow up studies LDR-BT was found to improve biochemical control and overall survival

rates after 5, 7 and 12 years compared to EBRT within single institutions [69, 70], although

this has yet to be verified within a randomised clinical trial. HDR-BT studies have similarly

shown either comparable or even improved biochemical and survival outcomes compared

to other prostate cancer treatment techniques [67, 71].

2.2.3 Non-Radiotherapy Treatments

2.2.3.1 Active Surveillance

Active surveillance was introduced as a treatment technique to counter the large number

of over-diagnosed patients detected from PSA screening [72]. Nearly half of the men aged

between 55 and 67 with elevated PSA readings were estimated not to require additional

treatments [73]. Examples include benign and slow growing cancers, where the patient

would pass from other circumstances before encountering any symptoms of the cancer [74].


Typically, only low-risk prostate cancers are eligible for active surveillance [75], although

variations are observed between institutions [76]. Older patients with intermediate-risk

prostate cancer, while suffering concurrent illnesses, may also be offered active surveillance

as an alternative treatment [77].

Active surveillance involves the continual monitoring of the patient through periodic biop-

sies, PSA screening, and magnetic resonance imaging (MRI) [77, 78]. Low-risk prostate

cancer progresses slowly, often going undetected in patients until post-mortem analysis.

Many men will subsequently never suffer the associated symptoms, and therefore not re-

quire radical treatment from radiotherapy or surgery [77]. Long-term follow up data from

an active surveillance clinical trial found that, of the 993 prostate patients receiving active

surveillance treatment, only 15 patient’s (1.5%) deaths were attributed to prostate cancer,

while an additional 13 patients (1.3%) developed metastases [79]. These figures mirror re-

sults reported from other clinical trials involving low-risk patients undergoing more radical

treatments [80].

Regular surveillance is required due to the potential for the cancer to progress suddenly.

Whilst the optimal timeframe for surveillance is still a matter of contention, most studies

recommend a combination of PSA testing performed every 3 - 6 months combined with

biopsies conducted every 12 - 24 months [75, 77, 79]. Regular MRI scanning has been

shown to aid in determination of disease progression, although this work is ongoing [81].

If at any stage the cancer has been shown to progress, the treatment strategy for the

patient must be altered.

2.2.3.2 Surgery

The most commonly utilised treatment technique for prostate cancer is a radical prostate-

ctomy. A prostatectomy is suitable for localised prostate cancer at any staging, however

for low-risk cancers it shows no significant improvements on overall survival compared

with active surveillance [82]. Early studies have shown that most prostate cancers are

both multifocal and bilateral [83]. Therefore the prostate and seminal vesicles must be

completely removed during a prostatectomy to provide confidence that all cancer cells

have been removed [84].


During surgery preservation of the neurovascular bundle is critical, as damage to these

nerves can lead to erectile dysfunction [85]. The likelihood of developing erectile dysfunc-

tion following a prostatectomy varies within the literature, ranging from less than 10% up

to nearly 100% [86]. Recent advancements made in surgical techniques include the develop-

ment of laparoscopic radical prostatectomy and implementation of robotic approaches [87].

Laparoscopic surgery is minimally invasive surgery shown to reduce morbidity and blood

loss [88], whilst robotic processes have reduced the risk of perioperative complications such

as death and nerve injury [89].

2.2.3.3 Systemic Therapies

Androgen Deprivation Therapy (ADT) is often used in conjunction with radiotherapy,

with multiple clinical trials demonstrating the benefits of adjuvant ADT with radiother-

apy for patients with localised prostate cancer [90–97]. Primary ADT trials have been

less promising, showing little to no improvements in survival and quality of life [98, 99].

Primary ADT trials additionally saw increased cancer morbidity rates compared to radical

prostatectomy and EBRT [100].

A patient’s quality of life can be significantly impaired following ADT treatment. Common

side-effects include loss of sexual prowess (reduced libido and/or erectile dysfunction), gy-

naecomastia (body feminisation), loss of bone density, loss of cognitive function, fatigue,

and depression [101]. These physiological and sexual changes can be detrimental to a

relationship, with the patient’s partner often placed under significant distress [102]. Con-

sequently, utilisation of ADT as a primary treatment technique has significantly reduced

in recent times (Figure 2.5). However, it is still utilised if the prostate cancer is sufficiently

advanced with bone metastases also present [90].

If a tumour has metastasised and is no longer localised, many of the previous treatment

techniques will no longer be suitable. The treatment intent will often move from curative

to palliative, with chemotherapy commonly utilised in these situations. The administra-

tion of docetaxel every three weeks has been shown to improve overall patient survival

while reducing pain in clinical trials [103, 104], and is now widely used within the clinic

[105]. Significant side effects are observed with docetaxel, with gastrointestinal upset and


peripheral neuropathy common with chemotherapy [21]. If the patient becomes resistant

to docetaxel, cabazitaxel combined with prednisone can also be utilised [106, 107]. It was

shown that utilisation of cabazitaxel helped improve overall survival in patients who had

previously been treated with docetaxel [108, 109].

Immunotherapy utilises the body’s own immune system to treat tumours by suppressing

protein receptors that would ordinarily inhibit the immune system from eradicating can-

cerous cells [110]. Immunotherapy using multiple agents has been shown to be effective

for treating prostate cancer patients with tumours that have become resistant to ADT

[111]. Immunotherapy agent Provenge has been shown to be safe and improve patient

outcomes in phase I and II clinical trials [112]. Meanwhile, Sipuleucel-T was shown to

prolong overall survival in patients with metastatic prostate cancer in a phase III clinical

trial, however had no significant impact on time to disease progression [113].

2.2.3.4 Focal Therapy

A less popular treatment method for localised prostate cancer is focal therapy [114], which

incorporates either cryotherapy or high-intensity focused ultrasound (HIFU) [115, 116].

Focal therapy methods attempt to preserve the tissue surrounding the cancer to reduce

potential toxicities, whilst maintaining an ability to provide curative treatment. Focal

therapy has been adopted in other tumour locations, however currently there is a lack of

evidence supporting its use within the clinic for prostate cancer patients [114].

2.3 Principles of Radiotherapy

2.3.1 Cellular Damage and Death

Radiotherapy aims to induce cellular death of cancers, with lethal damage to the cell’s

DNA a crucial component of this [6, 117]. Radiation damage to DNA is observed as

the development of DNA crosslinks, damaged DNA bases, single-strand DNA breaks, or

double-strand DNA breaks [6]. Of these, double-strand DNA breaks are of the most

importance in radiotherapy, as repair is often not possible [118]. This compromises the


cell’s ability to repopulate, resulting in the eventual death of the cell. However, the

quantity of double-strand breaks during radiotherapy is comparatively small compared to

other forms of damage. For a cell receiving 1 Gray (Gy) of absorbed dose, only 25 – 50

double-strand breaks will be observed. This is significantly smaller than the 500 – 1000

single-strand breaks that will occur [6].

Radiation damages DNA either directly or indirectly. Direct damage arises when incident

radiation interacts with the hydrocarbons present in DNA strands, depositing energy that

directly ionises the molecules within the DNA [117]. The resulting free radicals are highly

reactive, resulting in a chain reaction of additional damage to hydrocarbons present in the

DNA. It has been found that direct interactions account for only 30% of all DNA damage,

with indirect interactions accounting for the remaining 70% [118].

Indirect damage to DNA arises from the ionisation of water molecules within the cell

[117]. Upwards of 80% of the mass of a typical cell is made up of water, resulting in

the majority of deposited dose being absorbed by these molecules. Ionisation of water

molecules results in highly unstable H2O+ cations, which subsequently decompose into

hydrogen (H) and hydroxyl (OH) radicals. These hydroxyl radicals are highly reactive

due to their unpaired electron, often diffusing short distances before undergoing further

reactions. Therefore, indirect damage requires the DNA strands to lie within the diffusion

distances of the radicals (approximately 10 – 20 Å) [6]. The ensuing chemical reactions

produce the aforementioned single- and double-strand breaks.

A cancerous cell is defined to have died once it has lost its ability to proliferate [6]. This

can arise from the complete destruction and breakdown of the cell, or from the cell losing

the ability to divide and repopulate. This latter process, known as senescence, occurs

naturally in healthy mitotic cells of a certain age. Senescence can be induced following

radiation damage to cancerous cells, usually following double-strand breaks, often resulting

in increased polyploidy [117, 119].

Physical destruction a cell can occur from apoptosis, necrosis, and mitotic catastrophe

[120]. Of these, the mitotic catastrophe is the most common following radiotherapy [6],

although it represents a pathway as opposed to a form of cell death [120]. As the name

suggests, mitotic catastrophe occurs during mitosis, where a cell’s chromosomes divide in


preparation for cell proliferation [121]. If defects and damage of the DNA have been left

unrepaired in the original cell, the resultant cells following proliferation will retain the

same damaged DNA structure. These abnormalities will result in an eventual breakdown

in the cell’s functional integrity, although this may take several iterations of proliferation

before the cell reaches this stage. Apoptosis is the most common form of cellular death

following mitotic catastrophe [6, 120].

Apoptosis, also known as ‘programmed cellular death’, is a genetically regulated form of

cellular death [6, 117, 120]. It is governed by the tumour-suppressor protein p53, which

stabilises upon stresses and damage to DNA and begins a controlled destruction of the cell

[120]. Apoptosis is a non-inflammatory process, resulting in a shrinkage of the cytoplasm,

blebbing, and condensation of the nucleus (amongst other processes) [117]. Cells at the

end of apoptosis will fragment into smaller bodies, which are sequentially destroyed by

phagocytic cells in the body [6].

Conversely, cells undergoing necrosis will swell significantly, resulting in an eventual rup-

ture of the cellular membrane [6]. Necrosis was initially believed to be non-regulated

(passive), although recent evidence suggests that DNA pathways exist that regulate this

behaviour [122, 123]. Inflammation follows the rupturing of a cell, with the leakage of

cellular components creating a damaging environment for neighbouring cells [6].

2.3.2 Radiobiological Models for Radiotherapy

The fundamental goal for cancer cure requires the cellular death of all cancerous cells

within the patient. If this is not achieved, any remaining clonogenic cells will proliferate,

resulting in the eventual return of the tumour. As cellular death from radiation is a

stochastic process, cell assays are used to evaluate how cells respond to radiation [6, 117].

A known quantity of cancerous cells are irradiated with a specific radiation dose and

allowed to incubate. Cells are defined to have survived the treatment if they can continue

to proliferate and form cell colonies. The ratio between total cell colonies and the initial

number of cancerous cells is the plating efficiency.


Figure 2.6: Murine melanoma cell survival curve. Multiple independent trialsare plotted (boxes), along with the geometric mean values (triangle) to generatea cell survival curve. The solid curve corresponds to the linear quadratic model

[6].

As multiple factors will affect a cell’s ability to proliferate and form a colony, a second cell

assay is prepared. This control group is set up in identical fashion to the original assay,

except it receives no radiation. The ratio between plating efficiencies gives the percentage

of cells that would survive this dose of radiation. This cell survival can be calculated

for many combinations of doses and cell types and plotted on a semi-logarithmic scale to

generate a cell survival curve (Figure 2.6).

The overall shape of a survival curve is influenced by many factors, however all curves

display a low dose ‘shoulder’ region, followed by a linear response at higher doses [6]. Cell

survival curve characteristics are dependent on cell type, quality of radiation, and whether

the cell is healthy or cancerous. Multiple radiobiological models exist for cell survival

curves, the most popular being the Linear Quadratic Model [124]. Whilst the model does

not exactly match the observed survival curves, especially at higher doses, it does allow

for quick and simple comparisons between different treatment regimes.


Figure 2.7: Linear Quadratic Model components. Each individual componentof the model are shown in dashed blue, with the final survival curve shown in red

[6].

2.3.2.1 Linear-Quadratic Model

The linear quadratic model assumes that cell death is caused by double-strand breaks in

DNA from either a single or multiple interactions [125]. The rate of cell death is therefore

dependent on the probability that the double-strand break is the result of a single particle

interaction (α), or two independent particles produced two single-strand breaks in close

proximity (β). Mathematically, the survival probability is expressed as:

S = e−(αD−βD2) (2.1)

The constants α and β are unique for each cell type and differ between healthy and

cancerous tissues. Values range from 0.1 – 1 Gy−1 for α, and 0.01 – 0.1 Gy−2 for β [6].

Figure 2.7 shows the effect of these parameters on the resulting survival curve. The initial

‘shoulder’ region of the curve is primarily due to the α term, with this term describing

cell death due to irreparable damage. Conversely, β corresponds to cell death due to the

accumulation of otherwise repairable damage [126]. As the delivered dose increases, the

number of single-strand breaks increases. The probability that two of these lie within close


proximity, resulting in a double-strand break, subsequently increases [6, 117]. At larger

doses this term therefore becomes the dominant contributor to the survival curve.

The ratio α/β equates to the dose where both linear and quadratic components within the

model equally contribute to cell death (i.e. the intersection between the two blue lines in

Figure 2.7). This quantity describes the shape of the shoulder region of the survival curve,

where the largest variations between cell survival curves are observed [117]. The shape of

this shoulder region is important when considering fractionated treatment regimes.

2.3.3 Therapeutic Ratio

Radiotherapy aims to provide a lethal dose to cancerous cells, whilst simultaneously min-

imising the impact of radiation damage to neighbouring healthy tissues. This is achieved

by maximising the therapeutic ratio; a ratio of tumour control and normal tissue compli-

cations at a specific dose [6]. This is illustrated further in Figure 2.8, where dose-response

curves for tumours and normal tissues are drawn in red and blue respectively. These curves

give the respective probabilities of a tumour being treated (tumour control probability -

TCP) and healthy tissue reacting with a specified toxicity (normal tissue complication

probability - NTCP) at specific doses.

An increase in the therapeutic ratio, and subsequent improvement in radiotherapy treat-

ment outcome, can be achieved by shifting either the tumour dose response curve (red)

to the left, or the normal tissue response curve (blue) to the right. The most common

method of achieving an improved therapeutic ratio involves dividing the total treatment

into multiple smaller doses in a process known as fractionation [6, 117].

2.3.4 Fractionation

As mentioned earlier, the α/β ratio is a useful quantity for describing the shoulder region of

a cell’s survival curve. As β is a measure of how well a cell can recover following repairable

damage, cells with larger values of β (and smaller α/β ratios) have large shoulder regions

and are classed as late responders to radiation. Conversely, smaller β values (and larger


Figure 2.8: The therapeutic Ratio. If the dose required for 50% TCP (Dt)is less than the dose required for 50% NTCP (Dn), then the therapeutic ratiowill be greater than one, and treatment will overall be beneficial to the patient(A). Conversely, a therapeutic ratio less than one corresponds to treatment with

greater complications to normal tissue than tumour control (B) [6].

α/β ratios) have smaller shoulder regions, with these tissues labelled as early responders

[117].

Most tumours are early responders, partly due to their enhanced proliferation rates com-

pared to normal tissues, with corresponding α/β ratios of approximately 8 – 12 Gy. How-

ever, recent evidence suggests that slow proliferating tumours (such as prostate cancer)

display characteristics similar to late responders with smaller α/β ratios [127–132].

The premise behind fractionation is early responding tissues are less resistant to radiation

damage at lower doses than late responding tissues (Figure 2.9). At these low doses this

equates to the therapeutic ratio being favourable to treatment. With increasing dose the

survival curves eventually cross, corresponding to the therapeutic ratio no longer being

favourable for treatment (Figure 2.8).

If the total dose is divided into multiple fractions of small dose, then cells will be able

to repair sub-lethal damage caused by the single strand breaks dictated by the β term in

the linear-quadratic model. Each additional fractionated treatment will therefore display

a pronounced shoulder region due to cell death caused by a single track once again being

the dominant component (Figure 2.10) [6].


Figure 2.9: Early versus late responding tissues. At low doses early respondingtissues have a smaller surviving fraction. When the dose is increased the sur-vival lines cross, resulting in a significantly reduced surviving fraction for late

responding tissues at increased doses [7].

Whilst repair is the major radiobiological process that alters fractionated survival curve,

additional cellular processes must be considered as well. These include repopulation, redis-

tribution, reoxygenation, and radioresistance; and combined with repair are colloquially

are known as ‘The Five Rs of Radiotherapy’ [8, 133]. Significantly, not all processes im-

prove the therapeutic ratio, so all must be considered when assessing the benefits and

extent of fractionation for each cancer site.

2.3.4.1 Repair

A cell’s ability to repair DNA is dependent on the type of damage that has transpired.

Double-strand breaks are difficult to repair, with attempts often producing erroneous

results that eventually leads to cell death. Single-strand breaks are much easier to repair

due to the reciprocal base pair still being present in the DNA, allowing for knowledge of

which base pair was severed. The repair process starts instantaneously, and depending on

cell type can take between 6 and 12 hours to finish [117].


Cellular repair between treatment fractions results in a reduction in the steepness of the

effective survival curve. This is observed in Figure 2.11, where two fractionation schedules

(blue and green) produce effective survival curves of differing steepness. As the fractionated

dose is reduced, the quantity of sub-lethal single-strand breaks combining to form double-

strand breaks is reduced, resulting in a comparative increase in the number of single-strand

breaks that get repaired. This results in a shallower survival curve, and with continual

reduction in fractionated dose this asymptotically approaches the limit of a survival curve

where only single track damage occurs during treatment [6].

As cancerous cells typically have larger α/β ratios than normal tissues, and hence a reduced

shoulder region, sub-lethal damage is a significant contributor to cellular death at these

low fractionation doses. Consequently, a smaller proportion of cancer cells sufficiently

repair between fractions, resulting in a significant separation between the survival curves of

healthy and cancerous cells. This separation is at its greatest when individual fractionated

doses are equal [6, 8]. As a result, cellular repair results in an increase in the therapeutic

Figure 2.10: Hypothetical cellular survival curve undergoing fractionated treat-ment. At each point of fractionation, a new shoulder region appears followingrepair of sub-lethal damage. Consequently, the dose required to produce anequivalent effect (for example, 2% cell survival) is substantially larger than if

fractionation weren’t utilised [8].


Figure 2.11: Cell repair following differing fractionation regimes. The red curvecorresponds to survival if no fractionation was utilised (i.e., a large single dose).Blue and green curves correspond to treatments of differing dose per fraction. Asthe dose per fraction is further reduced, the equivalent survival curves (dashedlines) will approach the limit where only lethal damage from a single track is

observed (yellow line) [6].

ratio during fractionation.

2.3.4.2 Repopulation

Throughout treatment both healthy and cancerous cells continue to proliferate and re-

populate. However, as the surviving fraction of cells is reduced during treatment, the

proliferation rate of surviving cells begins to accelerate after specific time period to coun-

teract this [134]. As a smaller fraction of cancer cells survive each fractionated dose, they

will experience enhanced repopulation earlier than healthy cells will [6]. This reduces the

steepness of cancer cell survival curves, resulting in a reduction of the therapeutic ratio.

To counteract the effects of repopulation, fractionation treatments must be completed in

as short a time as allowable. Estimates indicate that every additional day of treatment

reduces tumour control by 0.5% [117]. However, cancers with small α/β ratios (such as


prostate) have significantly reduced proliferation rates, and thus the effect of repopulation

are significantly reduced for these cancers [6].

2.3.4.3 Redistribution

The cell cycle is a series of checkpoints that a cell passes through before it begins prolifer-

ating. It consists of interphase (consisting of the G1, S, and G2 phases) and mitosis (M).

During the gap phases (G1 and G2) the cell increases in size in anticipation of proliferation,

whilst DNA splitting occurs during the synthesis (S) phase. Finally, the cell splits into

two copies of itself during mitosis. If a cell is not set to divide, it can exist in stasis within

the G0 phase, separate from the rest of the cell cycle. The cell can move in and out of this

state, depending on the need for it to divide and repopulate [2, 6].

A cell’s radiosensitivity depends on the phase of the cell cycle that it resides in [135]. Cells

in the S-phase will have the greatest resistance to radiation damage, whilst cell death is

increased if the cell is either undergoing mitosis or is towards the end of the G1 phase.

Additionally, it has been observed that a cell’s progression through the cell cycle is halted if

DNA damage is detected [136]. This occurs at the G2 checkpoint, and therefore radiation

will see an accumulation of cells with an increased radiosensitivity.

Therefore, redistribution during fractionated treatment has the overall effect of increas-

ing a cell’s radiosensitivity. As the effects of redistribution depend on a cell’s ability to

move through the cell cycle and proliferate, highly proliferating cells are therefore more

radiosensitive than slowly proliferating cells [6]. Consequently, redistribution will produce

a steeper survival curve for cancerous cells than healthy tissues (where most cells are lo-

cated in the more radioresistant G0 phase), leading to an increase in the therapeutic ratio

during fractionation.

2.3.4.4 Reoxygenation

Indirect damage through ionisation of water molecules into hydroxyl radicals accounts for

approximately 70% of all cellular damage. Repair mechanisms can overcome the single-

strand breaks formed by these radicals, however the presence of nearby oxygen hinders


this process. Oxygen reacts with the damaged DNA, producing stable molecules that can

only be repaired with enzymes [6]. This repair mechanism is slower than repair through

chemical means, increasing the likelihood of multiple single-strand breaks combining to

form a double-strand break. This effect is highly significant, with a well oxygenated cell

three times more radiosensitive than a poorly oxygenated (hypoxic) cell [6].

To remain well oxygenated, cells must be located within 70 µm (i.e. the diffusion distance

of oxygen) of a capillary [6]. Any cells located further than this will either be hypoxic or

dead. This is commonly seen in tumours, where the rapid uncontrolled proliferation of

cells results in regions of cells located large distances from nearby capillaries. Angiogenesis,

the mechanism of creating of new blood vessels, can reoxygenate these cells. This process

is not well structured though, and consequently large hypoxic regions are commonly found

within a tumour [6]. This increases the radioresistance of the tumour, resulting in a

subsequent reduction of the therapeutic ratio.

During fractionated treatment the well oxygenated cancer cells are initially killed, result-

ing in the increase of the relative proportion of hypoxic cells within a tumour. However,

due to the outer cancer cells dying off, a portion of these hypoxic cells can now become

oxygenated. These newly oxygenated cells become more radiosensitive, with an increased

likelihood of receiving a lethal dose during the next treatment fraction. This process

repeats, resulting in a significant decrease in total surviving cells following fractionated

treatment [6]. Consequently, reoxygenation of hypoxic cancerous cells because of fraction-

ation results in an increase of the therapeutic ratio.

2.3.4.5 Radioresistance

Repair, repopulation, redistribution, and reoxygenation were initially the only mechanisms

considered when describing the biological effects of fractionation [8]. It was later shown

that cancer cells from different cell lines exhibit unique radiosensitivities at low doses [133].

This effect is magnified during fractionation, consequently affecting the therapeutic ratio

for treatment.


2.4 Radiotherapy Workflow

The radiotherapy workflow consists of four key components: imaging, contouring, treat-

ment planning, and treatment delivery. The methods incorporated at each step have

evolved over time, with improvements in both available technology and quality assurance

(QA) protocols implemented in the clinic.

The International Commission on Radiation Units and Measurements (ICRU) have pub-

lished multiple protocols covering a range of radiotherapy treatment techniques [137–139].

Due to differences in available resources within separate clinics, protocol recommendations

have been divided into three separate tiers. Level 1 recommendations correspond to the

minimum required standards of QA a clinic must meet, with only conventional radiother-

apy to be implemented at this level. Level 2 protocols must be met for more modern

treatment methods such as Intensity Modulated Radiotherapy (IMRT) and Stereotactic

Body Radiotherapy (SBRT), whilst level 3 protocols correspond to reporting for clinical

trial research [137].

2.4.1 Imaging

All patients receiving radiotherapy undergo an initial planning scan, which is required

for subsequent treatment planning. During this scan the patient will be oriented in the

treatment position, with the tumour within the scanned field-of-view. Treatment plan-

ning systems (TPS) require electron density information to accurately simulate delivered

radiation dose throughout the body. Currently only Computed Tomography (CT) scans

provide this information, although investigations are underway determining whether spe-

cialised workflows incorporating MRI scans can also provide this information [140, 141].

As well as this primary scan, often a patient will receive a secondary MRI or Positron-

Emission Tomography (PET) scan. This is particularly true for prostate cancer patients,

where an MRI scan is often incorporated due to the increased soft tissue contrast these

scans provide. This has increased accuracy in identification and contouring of the prostate,


resulting in improved dose coverage of the target volume and reduced dose delivered to

surrounding organs at risk [142, 143].

All secondary scans must be registered to the primary scan within the TPS. For prostate

cancer, this can be easily accomplished if fiducial gold seeds had been implanted within

the prostate gland [144]. These seeds can be easily identified on both CT and T1-weighted

MRI scans, allowing for quick and accurate registration of images within the TPS.

2.4.2 Contouring

Following image registration, the radiation oncologist will outline multiple structures and

regions within the scans. This process of contouring (otherwise known as delineation

or segmentation) is completed either within the TPS or an additional vendor software.

Contoured regions include the target volume(s), where dose is to be prescribed, as well as

neighbouring organs-at-risk.

2.4.2.1 Target Volumes

ICRU 83 defines multiple target volumes that are to be implemented during radiotherapy

treatment planning and reporting [137]. These volumes are derived from anatomical,

physiological, and geometric concepts, and contouring of these structures are an integral

component of the prostate radiotherapy workflow.

The Gross Tumour Volume (GTV) is defined as ’the gross demonstrable extent and location

of the tumour’. This corresponds to the volume of known tumour infiltration that can

clearly be seen on the scans. It may consist of a primary tumour site, as well as nodal or

metastatic regions. It is imperative during radiotherapy treatment that the entire GTV

receives adequate dose to obtain local tumour control [137].

Following this definition of the GTV, the Clinical Target Volume (CTV) is then defined

as the ‘volume of tissue that contains demonstrable GTV and/or subclinical malignant

disease with a certain probability of occurrence…’ [137]. This is the volume of suspected

microscopic tumour infiltration which cannot be observed on the scan. Consequently,


Risk CTV PTV

Low Prostate CTV + 7mmIntermediate Prostate + proximal 1cm of seminal vesi-

clesCTV + 7mm

High Prostate + entire seminal vesicles CTV + 7mm

Table 2.4: South Western Sydney Cancer Therapy CTV and PTV contouring guidelines.

pathological knowledge is required to account for potential avenues of microscopic spread

of the tumour. Margins required to define the CTV from the GTV are based on clinical

trial data.

Studies investigating prostate cancer found that 80% of all carcinomas were bilateral, with

nearly all carcinomas either peripheral or both peripheral and central [83]. Consequently,

due to the high probability of the cancer being bilateral, the CTV will typically be com-

prised of the entire prostate gland. Depending on the staging of the cancer, seminal vesicles

may additionally be included within the CTV. Contouring guidelines to aid in defining the

CTV for prostate cancer patients are available [145], while guidelines for CTV contouring

of prostate cancer patients at South Western Sydney Cancer Therapy are shown in Table

2.4.

Finally, the Planning Target Volume (PTV) was introduced to account for geometrical un-

certainties derived from organ motion and setup variations during treatment [137]. Each of

these could result in a significant shift of the CTV between individual treatment fractions.

To ensure that the CTV always receives adequate dose coverage, an additional margin

is applied to the CTV to generate the PTV. By ensuring that the entirety of the PTV

receives the prescription dose, this ensures that the CTV receives the prescribed dose,

regardless of potential variations in geometric setup between treatment fractions.

Care must be taken when defining PTV margins. While large margins will ensure satis-

factory dose coverage, there will be a resultant increase in dose delivered to neighbouring

healthy tissue. Conversely, too small a margin would preserve neighbouring tissue, but

could result in inadequate dose coverage for the CTV. PTV margins implemented for

prostate cancer patients at Liverpool Hospital are listed in Table 2.4.


2.4.2.2 Organs-at-Risk

Organs-at-Risk (OARs) are healthy organs neighbouring the PTV where delivered dose

is to be minimised to preserve their functional integrity and reduce potential toxicities

developing. For prostate cancer commonly considered OARs are the bladder, rectum,

and femoral heads. Multiple studies have investigated the consequences of large doses

being delivered to these organs [146]. This information has allowed for appropriate dose

constraints during treatment planning to be determined for these OARs.

Guidelines are available to improve OAR contouring accuracy, such as those for the male

pelvic anatomy [147]. However, just as for the target volumes where a margin was added

to account for organ motion and setup uncertainties, an additional margin may also be

applied to the OAR. The Planning Organ-at-Risk Volume (PRV) will therefore be used

during treatment planning to ensure that the corresponding OAR will never be located

within a high dose region, and consequently will satisfy corresponding dose constraints,

during all treatment fractions.

2.4.3 Treatment Planning

Following contouring, all patient scans and contours are transferred to the TPS (If not

already located there) for treatment plan optimisation. Additional information required

includes the prescription dose and treatment method, OAR dose constraints, beam quality,

and the planned fractionation schedule. The TPS will then calculate a treatment plan for

the patient based on these parameters. Prescription doses must be met in order for the

treatment plan be clinically accepted, and are prescribed to either a single point or a target

volume depending on treatment method [137].

Two methods of treatment planning can be performed by the TPS, forward planning

and inverse planning. Forward planning is the simpler of the two, with treatment beams

inserted into the TPS first, and dose delivered to target volumes and OARs calculated

based on the set beam geometry. Beams are iteratively adjusted, with the resultant dose

distribution recalculated each time, until a clinically acceptable plan is created. This is

typically used for simpler treatment setups and palliative treatments.


However, this process cannot be used for more advanced treatment techniques such as

IMRT and Volumetric Modulated Arc Therapy (VMAT), which subsequently utilise in-

verse planning. Inverse planning differs from forward planning by starting with the desired

target volume dose objectives and OAR constraints. Beam weightings and orientations

are then determined by the TPS following automatic optimisation to best satisfy these

objectives and constraints.

The initial mathematical algorithms utilised by inverse planning systems were developed by

Brahme et al. [148, 149], with modern dynamic processes optimising fluence maps based on

multi-leaf collimator (MLC) segments [150]. This method is utilised by the SmartArc™

planning module for VMAT treatment planning within the Pinnacle3 TPS [151]. This

module is an extension of the Direct Machine Parameter Optimisation (DMPO) module,

which is based on direct aperture optimisation [152].

Dose objectives for target volumes are chosen to ensure the developed treatment plan satis-

fies the prescription. Dose constraints for OARs are also assigned at this stage. Weightings

are assigned to dictate how the TPS prioritises balancing the corresponding dose objectives

and constraints during optimisation. Additional planning volumes, such as expansions or

contractions of original contours, can also be utilised to improve treatment plan quality.

Once beam parameters, dose objectives and constraints have been set, an initial treatment

plan is generated by the TPS and an assessment is made of each dose objective. This

assessment will be a measure of how well the individual dose objective was met by the

treatment plan, weighted by the dose objective weighting. The TPS will then iteratively

adjust beam weightings, characteristics (such as dose rate and gantry angle dwell time),

and MLC positions until a treatment plan that best meets all dose objectives is derived.

If an adequate treatment plan is not obtained during this initial optimisation process,

additional dose constraints and objectives can be specified to guide the treatment plan to

an optimal solution [153].

Figure 2.12 shows an example of a Dose-Volume Histogram (DVH), where each curve

corresponds to the percentage volume of an organ that would receive the stated dose or

higher. Dose objectives for each target volume or OAR will usually correspond to a single

position on the curve, which makes DVHs useful to quickly assess the acceptability of a


treatment plan. However, DVHs provide no information regarding the spatial distribution

of the treatment dose. Consequently, a DVH should never be used in isolation when

assessing the suitability of a treatment plan, as knowledge of the spatial regions of high

or low dose within a contoured volume may be required before a treatment plan can be

considered acceptable.

2.4.3.1 3D-Conformal Radiotherapy

Three-dimensional conformal radiotherapy (3D-CRT) involves shaping the radiation beam

to match the target volume using MLCs within the LINAC gantry head [154]. A fixed

number of beams are selected for treatment, with the field shape and weighting of each

beam iteratively adjusted to assess the resultant dose distribution (i.e. forward planning).

The introduction of 3D-CRT resulted in treatment plans that delivered radiation that

tightly conformed to the target volume boundaries, reducing the dose delivered to neigh-

bouring OARs. For prostate cancer, utilisation of 3D-CRT resulted in significant reduc-

tions in dose delivered to the bladder and rectum [155, 156], significantly reducing the

Figure 2.12: Example DVH for a prostate cancer patient undergoing VMATtreatment. Curves corresponding to dose distributions for CTV (red), PTV (blue),bladder (yellow), and rectum (green) are shown. Treatment plan characteristicssuch as CTV dose uniformity and reduced high doses within the rectum can be

quickly and concisely observed.


number of patients developing acute and late gastrointestinal and genitourinary toxicities

[157]. By producing treatment plans with dose coverage conformal to the target volume,

the introduction of 3D-CRT also provided an opportunity to dose escalate, potentially

improving the treatment efficacy [158].

However, genitourinary and gastrointestinal toxicities were still observed when utilising

3D-CRT for prostate cancer treatment, particularly during dose escalation trials [159].

Attempts at reducing high rectal doses were often compromised by the need to provide

adequate dose coverage to the target volume [160]. Optimal dose distributions for 3D-CRT

are only possible if the tumour is either spherical in shape, or if a large number of beams

are selected for treatment. Furthermore, developed treatment plans sometimes required

delivery of radiation at clinically impossible gantry angles. For these reasons the use of

3D-CRT to treat prostate cancer has been replaced by more modern treatment techniques

such as IMRT and VMAT.

2.4.3.2 Intensity Modulated Radiotherapy

IMRT is an extension of 3D-CRT, where the intensity profile across each treatment beam’s

cross section is varied throughout treatment [161]. This modulation is performed using

the MLCs, which either move while the beam is on (dynamic IMRT), or incrementally

move into whilst the beam is off (step-and-shoot IMRT) [153]. Treatment planning for

IMRT requires inverse planning, compared to forward planning utilised for 3D-CRT.

Figure 2.13 demonstrates how IMRT treatment plans can achieve highly conformal dose

distributions around the PTV with fewer beams than 3D-CRT. Again, the resultant con-

formal dose distributions allow for the possibility to escalate delivered dose to the target

volumes without significantly increasing OAR dose, which has been investigated in multi-

ple clinical trials [9, 162–165].

While utilisation of IMRT was hypothesised to lead to improvements in relapse-free sur-

vival for prostate cancer patients [163, 166], observed trial outcomes found that these

improvements were either minimal [167, 168] to non-existent [169]. This was due to 3D-

CRT treatment plans still achieving conformal dose distributions across the PTV (Figure


Figure 2.13: Comparison between 3D-CRT (left) and IMRT (right) prostatecancer treatment plans. The 95% isodose level (i.e. the region receiving at least95% of the prescription dose) is shaded red. IMRT produces a more conformal

treatment plan, with faster dose drop-off outside the target volume [9].

2.13). The most significant advantage of IMRT over 3D-CRT is the reduction of gastroin-

testinal and genitourinary toxicities [170–173]. Dose delivered to the rectum, bladder, and

femoral heads are significantly reduced when using IMRT [167, 168], which has resulted

in reduced acute and late grade 1 and 2 rectal toxicities such as rectal bleeding and gas-

trointestinal morbidity [165, 174]. However, utilisation of IMRT has been seen to result

in a small increase in erectile dysfunction for patients [175].

Concerns have been raised over the increased leakage radiation and low dose wash that

arises from IMRT. It is estimated that the utilisation of IMRT doubles the rate of secondary

cancers following treatment compared to 3D-CRT [176]. Additionally, a disadvantage of

IMRT is the significant increase in treatment times, with patients sometimes required

to remain stationary for 15-20 minutes. During this time factors such as internal organ

movement due to bladder filling will result in an increased uncertainty in the accuracy of

treatment delivery. Additionally, the increased time on couch (and subsequent increase in

required hospital resources) results in an increased cost for IMRT treatment [169, 177],

although this is offset by the reduction of late toxicities. These concerns have since been

alleviated following the introduction of VMAT into clinical practice.


2.4.3.3 Volumetric Modulated Arc Therapy

VMAT is an extension of IMRT that substantially reduces treatment time for the patient,

whilst maintaining comparable treatment plan quality [178, 179]. Rather than modulating

the intensity of a fixed number of beams, VMAT continuously irradiates the patient as

the gantry rotates in a pre-defined arc [161]. During gantry rotation the dose rate contin-

uously varies, while MLCs dynamically modulate the intensity of the beam’s cross section.

Consequently, the entire treatment is able to be completed in a single arc [180].

Studies investigating the efficacy of VMAT for prostate cancer patients have shown it

offers either comparable or superior treatment plan quality compared to IMRT and 3D-

CRT, with boosted CTV and reduced rectum doses commonly observed [179, 181–184].

More significantly, large reductions were observed in treatment times (2.2 minutes versus

8.1 minutes [185]) and delivered MU for VMAT compared to IMRT [181, 184, 186, 187].

While VMAT adds extra complexity during treatment, and consequently results in an

increase in required Quality Assurance (QA), for prostate cancer patients it is regarded as

the current EBRT treatment method of choice.

2.4.3.4 Stereotactic Body Radiotherapy

Stereotactic Body Radiotherapy (SBRT) is an additional treatment method, where large

doses are delivered to the treatment volume over only a few fractions. This can either be

as monotherapy, or as a boost dose [188, 189]. Utilisation of SBRT requires tumours with

well-defined boundaries, with margins in the order of millimetres applied to the target

volume. This contrasts with 3D-CRT and IMRT, where applied margins will be in the

order of a centimetre [190].

Compared to other treatment methods, SBRT treatment has a difference in treatment

philosophy and intent. While 3D-CRT, IMRT, and VMAT attempts to deliver a dose that

maximises the therapeutic ratio between tumour and normal tissue cells, SBRT instead

attempts to deliver an ablative dose to the tumour volume that completely kills off all

present cancerous cells. Consequently, dose delivered to normal tissue during SBRT must

be minimised due to the high likelihood of significant toxicities developing [191]. For this


reason, regular imaging must be incorporated for SBRT during treatment delivery (see

subsubsection 2.4.4.1. Image Guided Radiotherapy).

SBRT represents a promising treatment method for prostate cancer due to the prostate

having a suspected low α/β ratio [192]. As described in subsubsection 2.3.2.1. Linear-

Quadratic Model, a lower α/β ratio means that prostate tumours will be more sensitive to

larger doses per fraction. So much so, that some models have predicted an increase of up to

15-20% in biochemical control when SBRT is utilised [193]. Additionally, a smaller number

of fractions reduces overall treatment times from 8 – 9 weeks down to 1 - 2 weeks [194].

This could have significant benefits for patients that would have difficulty committing to

long fractionation schedules, along with reducing the overall cost of treatment.

Preliminary trials have demonstrated the safety of current SBRT techniques for treating

prostate cancer [194–197], with survival and toxicity rates comparable to other forms of

radiotherapy [194, 195, 197–204]. Five year follow up of the HYPO-RT-PC trial, inves-

tigating intermediate- and high-risk prostate cancer patients found that a 42.7 Gy in 7

fraction treatment regime was non-inferior in patient outcome compared to a traditional 78

Gy in 39 fraction treatment schedule [205]. Additionally, the PACE clinical trial found no

significant increase in gastrointestinal or genitourinary acute toxicities for prostate cancer

patients treated using SBRT [206].

2.4.4 Treatment Delivery

Following treatment plan optimisation and acceptance, the final step in the radiotherapy

workflow is the accurate delivery of the developed treatment plan to the patient over

multiple fractions. It is imperative that the patient is setup in the same position on the

treatment couch as they were during the planning scan. Tattoos or radiopaque markers

on the patient are used as reference points to align the patient with the in-room geometry,

while most modern LINACs incorporate on-board imaging to further improve patient setup

accuracy.

In order to ensure the accuracy of treatment, an intensive QA protocol is implemented

to determine whether the dose distribution delivered during each fraction of treatment


matches the distribution prescribed during treatment planning [207]. However even with

perfect patient setup and radiotherapy QA, the accuracy of treatment is still compromised

by the variability in position and shape of the target volumes. For prostate cancer this

can be due to internal motion of the prostate, as well as variable rectal and bladder filling

between treatment fractions [208, 209].

2.4.4.1 Image Guided Radiotherapy

Modern radiotherapy treatments incorporate Image Guided Radiotherapy (IGRT) to vali-

date patient setup (and consequently, target volume location) before and during treatment.

Imaging choices include Electronic Portal Imaging Devices (EPIDs), kilovoltage imaging

using fiducial markers or Cone-Beam CT, and transabdominal ultrasound for prostate po-

sition validation [210]. It has been shown that utilisation of IGRT significantly improves

toxicity free survival for prostate cancer radiotherapy [211, 212].

EPIDs are semiconductor detector systems included with nearly all modern LINACs. They

utilise the megavoltage treatment beam of the LINAC to generate a two-dimensional x-

ray image of the patient. A major benefit of EPIDs for IGRT is that images are created

from the beams eye view, providing a direct confirmation positioning of the target volume

with respect to the treatment beam. However, images obtained from megavoltage x-

rays display poor to negligible soft tissue contrast. Bony structures or fiducial markers

implanted within the prostate will instead be used as surrogates to verify the position of

the target volume. This does introduce an uncertainty in patient setup and alignment, as

it requires knowledge of the relationship between these structure’s locations and the PTV

[210].

Conversely, kilovoltage imaging systems produce images with superior soft tissue contrast,

but are mounted perpendicular to the delivered treatment beam. Acquired images are

reconstructed into a digitally reconstructed radiograph to validate the target volume po-

sition. Validation is performed either through the alignment of fiducial markers that were

surgically implanted in the prostate, or via the acquisition of a Cone-Beam CT scan that

allows for a direct registration with the original planning CT scan. Whilst this process


will introduce some uncertainty in target volume alignment, the benefit is that a surro-

gate is no longer required for target volume alignment. For prostate cancer radiotherapy,

alignment utilising EPIDs is generally considered acceptable for IGRT [210], however for

stereotactic treatments Cone-Beam CTs will be required.

2.5 Quality Assurance in Radiotherapy

Advancements in technology and medical knowledge have allowed for the continual im-

provement in patient outcome following radiotherapy treatment. Improvements in dose

distribution conformality have resulted in greater loco-regional control and relapse-free sur-

vival rates, whilst simultaneously reducing complications arising in neighbouring tissues

[213]. However, technological advancements result in an inherent increase in treatment

complexity, with each additional step within the treatment workflow providing additional

sources of uncertainty during treatment delivery [213].

Quality assurance (QA) protocols within radiotherapy are the implemented procedures

undertaken to ensure no major deviations are observed between the desired and actual

delivered treatment [214, 215]. Most radiotherapy centres have local QA protocols based on

working group recommendations [216], ensuring that each patient’s treatment is delivered

to a high standard. However, additional QA protocols are required when multiple centres

participate in a large clinical trial. Clinical trials are used in radiotherapy to investigate

the efficacy of new treatment technologies or techniques to improve patient outcome. In

order to obtain significant results, high quality data must be obtained to ensure potential

confounding factors are kept to a minimum. For these reasons, clinical trials incorporate

extensive QA not only ensures the safety of the patient and efficacy of treatment when

investigating a new technique, but additionally ensures that statistically significant results

can be derived from a limited sample size [217, 218].

2.5.1 Quality Assurance in Clinical Trials

Guidelines can be generated to assess the level of QA provided by a clinic in a clinical

trial [19, 214, 219], with Table 2.5 providing an example of these. Most uncertainties in


Clinical Trial QA

Level 1 Facility questionnaireExternal reference dosimetry audit

Level 2 Dummy runLevel 3 Limited individual case reviewLevel 4 Extensive individual case reviewLevel 5 Complex dosimetry check

Table 2.5: EORTC defined QA levels for a clinical trial [19].

radiotherapy arise from human error, so peer review is a critical component of a strong

QA protocol [220].

It has been shown that high levels of QA within a clinical trial improves patient care

and quality of life, while reducing toxicities from treatment [11, 214, 221]. Additionally,

implementation of a strong QA protocol can improve a clinical trial’s cost effectiveness by

reducing the required patient numbers to be recruited [214, 218]. Substandard QA may

result in an underpowered study, where subtle differences in treatment outcome would not

be observed [218].

Despite the evidence validating the need for strong QA protocols to be employed within

a clinical trial, centres are often unable or even unwilling to meet these levels of QA.

Funding issues, perceived lack of benefit, significant increases in workload, and additional

delays in patient accrual due to increased QA procedures are reasons given by centres

for not meeting the required levels of QA for a clinical trial [217]. Adherence to QA

protocols has been shown to range from 8% to 71% [221], with studies confirming that

patients receiving a reduced level of QA during treatment had a resultant poorer treatment

efficacy and increased rate of toxicities [10, 11, 23, 222–224].

In a study by Abrams et al., patients with resected pancreatic adenocarcinomas received

chemotherapy with concurrent radiotherapy [225]. Following the study, correlations be-

tween protocol compliance and patient survival and toxicity were investigated [222]. Com-

pliance was assessed in a binary fashion as either per protocol, or less than protocol de-

manded. Of the 416 patients within the study, only 216 received treatment per protocol.


It was found that patients that received treatment considered less than protocol demanded

had an increased risk of treatment failure, as well as an increased risk of developing grade

4 or 5 nonhematologic toxicity.

A separate study investigating unresectable pancreatic cancer measured survival, toxicity,

and response rates for patients treated with bevacizumab, capecitabine, and concurrent

radiotherapy [23]. It was found that 13.4% of patients in this study had unacceptably

large tumour contour volumes, resulting in a significant volume of healthy normal tissue

included within the target volume. This manifested in a significant increase grade 3+

gastrointestinal toxicities for these patients both during chemoradiotherapy (45% versus.

18%) and maintenance chemotherapy (45% versus. 13%) [23].

Dühmke et al. expanded on previous work investigating the role of radiotherapy following

chemotherapy for advanced Hodgkin’s disease patients [226]. The primary endpoint in

this trial was to assess whether reduction in radiotherapy dose would not compromise

treatment efficacy. Protocol violations, such as incorrectly contoured target volumes and

incorrect dose distributions, were identified in 127 of the 369 patients within the study

[10]. Patients who received treatment that violated protocol guidelines were observed to

have a statistically significant reduction in relapse free survival (72% versus. 84% after 7

years, Figure 2.14).

Another study retrospectively investigated 135 Hodgkin’s lymphoma patients where QA

was performed during radiotherapy, in order to assess relationships between major protocol

violations and recorded patient outcome [224]. Major protocol violations were observed in

47% of treated patients, with incorrect target volume contours the most common source.

However, 6.5 years following treatment no statistically significant correlation was observed

between protocol compliance and treatment failure rates. Notably, this is one of the only

studies where correlations were not observed, and has since been criticised due to the

small number of events recorded in the study (only 17 patients receiving radiotherapy

failed treatment) [217].

Eisbruch et al. investigated the efficacy of hypofractionated IMRT treatment for patients

with oropharyngeal cancer [223]. The QA protocol implemented in this study required

the study chair to review a number of submitted cases from each participating institution.


Figure 2.14: Relapse free survival for patients with and without treatment pro-tocol violations (PV). Factors affecting relapse-free survival included inadequatecontouring, insufficient dose to the target volume during planning, dose-rate being

too slow for treatment, and encountered technical difficulties [10].

Both target volume and organ-at-risk contours were assessed, with minor and major vari-

ation gradings attributed to cases where the prescription criteria was not met. 2 out of 4

patients assessed as having major contouring variations resulting in significant underdosage

were observed to have locoregional failure following treatment. Conversely, locoregional

failure was only observed in 3 of the 49 patients without major contouring variations.

Finally, substantial evidence linking poor quality QA with treatment efficacy comes from

a phase III trial investigating the addition of tirapazamine in conjunction with chemora-

diotherapy for advanced squamous cell carcinomas in head and neck patients [227]. The

QA protocol consisted of interventional review of treatment plans throughout the trial, as

well as retrospective analysis of radiotherapy plans and documents by the Trial Manage-

ment Committee (TMC) [11]. Initial results from the trial found no significant difference

between trial arms with respect to overall survival, failure-free survival rates, and time to

locoregional failure. Only 166 of the 857 patient plans were found to be protocol compliant

during interventional review. Recommendations were returned to clinics for the remaining


Figure 2.15: Locoregional control amongst patients whose plans were initiallycompliant (yellow), made compliant following QARC assessment (blue), non-compliant but assessed as having no major impact on tumour control (purple),and non-compliant but assessed as having a major impact on tumour control (red).The only statistically significant difference arose between the first three and the

final grouping [11].

687 patients from the Quality Assurance Review Centre (QARC). Following treatment,

retrospective review found that 208 plans still failed to meet QA protocol. Of these plans,

it was predicted that non-compliance of QA protocol would majorly impact tumour control

probability in 97 patients [11].

Non-compliance in order of prevalence was attributed to poor dose distributions in plan-

ning, incorrect dose descriptions, incorrect contouring, and excessively prolonged treat-

ment. However, incorrect contouring was adjudged to result in the most serious ramifica-

tions for tumour control [11]. An inverse correlation was observed to exist between the size

of the clinic and the percentage of patient plans that failed to meet QA protocol. Clinics

treating more than 20 patients in the trial had a non-compliance rate of 5.4%, compared

to 29.8% for clinics with less than 5 patients enrolled.

Figure 2.15 illustrates the locoregional failure rates from the study, where patients have

been divided into four cohorts following review by the TMC. The first three cohorts consist


of patients with initially compliant treatment plans (yellow), plans that were initially non-

compliant but made compliant following QARC review (blue), and plans that were non-

compliant at TMC review, but assessed to not have any significant effect on tumour control

probability (purple). No statistically significant difference in overall survival, failure-free

survival rates, and time to locoregional failure were found between these patient cohorts.

The fourth subgroup (red) consisted of non-compliant plans at TMC review, where it was

anticipated that non-compliance would result in a significant impact on tumour control.

These patients were observed to have significantly poorer outcomes than the initial three

cohorts across all three end points. After two years freedom from locoregional failure

and overall survival had reduced to 54% and 50% respectively. This corresponded to

20% reduction when compared to patients with initially compliant plans (78% and 70%

respectively), and was twice the survival benefit hypothesised using tirapazamine within

the trial [227]. Consequently, not only does poor QA in radiotherapy mask potential

benefits form within a clinical trial, but conversely a strong QA protocol can result in

clinical benefits equivalent to an advancement in cancer treatment [11].

2.5.2 Contouring Variability and Uncertainty

QA protocols attempt to minimise the errors and uncertainties in treatment that may

arise during any stage of the radiotherapy process. These uncertainties may originate

from mechanical or human processes. Of all the uncertainties that may arise, uncertainty

in contouring target volumes and organs-at-risk is considered the greatest source of in-

accuracy within radiotherapy [24, 213, 228]. Poorly defined target volume contours may

result in significant regions of the tumour not receiving adequate dose, leading to poorer

treatment efficacy and increased risk of recurrence [228].

Contouring variation can be sub-categorised as being either inter-observer (between ob-

servers utilising the same image set) or intra-observer (same observer across different

time points). Inter-observer variation constitutes a random error, however intra-observer

contouring variation can be considered both systematic and random. An example of a


systematic error is the inherent bias of a clinician when contouring the datasets. The vari-

ations in contours can be significant, with contoured prostate volumes observed to vary

by up to a factor of 1.6 [228].

Multiple studies have investigated the extent of contouring variations in radiotherapy

across a wide range of treatment sites including lung [229, 230], cervix [231–233], breast

[234, 235], bladder [236, 237], oropharynx [238], and prostate [239, 240]. Common reasons

cited for these variations include:

• Poor quality scans or incorrect imaging modality for treatment planning [24, 228,

229, 239, 241–243]

• Inadequate observer training [11, 24, 228, 231, 240]

• A lack of, or substandard QA protocols or guidelines [228, 230, 234, 243–248]

• Personal bias and interpretation of the observer (i.e. an unwillingness to include a

portion of volume due to its proximity to a critical structure) [24, 249, 250]

Consequently, because of its prevalence, peer review of a clinician’s contours is regarded

as one of the most important portions of QA during the radiotherapy workflow [220, 251].

2.5.2.1 Origin of Contouring Uncertainties

Studies investigating inter-observer contouring variability of the prostate on CT found the

largest uncertainties near the prostatic apex and seminal vesicles [12, 252–257]. This is

illustrated in Figure 2.16, where a large variation between five clinicians contours at the

prostatic apex is observed [12]. Uncertainties in contouring the prostate base have also

been observed, but typically are not as prevalent [252, 254, 256].

Inclusion of rectal tissue in the prostate contour results in erroneous treatment volumes,

and it has been shown only 84% of the true pathologically defined treatment volume is

properly contoured in a prostate contouring study [249]. An interesting observation from

this same study was that the contoured volume was 30% larger than the true volume. It


Figure 2.16: Inter-observer contouring variation at the prostatic apex by fiveobservers [12].

was suggested by the study that a large portion of neighbouring tissue, such as rectal wall,

was incorrectly contoured as the prostate.

When comparing imaging modalities, prostate volumes contoured on CT images have been

shown to be up to 40% larger than their equivalent volumes contoured on MRI [239, 256–

258]. Additionally, prostate volumes contoured on MR showed considerably less inter-

observer variation than corresponding CT contours [239, 258, 259]. This is attributed to

the clearer soft tissue resolution of structures, such as the prostatic apex, provided by MR

scans [142, 260]. Similar trends were observed for prostatectomy patients, where target

volumes defined using MR were reduced by 9 – 13% compared to those defined using CT

[261].

2.5.2.2 Efforts to Reduce Contouring Uncertainties

High quality radiotherapy requires precise contouring of the target volumes for treatment

planning [24], therefore, efforts have been made to reduce the variability inherent in the


process. Urethral and bladder contrast agents have been shown to reduce contouring

variation on CT scans [262]. As poor image resolution contributes to increased variability

[263], advances in imaging technology and software capabilities also lead to a reduction in

contouring uncertainty [26].

Multimodality imaging incorporating MR scans have now become standard-of-care for

prostate cancer treatments [243, 264]. Typically, gold seeds are surgically implanted into

the patient’s prostate prior to scanning. These seeds are easily detected on both MR and

CT scans, which allows for simple registration of the two scans. This is complicated if

significant differences in bladder and rectum filling are observed between scans, as well

as potential calcifications and poor seed placement, all of which affect the quality of the

registration [264].

Explicit guidelines and clinician training have been shown to be the most significant factors

in reducing inter-observer contouring variations [228], with poor guidelines shown to have

a greater impact on contouring variability than incorrect imaging modality [243]. For this

reason, clinical trial guidelines will explicitly clarify the regions to be contoured prior to

study commencement [244–247]. Implementation of ‘evidence based’ protocols increases

the mean target contour volume, whilst simultaneously decreasing the inter-observer vari-

ability [244]. Training programs also reduce inter-observer contouring variation within the

clinic [231, 232, 235, 240, 256, 265].

2.5.2.3 Clinical Impact of Contouring Variations

Studies investigating the clinical and radiobiological impact of inter-observer contouring

variations for prostate cancer patients are limited. Fiorino et al. investigated inter-observer

contouring variations of rectal contours from three observers for prostate cancer patients,

and found significant variation in rectum contours in only two patients [266]. These

patients both had large spacings for the CT scans (10 mm slice thickness), and variations

in NTCP were found to correlate with deviations in the anterior boundary of the rectum.

Two studies have investigated the impact on dosimetry of inter-observer contouring vari-

ability of the target volume for prostate cancer patients [12, 244]. The first incorporated


five patients datasets, with target volumes contoured by five observers [12]. Significant

differences in contoured volumes were observed around the prostatic apex and seminal

vesicles. No correlation between contouring variation and organ-at-risk dose metrics were

observed. The study concluded that regions where contouring variation was greatest cor-

responded to regions with the smallest impact on rectal and bladder toxicity.

Meanwhile, Mitchell et al. investigated the impact of introducing a protocol on inter-

observer contouring variations for postprostatectomy patients [244]. While introduction

of the protocol reduced variability between observers, an increase in mean CTV volume

across all three patients was observed. This corresponded to an increase in the number of

plans with unacceptable rectal dosimetry.

Van Herk et al. assessed the impact of contouring variability through insertion of random

and systematic errors into target volume contours within the treatment plan [267]. The

target volume contour was randomly displaced or rotated during dose computation, with

the resultant change in EUD and TCP assessed. The required planning margin that must

be applied to the target volume to counter these contouring variations was empirically

determined. It was found that a margin of 10 mm ensured that TCP reduced by less than

1% due to these introduced geometric errors.

Subsequently, no modern studies with large patient numbers have adequately investigated

the clinical and dosimetric effects of inter-observer contouring variation for prostate cancer

patients. It is known that dose distributions for more modern techniques such as VMAT

and SBRT vary from these, where tighter dose profiles are required for safe treatment.

Consequently, inter-observer contouring variation could produce substantial clinical and

dosimetric errors in treatment with these techniques, compared to those observed with

conformal radiotherapy or brachytherapy.

2.5.3 Treatment Planning Variability

While inter-observer contouring variability represents a major uncertainty when delivering

high quality radiotherapy, another source of uncertainty is the development of the clinically

acceptable treatment plan within the Treatment Planning System (TPS). Variability in


quality between planners is a consequence of human-driven processes generating these

plans, as well as the often-subjective measures used to evaluate the suitability of these

plans. For these reasons, quantitative metrics to explicitly describe plan quality should be

incorporated in conjunction with other quality control measures such as peer review [268].

Inter-centre IMRT planning exercises have been performed for multiple sites, with CT

image sets and pre-defined contours distributed to participating centres to remove the

impact of contouring variation from the study [36, 269–272]. Dose objectives or metrics

for target volumes and neighbouring OARs are used to assess the acceptability of these

treatment plans, while some studies also incorporate a qualitative assessment from a ra-

diation oncologist [273, 274]. Technical factors associated with the deliverability of the

treatment plan, such as treatment time and monitor unit number, have also been assessed

when assessing plan quality [269].

Significantly, in all these studies there was at least a single plan that was able to fulfil

all dose objectives strictly. The feasibility of planning delivery systems to meet required

dose objectives is therefore feasible as long as appropriate guidelines and protocols have

been clearly outlined prior to treatment planning [269, 271]. Retrospective review of 803

IMRT plans found that head and neck patients displayed the largest dosimetric variation

between patients, while prostate plans displayed the least [271].

Identifying the cause of variability is difficult, with conflicting results within the literature.

Batumalai et al. found that planning experience significantly affected plan quality for

IMRT head and neck plans, whereby only the two most experienced dosimetrists were

able to meet all PTV and OAR dose constraints [273]. This was not seen in other studies

[36, 37], where an intrinsic ‘planner skill’ was proposed as the most significant contributor

to planning quality [36]. Multicriteria optimisation processes [274, 275] and feasibility

estimation tools [276] have been shown to reduce planning variability while increasing

treatment plan quality from inexperienced planners.


2.6 Clinical Trials

2.6.1 RADAR

In 2003 the Trans-Tasman Radiation Oncology Group (TROG) began recruiting patients

into a randomised phase III multicentre clinical trial to determine the efficacy of 18 months

androgen deprivation treatment (ADT) following radiotherapy Trans-Tasman Radiation

Oncology Group [277]. The trial was planned as an extension of a previous study that

indicated 6 months adjuvant ADT improved patient outlook [278, 279]. The Randomised

Androgen Deprivation and Radiotherapy (RADAR) trial was a two-by-two factorial trial,

with patients randomly allocated to one of four treatment groups in a 1:1:1:1 split. Trial

arms consisted of either 6 or 18 months ADT treatment using Leuprorelin acetate, with or

without additional bisphosphonate therapy through the administration of zoledronic acid

Trans-Tasman Radiation Oncology Group [277].

A total of 1071 patients were enrolled into RADAR across 23 centres throughout Australia

and New Zealand between 2003 and 2007 [280]. Eligibility criteria was locally advanced

prostate cancer, defined as either stage T2b and greater, or T2a with Gleason score of

7+. Patients with lymph node involvement or metastases were excluded from the trial,

along with those with comorbidities that would compromise five years of survival [280].

Previous prostate cancer therapy, osteoporosis resulting in vertebrae height reduction of

30% or more, and bisphosphonate treatment also excluded patients from the trial.

Initial primary endpoints investigated in the RADAR trial were biochemical failure and

patient survival rates Trans-Tasman Radiation Oncology Group [277, 280], however these

changed to prostate cancer specific mortality after the initial ADT trial showed PSA

progression failed to correlate with these endpoints [279]. Ten year follow up results showed

that the addition of zoledronic acid when combined with ADT and radiotherapy had no

significant effect on patient survival rates [13]. Patients that received 18 months ADT

combined with radiotherapy had a 3.7% reduction in prostate cancer-specific mortality

compared to patients receiving 6 months ADT with radiotherapy (9.7% versus 13.3%

respectively, sub-hazard ratio of 0.70, Figure 2.17).


Figure 2.17: Cumulative incidence of prostate cancer mortality. Significantreductions were observed for patients received 18 months androgen deprivation

therapy in conjunction with radiotherapy [13].

Secondary endpoints included cumulative incidence of PSA progression, patient reported

outcome (PRO) assessments, and treatment related morbidities. Patients receiving 18

months of ADT reported poorer hormone treatment related symptoms, sexual activity,

social function, fatigue, and financial problems [281]. By 36 months post-treatment there

was no longer any significant difference between trial arms. PSA progression for patients

receiving 18 months ADT was reduced compared to patients receiving 6 months ADT

[280]. No differences in rectal or urinary dysfunction were observed between trial arms

[282], however an increase in urinary dysfunction was observed for patients treated with a

high dose rate brachytherapy boost.

At the commencement of the RADAR trial clinics around Australia and New Zealand

were installing and commissioning LINACs that were capable of dose escalation for three-

dimensional conformal radiotherapy treatment [283, 284]. Clinics could either prescribe

66, 70, or 74 Gy in 2 Gy fractions to the target volume, or 46 Gy combined with a 19.5

Gy high dose rate brachytherapy boost [284].

Multiple QA processes were incorporated into the RADAR trial. Before a clinic could

escalate the prescription dose within the RADAR trial they were required to participate

in a setup accuracy study (SUAS) [283]. A periodic audit of target volume and OAR


contours was also introduced. Following treatment, manual review of treatment planning

data for all patients was undertaken [285].

2.6.1.1 Setup Accuracy Study (SUAS)

The SUAS was a multifaceted QA program that required enrolled clinics to participate

in benchmarking exercises, electronic treatment plan review, structure audits, dosimetry

phantom studies, and treatment set-up verification studies [283]. A multi-centre QA study

of this magnitude had only been completed previously once [286], so software was developed

to analyse the large amounts of data provided by the study [287].

Clinics were to include at least 10 patient’s pelvic datasets to be treated according to the

RADAR trial protocol in the benchmarking exercise. During each fraction of treatment

setup accuracy was to be verified with either on-board imaging systems, electronic portal

imaging devices, or film [283]. Protocols related to patient setup and immobilisation, as

well as bladder and rectum filling, were left to the clinic’s discretion.

To participate in dose escalation, a clinic was required to demonstrate their capacity to

safely deliver radiation at each successive dose level. Assessment was based on Australasian

Consensus Guidelines for three dimensional conformal radiation therapy [288], which re-

quired at least 90% of treatment isocentres to lie within 10 mm of the intended isocentre.

Clinics that met only this requirement were only able to deliver 66 Gy treatment while

participating within the trial. If 90% of treatment isocentres were within 5 mm, the clinic

could escalate the treatment dose to 70 or 74 Gy. An option of escalating dose to 78 Gy

during treatment was also available, which required 90% of treatment isocentres lie within

3 mm of the intended isocentre [283].

All 22 centres enrolled within the setup accuracy study were able to meet the requirements

for 66 Gy treatment. Additionally, all but three of the centres satisfied the requirements

for dose escalation to 70 or 74 Gy. Four clinics were eligible to treat with a prescription

of 78 Gy, however due to personal choice or failing to satisfy other trial criteria, no clinics

escalated dose to this value [283]. By providing a range of dose prescriptions, the RADAR

trial generated enough data to simultaneously investigate the role of dose escalation on


treatment outcome. It was subsequently found that escalated dose led to a reduction in

the local progression, independent of which treatment arm the patient was enrolled in

[284].

Throughout the SUAS, questionnaires were administered to participating clinics, to allow

treatment and verification techniques from each clinic to be disclosed to the trial coordina-

tors. Setup data was analysed for random and systematic errors and associated with the

clinic’s procedures to identify potential sources for these errors. Feedback and recommen-

dations were then supplied to each clinic. A secondary survey found that 65% of audited

clinics changed clinical practice guidelines following participation within this trial [283].

Implementation of the SUAS resulted in the acquisition of a large, high quality dataset

for the RADAR trial, with lower protocol variation rates observed compared to previous

trials [278, 279]. Reduced rates of high-grade toxicity throughout the RADAR trial were

praised as indications of the effectiveness of this QA protocol [289]. However, secondary

analysis of the RADAR dataset identified correlations between treatment planning and

delivery factors and recorded patient outcome. It was found that patient treatments

calculated using rigorous dose calculation algorithms, as well as treatments involving 7

or more treatment beams, resulted in a significant higher incidence of local composite

progression [290]. Retrospective analysis additionally revealed clinical factors such as

smoking and utilisation of laxative correlated with increased urinary and gastrointestinal

toxicities respectively [291, 292].

2.6.2 RT01

The Medical Research Council (MRC) RT01 trial within the United Kingdom was ini-

tiated to investigate dose escalation for prostate cancer radiotherapy [293, 294]. At the

time, recent advancements in delivering conformal radiotherapy meant many centres were

investigating the efficacy on patient outcome when prescribed dose was increased from the

standard dose of 64 Gy in 32 fractions [295–297]. The RT01 trial was designed as a phase

3, open-labelled randomised controlled trial, with primary endpoints being biochemical

progression-free survival and overall survival [298].


Men with histologically confirmed T1b – T3a localised prostate cancer (i.e., not nodal

or metastatic involvement) with PSA < 50 ng/mL were eligible for the trial. Between

January 1998 and December 2001, a total of 843 men were randomly assigned to either a

control or dose escalated cohort. The control group received 64 Gy in 32 fractions, while

the dose escalated group received 74 Gy in 37 fractions. All patients were treated using 3D

Conformal Radiotherapy. Randomisation was stratified by risk of seminal vesicle invasion

and treatment centre. Patients in both trial arms also received neoadjuvant androgen

deprivation therapy for 3 – 6 months prior the commencement of radiotherapy [294].

Ten year follow up revealed that patient’s receiving escalated dose had a significantly

improved biochemical progression free survival compared to the control group (55% versus

43%, Hazard Ratio = 0.69). However, these improvements were not translated to overall

survival at ten years, which was identical between both trial arms. Additionally, significant

increases in bladder, bowel, and sexual toxicities were observed in the dose escalated

cohort immediately following radiotherapy [159, 299, 300]. Increased dose delivered to

neighbouring organs-at-risk were found to increase erectile dysfunction [301] and multiple

rectal toxicities [302–304].

Quality assurance procedures for the RT01 trial included the distribution of a question-

naire covering treatment planning and delivery [305], as well a purpose built phantom for

dosimetry auditing [306]. Both were found to help estimate and minimise potential setup

errors. Portal imaging was also utilised in the trial to ensure that treatment delivery was

accurate and precise [307]. Finally, a clinical planning exercise was employed to provide an

assessment of contouring variability amongst participating centres [254]. The variability

in target volume contours was found to be sufficient for the purposes of the trial.

2.6.3 CHHiP

The Conventional versus Hypofractionated High-dose intensity-modulated radiotherapy

for Prostate cancer (CHHiP) trial is a randomised, phase 3, non-inferiority trial also con-

ducted within the United Kingdom investigating the efficacy of hypofractionated treat-

ment for localised prostate cancer patients [308]. The study sought to confirm findings

from smaller randomised trials comparing a hypofractionated treatment schedule against


conventional fractionation [309–311]. Rationale for hypofractionation was derived from

recent findings that the α/β ratio for prostate cancer was small (1.4 – 1.9 Gy), compared

to standard tumours with α/β ratios of approximately 10 Gy [127–132].

Between October 2002 and June 2011, a total of 3216 men with localised prostate cancer

were enrolled into the CHHiP trial. Eligibility criteria for trial enrolment was histologically

confirmed T1b – T3a prostate cancer with no nodal or metastatic involvement. Patients

were randomly assigned into one of three cohorts, with randomisation stratified by National

Comprehensive Cancer Network (NCCN) risk group and treatment centre. The first cohort

received a conventional fractionation schedule of 74 Gy in 37 fractions. The second cohort

received a hypofractionated schedule of 60 Gy in 20 fractions, while the third also received

hypofractionated treatment of 57 Gy in 19 fractions [308]. All treatment plans were

delivered with intensity-modulated techniques, with some receiving optional IGRT.

Primary endpoint for the study was time to biochemical or clinical failure, with a goal of

showing that hypofractionated treatment of a large cohort of patients was non-inferior to

standard fractionation. Early assessments in 2006 of the first 457 patients revealed that

hypofractionated treatment was being well tolerated at 2 years follow up [312]. Five year

follow up found no significant difference in biochemical or clinical failure between the 74

Gy and 60 Gy cohorts (88.3% and 90.6% respectively) [308]. While non-inferiority was

established for the 60 Gy cohort, it could not be established for the 57 Gy cohort. No

significant differences in bladder or bowel toxicities (Radiation Therapy Oncology Group

(RTOG) Grade 2 or greater grading) between cohorts [308, 313].

Secondary analysis found no significant differences in patient outcome were observed

when patients were stratified by age [314]. Additionally, treatment plans developed using

inverse-planning methods resulted in reductions in bowel toxicity when compared against

forward-planning methods [315]. As a result, hypofractionated treatment of 60 Gy in 20

fractions was recommended as the new standard of treatment by the NHS. Clinical practice

throughout UK has subsequently changed based on these findings, with hypofractionated

treatment accounting for 49% of prostate cancer radiotherapy treatments by 2017 [316].

Quality assurance procedures developed for the CHHiP trial consisted of pre-trial process

documents, planning benchmark cases, and dosimetry site visits [317]. Minor protocol


deviations observed included incorrect bladder or bowel preparation, no patient immobil-

isation, and out of tolerance treatment verification. Prospective review of contouring was

completed for 100 patients, where major variations were observed in 71 of the cases. This

led to the clarification within the trial protocol of PTV margins to reduce the extent of

contouring uncertainty [317].

2.7 Automated Contouring

Due to the prevalence of variation within manual contouring, as well as the substantial

time and financial resources that are required, much work has been undertaken investi-

gating the efficacy and efficiency of automatic and semi-automatic contouring methods

[28]. Automatic contouring reduces the clinical workload, whilst providing a consistent

definition of the contour boundary. Some methods, such as thresholding based on an

intensity value, can be performed on a scan in isolation. However, most automatic con-

touring methods incorporate a priori clinical data, although how this data is utilised varies

between methods.

Commonly investigated automatic contouring methods include statistical shape models,

machine learning, and atlas-based methods [28]. Statistical shape models are comprised of

multiple surfaces and shapes corresponding to the structure to be contoured. An incoming

image to be contoured can then have features from these surfaces propagated via surface

meshes, vector fields, landmarks [31]. These methods are particularly stable against local

image artefacts and perturbations, due to the final contour being restricted to anatomically

plausible shapes. Statistical shape models have been shown to successfully contour bone

[318], brain [319], cardiac [320], and pelvic soft tissue structures [321, 322].

Machine learning, and in particular the subfield of deep learning, has seen an exponen-

tial rise in popularity in medical imaging contouring studies [323, 324]. The premise of

deep learning focuses on the construction of multiple processing layers to learn features

and representations of inputted data. Structure within these datasets is learned through

back-projection algorithms, which then update the internal parameters that represent the

data [32]. While many architectures for this process exist, one of the most common are


convolutional neural networks (CNN) [325, 326]. CNNs assign weights that are shared

between common features within an image that are spatially correlated, reducing the de-

grees of freedom within a model and considerably reducing computational requirements.

However, despite many advancements in recent times, a significant drawback for deep

learning are the large datasets required for training, along with the increased demands on

computational power [327].

Multiple studies have used deep learning to automatically contour soft tissue structures,

where results comparable to inter-observers have been regularly observed across multiple

sites [328, 329]. A popular CNN architecture for automatic contouring used in these studies

is the U-Net architecture [330]. For prostate segmentation, most studies have investigated

contouring on MRI [331–337], however deep learning has been applied to ultrasound [338]

and CT images [339, 340]. Dice Similarity Coefficients (DSC) of 0.87 – 0.88 were recorded

in these respective studies for prostate segmentation on CT, which is comparable to values

observed within inter-observer contouring variation studies. Accurate segmentation of the

prostate on MRI using deep learning could be utilised in future MRI-only planning of

prostate radiotherapy [341].

Automatic contouring utilising atlas-based methods requires a single or collection of refer-

ence images with the desired contours already segmented [16, 28]. By finding the optimal

mapping between the reference image and the input image, a transformation can be ap-

plied to the contours to match the anatomy of the input image. Atlases can incorporate

information from both CT and MRI scans [342–345], and multiple studies have investi-

gated constructing atlases for sites such as brain [346–348], prostate [342, 344, 349, 350],

and cardiac substructures [351, 352]. Additionally, atlases can be utilised as an initial

step in automatic contouring, before additional methods such as deep learning are used to

refine the contour boundaries [331, 333].

2.7.1 Atlas-Based Contouring

Atlas-based contouring allows for information regarding a structure’s spatial relationship

with neighbouring anatomy to be incorporated within the automatic segmentation [353].

Atlases consist of either a single or collection of reference images, each with previously


contoured structures by a trained expert [15]. Automatic contouring of an incoming pa-

tient scan proceeds by registering the atlas to match the anatomy of the new scan. This

transformation is applied to the contoured structures, which can then be propagated to the

incoming patient scan. This process is fast, reproducible, and can incorporate information

from multiple imaging modalities such as CT and MRI [342–345].

Multiple different methods for atlas construction and utilisation exist in the literature,

with examples shown in Figure 2.18 [354]. The earliest and simplest method utilises a

single template image as the atlas, with contoured structures directly propagated onto the

incoming scan following registration [355]. While easy to implement and not computa-

tionally intensive, single atlases fail to account for significant variation between incoming

patient anatomy [356].

Consequently, most atlases consist of multiple template images. Early investigations would

identify a single template image from within the atlas to use for automatic contouring

[357]. This soon evolved to probabilistic atlas-based contouring, whereby all template

images were registered into a single coordinate frame [358–360]. Utilisation of these atlases

provided information regarding the probability that the structure to be contoured was

present at different locations within the incoming image.

Finally, multi-atlases utilise all template images to define the contour [347, 354, 361].

Following registration of all template images, either all or some selection of the registered

template images will have contours propagated onto the incoming patient scan. The final

automatic contour will then be generated either through a voting or statistical method.

Utilisation of a multi-atlas has been shown to overcome the issue of observer bias present

when only a single template is used within the atlas.

Figure 2.19 shows a generic workflow for the development and implementation of a multi-

atlas [15]. Multi-atlas development requires the initial generation of template images

and associated contours to form the atlas, as well as an optional offline learning phase.

Keys steps when implementing an atlas to an incoming image are atlas registration, label

(or contour) propagation and fusion. While atlas selection, online learning, and post-

processing are strictly not required during atlas implementation, these optional steps will

improve the quality of the automatically contoured structure.


Figure 2.18: Examples of different atlases. From left to right; a single-atlaswith a single template image, a multi-atlas with all template images utilised dur-ing auto-contouring, a multi-atlas with a single best-fit template used for auto-contouring, and a multi-atlas with a subset of template images used for auto-

contouring. Image reproduced from [14].

2.7.2 Multi-Atlas Development

The fundamental property of atlases is that structures previously contoured by trained

experts on the template images are utilised to automatically contour the incoming patient

image. Consequently, the quality of automatic contour will be dependent on the quality

of the atlas contours (i.e. rubbish in equals rubbish out). Template images that are

representative of the anatomy within the incoming datasets to the contoured is imperative

[356]. This can be performed via visual inspection [362], or through automated clustering

[363, 364].

It has been shown that the accuracy of automatic contouring improves as the number of

template images within the multi-atlas is increased [365, 366]. However, each additional

template image will result in diminishing returns regarding contouring quality. Conse-

quently, as acquisition of expert-defined contours is a time and cost-intensive venture, a

pre-defined number of template images should be decided upon at commencement of multi-

atlas development. This number will be dependent on the imaging modality utilised, and

the structure (or structures) to be contoured [16, 367]. Disagreement exists within the

literature regarding the optimal number of template images required to construct an atlas


Figure 2.19: Steps required for multi-atlas development (above dashed line)and implementation (below dashed line). Steps are ordered chronologically, with

dashed blocks optional. Image adapted from [15].

[346, 354, 368]. Consequently, most studies will estimate this number based on available

computational resources and previous experiences [15].

To reduce the burden on a single expert when acquiring contours, it has been proposed

that contouring of the template images should be opened up to multiple observers who

aren’t experts [369]. The impact of inter-observer contouring variability within the atlas

can be abated through the utilisation of well-defined guidelines [251]. It has been shown

that utilisation of multiple observer’s contours within the atlas can lead to significant

improvements compared to utilisation of only a single observer’s contours [370].

2.7.3 Atlas Workflow

The workflow for applying a developed multi-atlas to automatically contour structure on

an incoming patient scan is shown in Figure 2.20. A collection of M template images are

each rigidly registered to the incoming query image, a similarity metric is then calculated to


Figure 2.20: Workflow for implementation of a multi-atlas, automatically con-touring structures on an incoming query image. Image is adapted from [16].

assess the similarity between registered template and incoming image (Step 1). Registered

template images are ranked according to this metric, and the top n template images are

selected for non-rigid registration based on a metric threshold (Step 2).

These n template images are deformably registered to the incoming image, with the non-

rigid deformations applied to associated atlas contours. These contours are subsequently

propagated onto the incoming patient scan (Step 3). A decision is made to fuse these

propagated contours for each structure, which defines the final contour (Step 4).

2.7.3.1 Registration

Registration is the spatial transformation of points from a deformable image mapped onto

homologous points on a target image [16]. This transformation can be applied either

to the entire image, or locally on a voxel-per-voxel basis. Rigid registration is the most

popular global registration technique, where rotations and translations are performed to


Figure 2.21: To deform the image on the left to match the image on the right,non-rigid registration is utilised. Vector fields correspond to the individual trans-

formations applied to each voxel. Image adapted from [16].

align the deformable image with the target image. Affine registration can also be employed,

whereby the deformable image can be skewed and scaled. Local registration, or non-rigid

registration, applies a unique transformation to each voxel across the deformable image.

This results in a vector mapping of voxels from the deformable image to the target image

(Figure 2.21).

The registration process consists of three key components; a measurement of similarity

between the images, determination of a deformation or transformation model, and an

optimization process [15, 371, 372]. These steps are related to a cost function:

C = S(F (x), T (M(x))) (2.2)

The cost function describes how well the deformable image M(x) correlates with the fixed

target image F (x) following the transformation T . This correlation is assessed using the

similarity metric s, which can be dependent on image points, structures, or intensity

differences between the images. Popular intensity similarity metrics utilised in atlases

include the sum of squared differences and normalised correlation coefficient, while mutual

information is utilised when registering between imaging modalities (i.e., from MRI to CT).

Registration is hence an optimisation process, with the required transformation model T

iteratively adjusted until the cost function value is minimised [16, 371].

The correct choice of transformation model is an important factor in the optimisation

process, and will be dependent on the site and desired application of the atlas [373]. The


two most popular models for non-rigid registration are the B-Spline free form deformation

[374, 375], and the diffeomorphic Demons Algorithm [376]. Additionally, many models

incorporate prior knowledge of the deformations in the form of regularisation constraints

to ensure that any transformations are still realistic [15, 372].

2.7.3.2 Contour Propagation and Fusion

Following registration of the deformable atlas images, associated contours are propagated

to the target image. This process will require interpolation of the deformable image

into the coordinate system of the target image, which if incorrectly applied can lead to

artefacts within the registration [16, 377]. Each atlas image transfers a label to every

voxel within the target image [363, 378], which can either be binary (voxel is or isn’t

within the contour), or take a value from a probability distribution. Nearest-neighbour

interpolation is frequently used [379], which provides a tissue consistency step along with

the propagation of information from the atlas images.

When utilising a multi-atlas, propagation of multiple atlas contours will result in differing

labels for voxels within the target image. Consequently, a decision will need to be made

regarding the final label attributed to each voxel. There are many available methods

for this decision, with a majority vote the simplest method utilised [15]. As the name

suggests, each voxel within the target image will take a final label corresponding to the

most frequent label provided by all propagated atlas contours. Further variations of this

set incorporate a threshold level of agreement, for example 30% agreement, rather than a

simple majority [16].

It should be noted that simple majority voting does not take into consideration the qual-

ity of registration between the atlas and target image. A simple extension was therefore

weighted voting, whereby weights are assigned based on the level of similarity between the

atlas and target image following registration. Initial investigations utilised normalised mu-

tual information to provide weightings [380], however these global weights were unable to

account for the localised variations in registration accuracy. Consequently, local-weighted

voting methods were developed to provide voxel-by-voxel weights to each atlas contour,


based on similarity within a small regions between the registered atlas and target image

[378, 381].

Aside from voting methods, probabilistic algorithms such as expectation-maximisation

have also been used to define the final contour from propagated atlas contours. The

STAPLE algorithm is an example of this, whereby the ‘true’ target image contour is

estimated through maximisation of the sensitivity and specificity of all propagated atlas

contours [382, 383]. A key assumption of the algorithm is that all information regarding

the true contour is provided by the propagated atlas contours. Further refinements of the

STAPLE algorithm integrate atlas selection, to ensure only atlas images with acceptable

similarity with the target image are included in the generation of the target image contour

[384].

2.7.4 Limitations of Atlases

Atlases exhibit great promise for improving the efficacy and efficiency of contouring in

clinical trials, and are now incorporated in many commercial treatment planning systems

[28]. However, work is still required in the design, implementation and validation of these

atlases before they display the accuracy required for clinical use [350, 385]. Currently,

atlases have been unable to generate contours that were an improvement on what a trained

expert would be able to produce [386]. In fact, for each of these investigated techniques, at

least one large mistake was made that an experienced observer or clinician would not make.

Consequently, when employed clinically, adjustments must often be made to automatic

contours, negating the time savings otherwise provided by the atlas [387, 388].

CHAPTER 3

Correlations between contouring

similarity metrics and simulated

treatment outcome for prostate

radiotherapy

69

© 2018 Institute of Physics and Engineering in Medicine

1. Introduction

Inter- and intra-observer contouring variability remains one of the largest sources of uncertainty in radiotherapy (Weiss and Hess 2003, Van Dyk et al 2013), with poor contouring significantly impacting the quality of treatment and patient outcome in clinical trials (Peters et al 2010). Uncertainties in contouring for prostate cancer radiotherapy have been attributed to imaging modality (Dubois et al 1998, Rasch et al 1999), observer training (Khoo et al 2012), and an over-cautiousness of clinicians limiting rectal tissue volume contoured by trimming the planning target volume (PTV) (Gao et al 2007). Implementation of clinical protocols and utilisation of MRI

D Roach et al

Printed in the UK

035001

PHMBA7

© 2018 Institute of Physics and Engineering in Medicine

63

Phys. Med. Biol.

PMB

1361-6560

10.1088/1361-6560/aaa50c

3

1

14

Physics in Medicine & Biology

IOP

23

January

2018

Correlations between contouring similarity metrics and simulated treatment outcome for prostate radiotherapy

D Roach1,2 , M G Jameson2,3, J A Dowling1,4,8,9,10, M A Ebert5,6,9, P B Greer7,8, A M Kennedy5, S Watt3 and L C Holloway1,2,3,9

1 South Western Sydney Clinical School, University of New South Wales, Sydney, Australia2 Ingham Institute for Applied Medical Research, Sydney, Australia3 Liverpool Cancer Therapy Centre, Liverpool Hospital, Sydney, Australia4 Australian e-Health Research Centre, CSIRO, Royal Brisbane Hospital, Herston, Australia5 Sir Charles Gairdner Hospital, Nedlands, Australia6 Faculty of Science, School of Physics, University of Western Australia, Crawley, Australia7 Calvary Mater Newcastle Hospital, Newcastle, Australia8 School of Mathematical and Physical Sciences, University of Newcastle, Newcastle, Australia9 Centre for Medical Radiation Physics, University of Wollongong, Wollongong, Australia10 Institute of Medical Physics, University of Sydney, Sydney, Australia

E-mail: [email protected]

Keywords: prostate, inter-observer variability, contouring, similarity metrics, VMAT

Supplementary material for this article is available online

AbstractMany similarity metrics exist for inter-observer contouring variation studies, however no correlation between metric choice and prostate cancer radiotherapy dosimetry has been explored. These correlations were investigated in this study. Two separate trials were undertaken, the first a thirty-five patient cohort with three observers, the second a five patient dataset with ten observers. Clinical and planning target volumes (CTV and PTV), rectum, and bladder were independently contoured by all observers in each trial. Structures were contoured on T2-weighted MRI and transferred onto CT following rigid registration for treatment planning in the first trial. Structures were contoured directly on CT in the second trial. STAPLE and majority voting volumes were generated as reference gold standard volumes for each structure for the two trials respectively. VMAT treatment plans (78 Gy to PTV) were simulated for observer and gold standard volumes, and dosimetry assessed using multiple radiobiological metrics. Correlations between contouring similarity metrics and dosimetry were calculated using Spearman’s rank correlation coefficient.

No correlations were observed between contouring similarity metrics and dosimetry for CTV within either trial. Volume similarity correlated most strongly with radiobiological metrics for PTV in both trials, including TCPPoisson (ρ = 0.57, 0.65), TCPLogit (ρ = 0.39, 0.62), and EUD (ρ = 0.43, 0.61) for each respective trial. Rectum and bladder metric correlations displayed no consistency for the two trials. PTV volume similarity was found to significantly correlate with rectum normal tissue complication probability (ρ = 0.33, 0.48). Minimal to no correlations with dosimetry were observed for overlap or boundary contouring metrics. Future inter-observer contouring variation studies for prostate cancer should incorporate volume similarity to provide additional insights into dosimetry during analysis.

PAPER2018

RECEIVED 28 April 2017

REVISED

18 December 2017

ACCEPTED FOR PUBLICATION

4 January 2018

PUBLISHED 22 January 2018

https://doi.org/10.1088/1361-6560/aaa50cPhys. Med. Biol. 63 (2018) 035001 (14pp)

Chapter 3. Prostate contouring metric correlations 70

2

D Roach et al

reduces inter-observer contouring variability (Dubois et al 1998, Rasch et al 1999, Mitchell et al 2009), although only limited agreement amongst observers is reached (Ost et al 2011).

The prevalence of inter-observer contouring variability for prostate cancer has been thoroughly investigated, however few studies have additionally assessed the impact of this variability on dosimetry (Vinod et al 2016a). Of the studies that have investigated this, only neighbouring organ-at-risk (OAR) dosimetry has been evaluated (Foppiano et al 2003, Livsey et al 2004, Mitchell et al 2009, Perna et al 2011). Consequently, no study has assessed the impact of inter-observer contouring variability on target volume dosimetry for prostate radiotherapy. Addi-tionally, studies to date utilised 3D-CRT, whereas most prostate cancer treatments now employ more modern techniques such as intensity modulated radiotherapy (IMRT), volumetric modulated arc therapy (VMAT), or stereotactic body radiotherapy (SBRT). These techniques generate tighter dose distributions matching the radiotherapy target volumes, and have been shown to reduce dose and subsequent toxicities to OARs (Palma et al 2008, Al-Mamgani et al 2009, Quan et al 2012). However, due to the steep dose gradients produced by these techniques, poorly contoured target volumes could result in larger impacts on dosimetry than has been observed in previous studies.

Similarity metrics are utilised to quantify contouring variations; however, no consensus exists over the choice of metric to incorporate during an investigation (Jameson et al 2010, Fotina et al 2012). This restricts compariso ns being made between studies, as similarity metrics cited may poorly correlate with one another (Sharp et al 2014). A combination of boundary and volume metrics is recommended (Fotina et al 2012), how-ever these metric choices may have little to no correlation with dosimetry. Studies investigating non-small cell lung cancers (Jameson et al 2014) and head and neck cancers (Beasley et al 2016) found that commonly utilised overlap metrics conformity index and DSC displayed weak or no correlations with simulated treatment outcome respectively. The aim of this study was to evaluate correlations between contouring similarity metrics and dosim-etry for prostate cancer planned for VMAT radiotherapy.

2. Materials and methods

2.1. Patient datasets and contouringTwo patient datasets were utilised for this study. The first trial incorporated forty-two patients from a prior study containing pre-treatment CT and MRI scans for localised prostate radiotherapy (Dowling et al 2015). Three observers (two experienced radiation oncologists, one experienced research radiation therapist) independently contoured clinical target volume (CTV) (Hodapp 2012), rectum, and bladder on T2-weighted MRI based on trial contouring protocol using Eclipse™ treatment planning software (Varian Medical Systems, Palo Alto, CA, USA). MRI scans were rigidly registered to CT with respect to gold fiducial markers implanted in the prostate, and CTV, rectum, and bladder contours were transferred to CT for treatment planning.

The second trial utilised a five patient subset from the previous trial, with patients selected based on prior clustering of post registration intensity based image similarities of potential atlas images with respect to the larger RADAR patient dataset (Trans-Tasman Radiation Oncology Group (TROG) 2005) using affinity propaga-tion (Frey and Dueck 2007, Kennedy et al 2016). This work was to be used during additional atlas based segmen-tation analyses. Ten observers across four treatment centres (two medical physicists, one radiation therapist, one radiographer, and six radiation oncologists) contoured CTV, rectum, and bladder on CT based on trial contour-ing protocol, with five observers using Eclipse™ (Varian Medical Systems, Palo Alto, CA) and five observers using Pinnacle3® (Philips Healthcare, Best, Netherlands) treatment planning software. A uniform 7 mm margin was applied to the CTV in both trials to define the PTV (Hodapp 2012). Following contouring, DICOM structure files were returned and imported into Pinnacle3® for treatment planning.

2.2. Gold standard volumesIn the absence of pathological information, ‘gold standard’ reference volumes were estimated for each structure from observer contours. The first trial utilised the simultaneous truth and performance level estimation (STAPLE) algorithm to generate these volumes (Warfield et al 2004). For the second trial, the large number of observers resulted in overlapping CTV and rectum STAPLE volumes that were deemed inappropriate for the study. Consequently, a majority vote was used to define the gold standard volumes for each structure in this trial. The impact the choice of gold standard volume has on analysis was investigated by a supplementary study of the first trial dataset, whereby observer contours were iteratively designated as additional gold standard volumes. Gold standard volumes were created within MilxView, an open-sourced image manipulation and processing platform developed by the Commonwealth Scientific and Industrial Research Organisation (CSIRO) biomedical informatics group (Burdett et al 2010).

Phys. Med. Biol. 63 (2018) 035001 (14pp)


3

D Roach et al

2.3. Treatment planningThe initial trial contained three patient datasets with corrupt DICOM structure set files, while an additional four patients were removed due to incorrect CTV delineation by a single observer (three included seminal vesicles within the CTV, one patient had prostate bed contoured). This resulted in thirty-five patient datasets being imported into Pinnacle3® for the first trial, and five patient datasets imported during the second trial.

VMAT treatment plans (78 Gy to PTV) incorporating gold standard contours were initially generated for each patient using Pinnacle3®’s Autoplanning module, and were assessed for quality by an experienced radiation therapist. Treatment plans considered clinically unacceptable had dose objectives manually adjusted and were resimulated, until all gold standard treatment plans were accepted in line with department prostate planning protocol (supplementary table 1 (stacks.iop.org/PMB/63/035001/mmedia)). VMAT treatment plans for each set of observer contours were subsequently generated using dose objectives matching each patient’s gold standard treatment plan.

2.4. Contouring similarity metricsStructure DICOM files were exported from Pinnacle3®, and converted into NifTI files within MilxView (Burdett et al 2010). Volumetric, statistical, and boundary similarity metrics, summarised in table 1, were calculated with respect to the corresponding gold standard contour for bladder, rectum, CTV, and PTV. Derivations for contouring similarity metrics are included in table 2.

2.5. Dosimetry and radiobiological analysisDose-volume histograms (DVHs) of gold standard volumes for all treatment plans were exported from Pinnacle3®, with radiobiological metrics for target volumes and OARs (table 1) calculated using in-house developed software Comp Plan (Holloway et al 2012). The difference between metrics calculated from gold standard and observer treatment plans provided a measure of the impact on dosimetry due to observer contouring variability.

2.6. Statistical considerationsAll statistical analysis was completed within MATLAB R2015b (The Mathworks Inc., Natick, MA). Spearman’s non-parametric rank correlation coefficient (ρ) was used to assess correlations between contouring similarity and radiobiological metrics. As 12 contouring similarity metrics were included in this study, and either 12 (CTV, PTV) or 16 (bladder, rectum) radiobiological metrics considered, Bonferroni corrections of 144 for CTV and PTV, and 192 for bladder and rectum were applied. Subsequently, p-values of p < 0.000 35 and p < 0.000 26 respectively were now considered significant. With 105 observer treatment plans, α = 0.05 (two-sided) and 1 − β = 0.9, the initial trial possessed the statistical power to detect correlations of |ρ| > 0.3. The second trial, with 50 observer treatment plans, could detect correlations of |ρ| > 0.4. Table 3 outlines the general interpretation of the strength of Spearman’s correlations within biomedical sciences, where the sign of ρ signifies the direction

of the correlation (Hinkle et al 2003).Final analysis of the contour datasets was then undertaken as detailed by Fotina et al, where the intra-class

correlation coefficient (ICC) was utilised to assess whether each trial possessed the number of observers required to result in a minimum acceptable level of study reliability (Fotina et al 2012).

Table 1. Contouring similarity and radiobiological metrics.

Contouring similarity metric Radiobiological metric

Dice similarity coefficient (DSC) Tumour control probability—poisson model (TCPPoisson)a

Volume similarity Tumour control probability—logit model (TCPLogit)a

Average relative volume difference Normal tissue complication probability (NTCP)b

Sensitivity Equivalent uniform dose (EUD)

Specificity Minimum dosea

C-Factor (Popovic et al 2007) Mean dose

Mean absolute surface distance Maximum doseb

95% Hausdorff distance Isodose volumes (IsoX)

Centroid distances Dose volume levels (DX)

Dose homogeneitya

a CTV, PTV only.b Bladder, rectum only.

Phys. Med. Biol. 63 (2018) 035001 (14pp)


4

D Roach et al

3. Results

3.1. Trial 1 resultsObserver and STAPLE volumes for CTV, PTV, bladder, and rectum are plotted in figure 1. Mean, standard deviation (SD), and coefficient of variation (COV) of these volumes are summarised in table 4. Observer B consistently contoured larger CTV (+11.03 cc), PTV (+22.44 cc), and rectum (+19.09 cc), although only CTV and PTV were considered statistically significant (figure 1). Minimal variations were seen between observers A, C, and STAPLE volumes for CTV, PTV, and rectum, and between all observer and STAPLE bladder volumes. Additional descriptive statistics can be found in supplementary tables 2 and 3. Figure 2 plots DSC for observer contours with respect to the corresponding STAPLE contour. The poorest overlapping PTV contour (DSC = 0.7612) is outlined purple in figure 3.

Mean differences between observer and STAPLE plan dosimetry for selected radiobiological metrics are shown in table 5. Negligible variations in CTV dosimetry were observed. Significant variations in dosimetry were seen for PTV, with several observer treatment plans failing to adequately treat the STAPLE volume (D95 < 76.2 Gy, supplementary table 1). One such treatment plan utilised observer C’s contours (orange) in figure 4, where observer C’s smaller PTV resulted in a significant portion of the STAPLE PTV receiving inadequate dose.

Table 2. Contouring similarity metric derivations.

Metric Equations and derivations

Dice similarity coefficient (DSC) DSC = 2|X∩Y||X|+|Y|

Volume similarity VOLSIM = Y−X(X+Y)/2

True positive (negative) Number of voxels lying within (outside) the observer contour that also lie within

(outside) the gold standard contour

False positive (negative) Number of voxels lying within (outside) the observer contour that lie outside

(within) the gold standard contour

Sensitivity p = True PositivesTrue Positives+False Negatives

Specificity q = True NegativesTrue Negatives+False Positives

Mean absolute surface distance MASD = 1NXS+NYS

(∑

x∈XS

miny∈YS

d (x, y) +∑

y∈YS

minx∈XS

d (y, x)

)

95% Hausdorff distance HDasym (XS, YS) = 95th percentilex∈XS

(miny∈YS

d(x, y))

HD (XS, YS) = max(HDasym (XS, YS) , HDasym(YS, XS))

Centroid (euclidean) Euclidean distance between centre-of-mass for gold standard and observer

Centroid (sagittal plane) Distance in sagittal plane between centre-of-mass for gold standard and observer

Centroid (coronal plane) Distance in coronal plane between centre-of-mass for gold standard and observer

Centroid (axial plane) Distance in axial plane between centre-of-mass for gold standard and observer

Absolute relative volume difference aRVD = |100 ×(

|X||Y| − 1

)|

C-Factor (Popovic et al 2007) d = 2p(1−q)p+(1−q) +

2(1−p)q(1−p)+q

C =

d, p � q̂ p > 1 − q

−d, p < q̂ p > 1 − q

undefined, p � 1 − q

X: Number of voxels within gold standard contour.

Y: Number of voxels within observer contour.

XS: Surface points of gold standard contour.

YS: Surface points of observer contour.

NXS: Number of surface points of gold standard contour.

NYS: Number of surface points of observer contour.

d(x,y): Euclidean distance from point x to point y.

Table 3. Strength of spearman’s correlation ρ.

Spearman’s |ρ| Strength of correlation

0.90–1.00 Very strong correlation

0.70–0.90 Strong correlation

0.50–0.70 Moderate correlation

0.30–0.50 Weak correlation

0.00–0.30 Negligible correlation

Phys. Med. Biol. 63 (2018) 035001 (14pp)


5

D Roach et al

Correlations between contouring similarity and radiobiological metrics were investigated using Spear-man’s ρ, with complete lists of correlations found in supplementary tables 4–7. No significant correlations were observed for CTV and bladder. Significant correlations for PTV and rectum are displayed in figure 5. Volume similarity exhibited the strongest correlation to dosimetry for both PTV and rectum. The strongest correlating radiobiological metrics for PTV volume similarity were minimum dose (ρ = 0.67), TCPPoisson (ρ = 0.57), dose homogeneity (ρ = −0.52), EUD (ρ = 0.43), and TCPLogit (ρ = 0.39). The strongest correlating radiobiological metrics for rectum volume similarity were mean dose (ρ = 0.45) and EUD (ρ = 0.43), with similar correlations for absolute relative volume difference also recorded (ρ = 0.46 and 0.38 respectively). No significant correlations between contouring similarity metrics and maximum dose or NTCP for the rectum were observed.

Figure 1. Observer and STAPLE volume spread for each structure across all 35 patients for trial 1. Notch box-plots show a significant difference in CTV and PTV median volumes by observer (B) compared to observers (A) and (C). Differences between observer median rectum volumes were found to be not statistically significant.

Table 4. Mean, standard deviation, and COV of structure volumes across trial 1 patient cohort.

Observer CTV PTV Bladder Rectum

A Mean (cc) 40.13 107.53 289.98 71.50

SD (cc) 19.32 37.97 141.24 33.50

COV 0.48 0.35 0.49 0.47

B Mean (cc) 53.74 135.07 285.72 95.51

SD (cc) 24.66 45.01 138.09 46.38

COV 0.46 0.33 0.48 0.49

C Mean (cc) 40.74 109.59 297.99 73.35

SD (cc) 21.17 41.54 144.15 40.17

COV 0.52 0.38 0.48 0.55

STAPLE Mean (cc) 42.71 112.63 291.63 76.42

SD (cc) 21.30 40.98 141.51 39.05

COV 0.50 0.36 0.49 0.51

Figure 2. Spread in observer dice similarity coefficient (DSC) for CTV, PTV, bladder, and rectum across all patients for trial 1.

Phys. Med. Biol. 63 (2018) 035001 (14pp)


6

D Roach et al

Sensitivity and specificity correlations (table 2) were comparable to volume similarity for rectum, but weaker and correlated with fewer radiobiological metrics for PTV. Unsurprisingly the C-Factor, described as a trade-off between sensitivity and specificity (Popovic et al 2007), exhibited similar correlations to these metrics. Significant correlations were observed between differences in centroid in the sagittal plane and minimum dose (ρ = 0.45) for PTV. Overlap metrics (DSC) and boundary metrics (mean absolute surface distance and 95% Hausdorff distance) showed no significant correlations with dosimetry for PTV, while DSC only showed minimal correla-tions (ρ = −0.35–0.38) with multiple rectum isodose curves (supplementary table 5). Analysis of additional gold standard volumes found the same similarity metrics exhibiting the strongest correlations with dosimetry (supplementary material tables 8–11).

Figure 3. Patient 32 containing the poorest overlapping PTV contour (DSC = 0.7612) on T2 MR. Observer (A) (red), (B) (purple), and (C) (orange) contours are outlined on (clockwise from top) transverse, coronal, and sagittal images. STAPLE volume is shaded light blue.

Table 5. Mean differences between observer and STAPLE plan dosimetry.

ΔTCPPoisson ΔTCPLogit ΔNTCP ΔEUD

Mean SD Mean SD Mean SD Mean SD

CTV −0.0001 0.0007 −0.0009 0.0062 — — −0.07 0.46

PTV −0.0545 0.2103 −0.0035 0.0107 — — −0.74 3.39

Bladder — — — — 0.0001 0.0006 0.86 1.95

Rectum — — — — 0.0042 0.0121 0.45 1.24

Figure 4. Patient 1 PTV contours on T2 MR. All observer contours recorded DSC > 0.9 with respect to the STAPLE volume shaded in light blue. However, due to significant portions of Observer C’s (orange) PTV failing to include the STAPLE volume, insufficient dose was delivered to the STAPLE PTV for these treatment plans.

Phys. Med. Biol. 63 (2018) 035001 (14pp)


7

D Roach et al

Correlations between PTV contouring similarity metrics and OAR dosimetry were also investigated. No sig-nificant correlations were observed for bladder dosimetry. Figure 6 plots PTV volume similarity against rectum NTCP, where a weak correlation (ρ = 0.33) was observed.

3.2. Trial 2 resultsFigure 7 plots CTV, PTV, bladder and rectum volumes for trial 2 datasets, while spread in DSC is illustrated in figure 8. Mean, standard deviation, and COV for these volumes are again summarised in table 6, with additional descriptive statistics found in supplementary tables 12 and 13. Differences between observer and majority voting plan dosimetry are shown in table 7. Patient 4 recorded the smallest mean CTV volume (32.88 cc), while patient 5 had the largest mean CTV (74.02 cc).

Similar trends were observed for PTV volumes (patient 4: 99.61 cc, patient 5: 177.41 cc). Differences between mean rectum volumes between patients was less pronounced, ranging from patient 2 (37.69 cc) through to patient 5 (58.94 cc). Bladder volumes varied significantly between patients, ranging from patient 1 (69.27 cc) through to patient 3 (400.13 cc). Figure 9 shows all observer PTV contours for patient 4, as well as the majority vote PTV shaded in light blue. Observer 1 (green contour) recorded the poorest overlap (DSC = 0.8167) of all PTV volumes.

Table 7 again shows significant differences in PTV dosimetry, due to multiple observer PTV contours failing to adequately cover the majority vote PTV. Mean rectum dosimetry was marginally decreased for observer treat-ment plans, while mean bladder dosimetry improved compared to the gold standard treatment plan. Observer A’s PTV from figure 9 is an example of a PTV failing to treat the majority vote PTV, with the corresponding treat-ment plan shown in figure 10. Complete lists of correlations between contouring similarity metrics and radio-biological metrics are given in supplementary tables 14–17.

No significant correlations between contouring similarity metrics and dosimetry were observed for either CTV or rectum. Bladder correlations were quite variable, a shifting of the coronal plane correlated moderately with NTCP (ρ = −0.51) and multiple isodose curves. Additionally, both DSC (ρ = 0.62) and aRVD (ρ = -0.54) were found to correlate with the maximum dose delivered to the bladder.

Figure 5. Significant spearman correlations for PTV (top image, p < 0.000 35) and rectum (bottom image, p < 0.000 26) for trial 1. Volume similarity, sensitivity, specificity and C-Factor significantly correlated with a range of radiobiological metrics for both structures. Most correlations identified were weak, although TCPPoisson, minimum dose, and dose homogeneity showed moderate correlations with volume similarity, sensitivity, and specificity for PTV.

Figure 6. PTV volume similarity versus rectum NTCP, ρ = 0.33.

Phys. Med. Biol. 63 (2018) 035001 (14pp)


8

D Roach et al

Specificity displayed the strongest correlations with dosimetry for PTV, with strong correlations observed for TCPPoisson (ρ = 0.81), TCPLogit (ρ = 0.87), EUD (ρ = 0.85), and dose homogeneity (ρ = −0.82). Comparatively, while multiple radiobiological correlations for sensitivity were significant, they were only moderate in strength (ρ = −0.57, −0.52, and 0.60 for TCPLogit, EUD, and dose homogeneity respectively). C-Factor correlations there-

Figure 7. Variations in contoured volumes for each structure across the five patients in trial 2. Gold standard volumes for each structure are shown as a black cross. Due to the use of a majority vote, each gold standard volume is by definition smaller than the median volume for each structure.

Figure 8. Spread in observer DSC for CTV, PTV, bladder, and rectum across all patients for trial 2.

Table 6. Mean, standard deviation, and COV of observer structure volumes for trial 2 patients.

Patient CTV PTV Bladder Rectum

1 Mean (cc) 54.43 141.99 69.27 51.13

SD (cc) 6.58 14.42 4.13 6.20

COV 0.12 0.10 0.06 0.12

Majority vote (cc) 51.32 131.79 68.24 46.23

2 Mean (cc) 43.38 118.88 144.81 37.69

SD (cc) 6.13 13.51 9.03 6.81

COV 0.14 0.11 0.06 0.18

Majority vote (cc) 42.13 113.44 143.33 32.89

3 Mean (cc) 52.64 133.17 400.13 38.61

SD (cc) 7.96 15.37 7.38 2.78

COV 0.15 0.12 0.01 0.07

Majority vote (cc) 49.92 127.59 396.46 36.19

4 Mean (cc) 32.88 99.61 143.21 54.79

SD (cc) 6.59 15.68 12.08 9.95

COV 0.20 0.16 0.08 0.18

Majority vote (cc) 32.81 96.41 142.10 56.35

5 Mean (cc) 74.02 177.41 84.91 58.94

SD (cc) 11.14 20.12 3.82 6.82

COV 0.15 0.11 0.05 0.12

Majority vote (cc) 70.75 165.25 84.11 55.96

Phys. Med. Biol. 63 (2018) 035001 (14pp)


9

D Roach et al

fore matched the poorer performing sensitivity correlations. Volume similarity again correlated strongly with a range of radiobiological metrics for PTV, with the strongest correlations being dose homogeneity (ρ = −0.76), TCPLogit (ρ = 0.70), EUD (ρ = 0.67), TCPPoisson (ρ = 0.65), and minimum dose (ρ = 0.61). These correlations are illustrated in figure 11.

Table 7. Mean differences between trial 2 observer and majority vote plan dosimetry.

ΔTCPPoisson ΔTCPLogit ΔNTCP ΔEUD

Mean SD Mean SD Mean SD Mean SD

CTV 0.0000 0.0001 −0.0014 0.0035 — — −0.11 0.27

PTV −0.3498 0.4547 −0.0150 0.0380 — — −4.34 11.35

Bladder — — — — 0.0003 0.0009 0.44 2.75

Rectum — — — — −0.0022 0.0248 −0.20 2.23

Figure 9. Patient 4 PTV contours on (clockwise from top) transverse, coronal, and sagittal images. Observer A (highlighted dark green) displayed the poorest overlapping PTV with respect to the majority vote PTV (shaded light blue), with a DSC of 0.8167.

Figure 10. Patient 4 dose distribution derived from observer A’s PTV contour (figure 9). The majority vote PTV is outlined in light blue, while the 78 Gy, 50 Gy, and 39 Gy isodose lines are shaded light green, red, and light blue respectively. This treatment plan resulted in zero tumour control probability for the majority vote PTV. It can clearly be seen on the sagittal and coronal slices that significant portions of the majority vote PTV were under-dosed during treatment planning.

Phys. Med. Biol. 63 (2018) 035001 (14pp)


10

D Roach et al

Correlations between PTV volume similarity and rectum NTCP were again investigated, where a stronger (yet still weak) correlation of ρ = 0.48 was observed.

4. Discussion

Following treatment plan simulation, it was found that volumetric metrics (volume similarity, sensitivity, and specificity) assessing inter-observer contouring variation for PTV routinely correlated significantly with multiple radiobiological metrics. This was assessed using two separate trials, the first comprising of a small number of observers and large number of patients, the second with a large number of observers and small number of patients, ensuring the study captured the breadth of contouring variation that would be observed clinically. Additionally, variations in volume similarity when assessing PTV contours were found to moderately correlate with rectum NTCP. This study provides evidence linking inter-observer contouring variation metrics with dosimetry and treatment plan quality for prostate cancer patients.

Vinod et al cited 16 studies investigating inter-observer contouring variation for GU structures, identifying the most popular similarity metrics for these studies as volume, surface, overlap, and centre-of-mass respectively (Vinod et al 2016a). Only a few studies included an assessment on dosimetry (Foppiano et al 2003, Livsey et al 2004, Mitchell et al 2009, Perna et al 2011), with these studies investigating impact on OAR dosimetry opposed to treatment target volumes. As stated by Vinod et al, an assessment of contouring variation based solely on geom-etry may have little to no clinical significance, especially where no correlations between contouring metrics and dosimetry have been performed. Additionally, commonly cited spatial overlap metrics conformity index and DSC have been shown to display only minimal to no significant correlations with dosimetry for non-small cell lung cancers (Jameson et al 2014) and head and neck cancers (Beasley et al 2016) respectively. As no study cor-relating contouring similarity metrics and dosimetry had been performed for prostate radiotherapy, this study was undertaken to bring clinical relevancy and allow insights into target volume and OAR dosimetry to other inter-observer contouring variation studies.

Two trials were incorporated in this study; the first involving a small number of observers with many patients, the second a large number of observers with a small number of patients. This multiple trial analysis allowed for a technique to determine whether the calculated correlations were invariant to the trial setup. Fotina et al introduced a method for calculating the minimum number of observers required within an inter-observer con-touring study in order for a minimal acceptable level of study reliability to be reached (Fotina et al 2012). This is derived from the intraclass correlation coefficient (ICC), and suggests that a minimum value of 0.8 for the ICC is required. Table 8 lists the calculated ICC and corresponding minimum number of observers required for each structure across both trials. It was shown that each trial satisfied this criterion, with the three observers used within the first trial considered sufficient. It should be noted that, due to the initial derivation of ICC requiring knowledge of the number of observers, the derived number of observers is meant only to assess whether the trial’s observer numbers are sufficient to produce reliable results. It does not give the recommended number of

observers required for contouring analysis within the study.No similarity metrics assessing CTV contouring variations correlated significantly with any radiobiological

metric across both trials. As department treatment planning required uniform PTV dose distributions, varia-tions in CTV dose distributions were negligible (tables 5 and 7), suggesting that the PTV margins applied ade-quately accounted for inter-observer CTV contouring variation. Within the initial trial substantial agreement between observer bladder contours was observed (figures 1 and 2), resulting in minimal variations in bladder dosimetry between observer treatment plans. Consequently, the insufficient spread in data prevented significant correlations for bladder being observed. Only when additional observers were included during the second trial were correlations between contouring variation metrics and dosimetry observed.

Figure 11. Significant spearman correlations for PTV with p < 0.000 35 for trial 2. Correlations for PTV were much stronger than those observed in trial 1 (figure 5), ranging from moderate (sensitivity, C-Factor) to strong (volume similarity, specificity).

Phys. Med. Biol. 63 (2018) 035001 (14pp)


11

D Roach et al

Curiously, additional observers within the second trial resulted in the loss of significance in correlations between contouring and radiobiological metrics for the rectum. A possible explanation could be the wider range of slices contoured by observers when defining the superior portion of the rectum during the second trial. Within the first trial all three observers contoured the rectum on approximately equivalent slices. Consequently, the variations in contouring were more significant within each slice, and hence within regions where higher dose would have been administered. As these correlations were only weak (table 3), they were subsequently masked when larger contouring variations were introduced during the second trial due to the variable number of slices contoured, coinciding with a region receiving less dose.

DSC and 95% Hausdorff distance are commonly quoted contouring similarity metrics within the literature, however in this study these metrics exhibited little or no significant correlation with dosimetry for all structures. An explanation for this is the inability for these metrics to differentiate between observer contours that lie either within or outside the gold standard volume. Figure 4 illustrates this. For this patient, all three observer PTV contours recorded DSC scores greater than 0.9, typically considered as excellent overlap (Zijdenbos et al 1994). Adequate treatment plans for the STAPLE target volumes were simulated based on observer A and B contours. However, due to observer C’s PTV lying entirely within the STAPLE volume, significant regions of the STAPLE PTV received inadequate dose coverage during treatment planning. Consequently, markedly different treat-ment outcomes were simulated between the observer treatment plans, despite nearly identical DSC values being recorded. This is also observed in figures 9 and 10, where the smaller PTV from observer 1 resulted in inadequate coverage of the gold standard PTV. Volume similarity, sensitivity, and specificity possess the ability to differenti-ate between contours lying within or outside the gold standard volume, consequently these similarity metrics correlated with target volume dosimetry.

A concern when utilising volume similarity may be the lack of spatial information this metric provides, such that two hypothetical observer contours could each have equal volume, but share no spatial overlap. However, assuming adequate scan resolution, this hypothetical should never occur between two experienced observers for structures considered in this study. Inspection of the patient datasets found that variations in contouring for rectum, bladder, and CTV were found around the perimeter of the contours, never in the localisation of the structures. Another point of contention within this study was the choice of either STAPLE or majority vote as reference gold standard volumes for each trial. Additional analysis involving multiple gold standard volumes uti-lising the trial 1 dataset produced equivalent correlations to those seen in trial 1 (see supplementary tables 8–11).

Larger PTV volumes compromises dose sparing to nearby OARs, with a statistically significant correlation between PTV volume similarity and rectum NTCP observed in both trials (ρ = 0.33, 0.48), and shown for trial 1 in figure 6. A previous study investigating the impact of CTV and PTV contouring variations on rectum dosim-etry was unable to observe statistically significant correlations (Livsey et al 2004). This discrepancy may be due to their study utilising 3D-CRT treatment plans, opposed to VMAT treatment planning incorporated in this study. VMAT treatment plans generate steeper dose gradients around the target volumes, increasing the sensitivity of target volume dosimetry to inter-observer contouring variations (also shown in figure 10). It is anticipated that future studies investigating SBRT treatment plans, where higher dose fractions are delivered, may yield stronger correlations than were observed in this study.

Within the initial trial observer B consistently contoured significantly larger CTV and PTV volumes, result-ing in this observer’s bias being over-represented during analysis. When investigating the cause of this discrep-ancy, it was found that observer A and C had prior experiences working with one another on prostate delineation projects. Consequently, both observers interpreted and contoured structures in this study in a similar fashion. This illustrates the importance of peer review during contouring for clinical trials, although as of yet no study has investigated the statistical impact of peer review on inter-observer contouring variation (Vinod et al 2016b). Contouring is routinely regarded as one of the highest priorities for peer review during a clinical trial (Brundage et al 2013, Marks et al 2013), with surveys showing up to 59% of observer contours undergo peer review (Hoopes et al 2015). For these reasons, it was decided that a second trial arm should be undertaken, to ensure that the range of contouring variation observed could be assumed to be representative of those found in general clinical practice.

Table 8. ICCs and minimum required number of observers for study reliability.

Trial 1 Trial 2

ICC Observers ICC Observers

CTV 0.9362 3 (2.45) 0.9752 1 (0.54)

PTV 0.9345 3 (2.99) 0.9719 1 (0.66)

Bladder 0.9969 1 (0.03) 0.9979 1 (0.03)

Rectum 0.9345 2 (1.82) 0.9495 1 (0.87)

Phys. Med. Biol. 63 (2018) 035001 (14pp)


12

D Roach et al

The second trial arm included contours from multiple volunteers across a range of treatment centres. The large number of observers was required for an ongoing study, however included medical physicists and radiogra-phers who lacked experience in contouring. As poor clinical training within this study could produce results not relevant to standard clinical practise, additional analysis utilising only the radiation oncologist and radiothera-pist contours was performed. Adjusted CTV, PTV, bladder and rectum mean volumes (52.38 cc, 135.83 cc, 168.50 cc, 48.98 cc) and DSC (0.8976, 0.9190, 0.9561, 0.8807) across the five patients were comparable to the complete dataset (51.85 cc, 134.92 cc, 168.98 cc, 48.10 cc and 0.8872, 0.9118, 0.9530, 0.8724 respectively). Wilcoxon rank sum testing revealed no significant difference (p < 0.05) between either dataset. Additional analysis of correla-tions using this subset resulted in near equivalent correlations for PTV between volume similarity, sensitivity, specificity, and the C-Factor with multiple radiobiological metrics (supplementary tables 18–21). It can there-fore be concluded that the lack of experience of some observers did not impact the findings of this study.

Additionally, as contouring within the second trial arm was performed on two separate platforms (Eclipse™ and Pinnacle3®), there was potential for bias to be introduced within the study. Analysis of CTV, PTV, blad-der, and rectum volumes and DSC overlap for Eclipse users (52.11 cc, 137.49 cc, 170.91 cc, 47.62 cc and 0.9013, 0.9228, 0.9554, 0.8913 respectively) compared to Pinnacle users (51.60 cc, 132.45 cc, 167.13 cc, 48.56 cc and 0.8737, 0.9013, 0.9507, 0.8543 respectively) revealed slightly improved overlap for CTV, PTV and Rectum from Eclipse users. Wilcoxon rank sum testing confirmed that distribution of DSC values for these structures differed significantly (p < 0.05) between the two treatment planning systems. While this confirmed a bias in spatial over-lap dependent on treatment planning system (and consequently, treatment centre), additional analysis where either only Pinnacle or Eclipse observer contours were investigated again revealed identical correlations between contouring and radiobiological metrics that were statistically significant. Consequently, this bias was deemed by the authors not to be impactful to the studies aims.

Treatment plan generation was another potential study limitation, as poor quality treatment planning could mask the impact inter-observer contouring variability has on dosimetry. By utilising the Autoplanning mod-ule within Pinnacle3®, and having treatment plans subsequently assessed by an experienced radiation therapist, treatment plans of similar quality satisfying department planning protocol were ensured across all observer con-tour sets.

Finally, it should be noted that a key difference between the two trial arms was the difference in scans used by observers for contouring. Trial 1 required contouring on T2-weighted MR scans, that were subsequently regis-tered and fused to planning CT, while trial 2 required contouring to be performed directly on CT. However, as gold standard volumes were constructed from observer contours, differences in CTV and PTV volume typically observed between CT and MR scans in prostate contouring studies (Debois et al 1999) would be reflected in dif-ferences in gold standard volumes between the trials. As the study was investigating differences between observer contours with respect to these gold standard volumes, the impact of differing CTV and PTV volumes between the trials were compensated for during the analysis. Additionally, figures 2 and 6 reveal the spread in DSC across all structures for both trials, where similar trends were observed. These DSC values lie within the range typically seen within clinical trials (Sharp et al 2014), consequently the different imaging modality used by each trial was again felt not to be impactful to the study.

It is important to consider the clinical impact, and not just effect on dosimetry, that inter-observer contour-ing variations are responsible for. Incorrect contouring has been shown to be a significant cause of poor quality radiotherapy delivered in both pancreatic (Abrams et al 2012) and head and neck cancer trials (Peters et al 2010). With the latter it was found that poor quality treatment plans, from which 25% were attributed to poor qual-ity contouring, were responsible for up to 20% reduction in both locoregional failure-free control and overall patient survival. Additional meta-analyses of clinical trials across multiple treatment sites concluded that radi-otherapy protocol deviations were associated with both reduced treatment efficacy and patient survival rates (Weber et al 2012, Ohri et al 2013). In each meta-analysis, poor quality contouring was again concluded to be a significant factor for protocol deviations.

Additionally, the use of large rectangular treatment fields for prostate radiotherapy resulted in fewer clinical failures compared to the tightly conformal fields used in intensity-modulated radiotherapy and VMAT (Heems-bergen et al 2013). It was concluded that increased dose delivered to regions just outside the defined prostate, where subclinical spread of disease could be present, were partly responsible for this increase in treatment effi-cacy. Consequently, the accuracy of contouring is paramount in ensuring adequate dose coverage across the entire treatment volume for radiotherapy, and an assessment of accuracy based on treatment efficacy is required to relate contouring variations with clinical relevancy. Thus, this study is not just of interest dosimetrically, but has direct clinical relevance in improving the largest source of uncertainty in radiation treatment (Weiss and Hess 2003).

As well as inter-observer contouring studies, contouring similarity metrics are routinely utilised during atlas development and validation (Acosta et al 2014). Automatically contoured structures generated by an atlas for a query patient must be assessed for accuracy and precision, which usually occurs using the clinician’s original contour and commonly utilised similarity metrics DSC and Hausdorff distance. A recent proof of concept study

Phys. Med. Biol. 63 (2018) 035001 (14pp)


13

D Roach et al

found that inclusion of dosimetry assessment during automatic contouring of the prostate improved rectum dose sparing and dosimetry (Chang et al 2017). Importantly, these improvements came despite the automatically con-toured structures recording nearly identical DSC and Hausdorff Distances to contours generated with no con-sideration of dosimetry. While these results were used to validate their automatic contouring algorithm, it also suggested that these metrics were unable to differentiate between contoured structures with varying dosimetry.

Volume similarity, sensitivity, and specificity significantly correlated with PTV dosimetry across both trials for prostate cancer radiotherapy. Conversely, correlations for rectum and bladder were only observed during trials 1 and 2 respectively. Due to correlations for these structures being only weak to moderate in magnitude, not being replicable across both trials, and due to no correlations for CTV being observed in either trial, a com-bination of contouring similarity metrics should still be cited during future inter-observer contouring variation studies. Fotina et al recommend reporting a combination of overlap and statistical measures of agreement during analysis (Fotina et al 2012). This study additionally advocates the reporting of volume similarity, as this would allow inter-observer contouring variation studies for prostate cancer radiotherapy to provide additional infor-mation on PTV dosimetry.

5. Conclusion

This study is the first to show statistically significant correlations between inter-observer contouring variations for prostate cancer radiotherapy and simulated patient dosimetry. Multiple significant correlations were observed between volume similarity, sensitivity, and sensitivity, and PTV dosimetry during both trials. Correlations between contouring similarity metrics and bladder and rectum dosimetry across the two trials were more variable, however variations in contouring PTV significantly correlated with differences in rectum dosimetry. No significant correlations between contouring similarity metrics and CTV dosimetry were observed. This study will greatly enhance future inter-observer contouring variation studies for prostate cancer radiotherapy, guiding contouring similarity metric choice to allow for insights into dosimetry and clinical relevancy during analysis.

Acknowledgment

The authors would like to thank Dr Wei Xuan for statistical advice, and Kirrily Cloak and Rohan Gray for treatment planning feedback. Additional thanks is given to Michelle Krawiec, Robba Rai, Professor Jim Denham, Dr Jeremiah De Leon, Dr Karen Lim, Dr Megan Berry, Dr Rohen White, Professor Sean Bydder, Dr Hendrick Tan, Dr Jeremy Croker, Dr Alycea McGrath, Dr Robert Jan Smeenk, and Dr John Matthews. The project was funded by NHMRC project grant number 1077788.

Conflict of interest

None.

ORCID iDs

D Roach https://orcid.org/0000-0002-5541-5217

References

Abrams R A, Winter K A, Regine W F, Safran H, Hoffman J P, Lustig R, Konski A A, Benson A B, Macdonald J S and Rich T A 2012 Failure to adhere to protocol specified radiation therapy guidelines was associated with decreased survival in RTOG 9704—a phase III trial of adjuvant chemotherapy and chemoradiotherapy for patients with resected adenocarcinoma of the pancreas Int. J. Radiat. Oncol. Biol. Phys. 82 809–16

Acosta O, Dowling J, Drean G, Simon A, De Crevoisier R and Haigron P 2014 Multi-atlas-based segmentation of pelvic structures from CT scans for planning in prostate cancer radiotherapy Abdomen and Thoracic Imaging (Berlin: Springer) pp 623–56

Al-Mamgani A, Heemsbergen W D, Peeters S T and Lebesque J V 2009 Role of intensity-modulated radiotherapy in reducing toxicity in dose escalation for localized prostate cancer Int. J. Radiat. Oncol. Biol. Phys. 73 685–91

Beasley W J, McWilliam A, Aitkenhead A, Mackay R I and Rowbottom C G 2016 The suitability of common metrics for assessing parotid and larynx autosegmentation accuracy J. Appl. Clin. Med. Phys. 17 41–9

Brundage M, Foxcroft S, McGowan T, Gutierrez E, Sharpe M and Warde P 2013 A survey of radiation treatment planning peer-review activities in a provincial radiation oncology programme: current practice and future directions BMJ Open 3 e003241

Burdett N, Fripp J, Bourgeat P and Salvado O 2010 MILXView: a medical imaging, analysis and visualization platform E-Health (Berlin: Springer) pp 177–86

Chang J, Tian Z, Lu W, Gu X, Chen M and Jiang S B 2017 A novel geometry-dosimetry label fusion method in multi-atlas segmentation for radiotherapy: a proof-of-concept study Phys. Med. Biol. 62 3656

Debois M, Oyen R, Maes F, Verswijvel G, Gatti G, Bosmans H, Feron M, Bellon E, Kutcher G and Van Poppel H 1999 The contribution of magnetic resonance imaging to the three-dimensional treatment planning of localized prostate cancer Int. J. Radiat. Oncol. Biol. Phys. 45 857–65

Phys. Med. Biol. 63 (2018) 035001 (14pp)


14

D Roach et al

Dowling J A, Sun J, Pichler P, Rivest-Hénault D, Ghose S, Richardson H, Wratten C, Martin J, Arm J and Best L 2015 Automatic substitute computed tomography generation and contouring for magnetic resonance imaging (MRI)-alone external beam radiation therapy from standard mri sequences Int. J. Radiat. Oncol. Biol. Phys. 93 1144–53

Dubois D F, Prestidge B R, Hotchkiss L A, Prete J J and Bice W S Jr 1998 Intraobserver and interobserver variability of MR imaging-and CT-derived prostate volumes after transperineal interstitial permanent prostate brachytherapy Radiology 207 785–9

Foppiano F, Fiorino C, Frezza G, Greco C, Valdagni R and AIRO National Working Group on Prostate Radiotherapy 2003 The impact of contouring uncertainty on rectal 3D dose–volume data: results of a dummy run in a multicenter trial (AIROPROS01–02) Int. J. Radiat. Oncol. Biol. Phys. 57 573–9

Fotina I, Lütgendorf-Caucig C, Stock M, Pötter R and Georg D 2012 Critical discussion of evaluation parameters for inter-observer variability in target definition for radiation therapy Strahlenther. Onkol. 188 160–7

Frey B J and Dueck D 2007 Clustering by passing messages between data points Science 315 972–6Gao Z, Wilkins D, Eapen L, Morash C, Wassef Y and Gerig L 2007 A study of prostate delineation referenced against a gold standard created

from the visible human data Radiother. Oncol. 85 239–46Heemsbergen W D, Al-Mamgani A, Witte M G, van Herk M and Lebesque J V 2013 Radiotherapy with rectangular fields is associated with

fewer clinical failures than conformal fields in the high-risk prostate cancer subgroup: results from a randomized trial Radiother. Oncol. 107 134–9

Hinkle D E, Wiersma W and Jurs S G 2003 Applied Statistics for the Behavioral Sciences (Boston, MA: Houghton Mifflin)Holloway L C, Miller J-A, Kumar S, Whelan B M and Vinod S K 2012 Comp plan: a computer program to generate dose and radiobiological

metrics from dose-volume histogram files Med. Dosim. 37 305–9Hoopes D J, Johnstone P A, Chapin P S, Kabban C M S, Lee W R, Chen A B, Fraass B A, Skinner W J and Marks L B 2015 Practice patterns for

peer review in radiation oncology Pract. Radiat. Oncol. 5 32–8Hodapp N 2012 The ICRU Report 83: prescribing, recording, and reporting photon-beam intensity-modulated radiation therapy (IMRT)

Strahlenther. Onkol. 188 97–99Jameson M G, Holloway L C, Vial P J, Vinod S K and Metcalfe P E 2010 A review of methods of analysis in contouring studies for radiation

oncology J. Med. Imaging Radiat. Oncol. 54 401–10Jameson M G, Kumar S, Vinod S K, Metcalfe P E and Holloway L C 2014 Correlation of contouring variation with modeled outcome for

conformal non-small cell lung cancer radiotherapy Radiother. Oncol. 112 332–6Kennedy A M, Dowling J A, Ghose S, Rivest-Hénault D and Ebert M A 2016 Cluster Based selection of CT exemplars from a clinical dataset

for inter-patient registration and dose mapping. Poster Int. Conf. on the use of Computers in Radiation Therapy (London)Khoo E L, Schick K, Plank A W, Poulsen M, Wong W W, Middleton M and Martin J M 2012 Prostate contouring variation: can it be fixed?

Int. J. Radiat. Oncol. Biol. Phys. 82 1923–9Livsey J E, Wylie J P, Swindell R, Khoo V S, Cowan R A and Logue J P 2004 Do differences in target volume definition in prostate cancer lead

to clinically relevant differences in normal tissue toxicity? Int. J. Radiat. Oncol. Biol. Phys. 60 1076–81Marks L B, Adams R D, Pawlicki T, Blumberg A L, Hoopes D, Brundage M D and Fraass B A 2013 Enhancing the role of case-oriented peer

review to improve quality and safety in radiation oncology: executive summary Pract. Radiat. Oncol. 3 149–56Mitchell D M, Perry L, Smith S, Elliott T, Wylie J P, Cowan R A, Livsey J E and Logue J P 2009 Assessing the effect of a contouring protocol on

postprostatectomy radiotherapy clinical target volumes and interphysician variation Int. J. Radiat. Oncol. Biol. Phys. 75 990–3Ohri N, Shen X, Dicker A P, Doyle L A, Harrison A S and Showalter T N 2013 Radiotherapy protocol deviations and clinical outcomes: a

meta-analysis of cooperative group clinical trials J. Natl Cancer Inst. 105 387–93Ost P, De Meerleer G, Vercauteren T, De Gersem W, Veldeman L, Vandecasteele K, Fonteyne V and Villeirs G 2011 Delineation of the

postprostatectomy prostate bed using computed tomography: interobserver variability following the EORTC delineation guidelines Int. J. Radiat. Oncol. Biol. Phys. 81 e143–9

Palma D, Vollans E, James K, Nakano S, Moiseenko V, Shaffer R, McKenzie M, Morris J and Otto K 2008 Volumetric modulated arc therapy for delivery of prostate radiotherapy: comparison with intensity-modulated radiotherapy and three-dimensional conformal radiotherapy Int. J. Radiat. Oncol. Biol. Phys. 72 996–1001

Perna L, Cozzarini C, Maggiulli E, Fellin G, Rancati T, Valdagni R, Vavassori V, Villa S and Fiorino C 2011 Inter-observer variability in contouring the penile bulb on CT images for prostate cancer treatment planning Radiat. Oncol. 6 1

Peters L J, O’Sullivan B, Giralt J, Fitzgerald T J, Trotti A, Bernier J, Bourhis J, Yuen K, Fisher R and Rischin D 2010 Critical impact of radiotherapy protocol compliance and quality in the treatment of advanced head and neck cancer: results from TROG 02.02 J. Clin. Oncol. 28 2996–3001

Popovic A, de la Fuente M, Engelhardt M and Radermacher K 2007 Statistical validation metric for accuracy assessment in medical image segmentation Int. J. Comput. Assist. Radiol. Surg. 2 169–81

Quan E M, Li X, Li Y, Wang X, Kudchadker R J, Johnson J L, Kuban D A, Lee A K and Zhang X 2012 A comprehensive comparison of IMRT and VMAT plan quality for prostate cancer treatment Int. J. Radiat. Oncol. Biol. Phys. 83 1169–78

Rasch C, Barillot I, Remeijer P, Touw A, van Herk M and Lebesque J V 1999 Definition of the prostate in CT and MRI: a multi-observer study Int. J. Radiat. Oncol. Biol. Phys. 43 57–66

Sharp G, Fritscher K D, Pekar V, Peroni M, Shusharina N, Veeraraghavan H and Yang J 2014 Vision 20/20: perspectives on automated image segmentation for radiotherapy Med. Phys. 41 050902

Trans-Tasman Radiation Oncology Group (TROG) 2005 RADAR trial—randomised androgen deprivation and radiotherapy (https://clinicaltrials.gov/ct2/show/study/NCT00193856)

Van Dyk J, Batista J and Bauman G S 2013 Accuracy and uncertainty considerations in modern radiation oncology Mod. Technol. Radiat. Oncol. 3 361–412

Vinod S K, Jameson M G, Min M and Holloway L C 2016a Uncertainties in volume delineation in radiation oncology: a systematic review and recommendations for future studies Radiother. Oncol. 121 169–79

Vinod S K, Min M, Jameson M G and Holloway L C 2016b A review of interventions to reduce inter-observer variability in volume delineation in radiation oncology J. Med. Imaging Radiat. Oncol. 60 393–406

Warfield S K, Zou K H and Wells W M 2004 Simultaneous truth and performance level estimation (STAPLE): an algorithm for the validation of image segmentation IEEE Trans. Med. Imaging 23 903–21

Weber D C, Tomsej M, Melidis C and Hurkmans C W 2012 QA makes a clinical trial stronger: evidence-based medicine in radiation therapy Radiother. Oncol. 105 4–8

Weiss E and Hess C F 2003 The impact of gross tumor volume (GTV) and clinical target volume (CTV) definition on the total accuracy in radiotherapy Strahlenther. Onkol. 179 21–30

Zijdenbos A P, Dawant B M, Margolin R and Palmer A C 1994 Morphometric analysis of white matter lesions in MR images: method and validation IEEE Trans. Med. Imaging 13 716–24

Phys. Med. Biol. 63 (2018) 035001 (14pp)


Table 1: Department prostate treatment planning protocol

Patient Name: Definitive Prostate +/- SV: 78Gy to D95

Structure Contour Name Constraint Per-Protocol Minor Variation Major Variation PTV78 PTV_78 D95 > 78 Gy (100%) 76.2 - 78 Gy (98-100%) < 76.2 Gy (98%)

PTV78 PTV_78 D50 78 – 81.5 Gy 81.5 - 82.0 Gy > 82 Gy

PTV78 PTV_78 D2 < 81.9 Gy (105%) 81.9 - 83.5Gy (105-107%) > 83.5Gy (107%)

CTV78 CTV_78 D99 > 78Gy (100%) 76.2 - 78Gy (98-100%) < 76.2 Gy (98%)

ANT PROSTATE CTV_ANTPROST Aim D90 ≥ 79Gy > 79 Gy 78.5 - 79 Gy < 78.5 Gy

SV CTV_SV Aim D90 ≥ 78.5Gy > 78.5 Gy 78 - 78.5 Gy < 78 Gy

Dmax - Dmax < 83.5 Gy (107%) 83.5 - 85.8Gy (107-110%) > 85.8 Gy (110%)

Dmax - Dmax location In CTV In PTV-CTV Outside PTV

Rectum RECTUM V40 < 50% 50 - 60% > 60%

Rectum RECTUM V50 < 40% 40 - 50% > 50%

Rectum RECTUM V60 < 30% 30 - 35% > 35%

Rectum RECTUM V65 < 20% 20 - 25% > 25%

Rectum RECTUM V70 < 10% 10 - 20% > 20%

Rectum RECTUM V75 < 5% 5 - 10% > 10%

Rectum PRV RECT_78_03_NS V80 < 1 cc must be less than 1cc > 1 cc

Rectum in PTV RECT_IN_PTV V79 < 1 cc 1 - 2 cc > 2 cc

Rectum outside PTV RECT-PTV V78 < 1 cc 1 - 2 cc > 2 cc

Rectum RECTUM Post Wall Max Dose < 39Gy 39 - 45Gy > 45Gy


3.1 Supplementary Material A

Rectum RECTUM Slices 50% crosses PRW 0 slices 1 - 5 slices > 5 slices

Urethra PRV URETHRA_PRV V82 < 0.1 cc 0.1 - 1cc > 1cc

Bladder BLADDER V50 < 40% 40 - 50% > 50%

Bladder BLADDER V60 < 30% 30 - 40% > 40%

Fem Heads FEMUR_L / R V40 < 10% 10 - 20% > 20%

Periph Tissue P_TISSUE V50 beyond PTV + 3.5cm < 1cc 1 - 2 cc > 2cc



A Mean 0.96 0.97 0.98 0.91

SD 0.03 0.02 0.01 0.04

COV 0.03 0.02 0.01 0.05

B Mean 0.87 0.90 0.97 0.87

SD 0.08 0.06 0.01 0.07

COV 0.09 0.06 0.01 0.08

C Mean 0.96 0.97 0.98 0.94

SD 0.04 0.02 0.01 0.03

COV 0.04 0.02 0.01 0.03

Table 2: Trial 1 DSC statistics


A Mean (voxels) 2.64 2.87 2.03 8.99

SD (voxels) 1.42 1.67 1.21 5.64

COV 0.54 0.58 0.59 0.63

B Mean (voxels) 5.24 5.35 2.31 7.78

SD (voxels) 2.29 2.27 0.74 4.15

COV 0.44 0.43 0.32 0.53

C Mean (voxels) 2.32 2.31 2.15 6.57

SD (voxels) 1.14 1.16 1.03 6.66

COV 0.49 0.50 0.48 1.01

Table 3: Trial 1 95% Hausdorff Statistics


Table 4: Bladder contouring similarity and radiobiological metric correlations, Trial 1: STAPLE gold standard

Table 5: Rectum contouring similarity and radiobiological metric correlations, Trial 1: STAPLE gold standard

NTCPlkb EUD mean max iso40 iso50 iso55 iso60 iso65 iso70 iso80 iso85 iso90 DX10 DX25 DX50

DSC -0.1874 -0.2629 -0.2766 0.0382 -0.2516 -0.2541 -0.2747 -0.2414 -0.2528 -0.2464 -0.2656 -0.2575 -0.2276 -0.3123 -0.3135 -0.2401

VOLSIM -0.1007 -0.1528 -0.1381 0.0311 -0.1492 -0.1833 -0.1761 -0.1404 -0.1379 -0.1664 -0.1535 -0.1355 -0.1316 -0.2142 -0.1115 -0.0740

SENS 0.0476 0.0655 0.0454 0.0232 0.0629 0.0797 0.0646 0.0545 0.0536 0.0870 0.0521 0.0425 0.0518 0.0799 -0.0055 -0.0001

SPEC -0.0883 -0.2139 -0.1906 0.1474 -0.1750 -0.2171 -0.2156 -0.2057 -0.2174 -0.2423 -0.2319 -0.2164 -0.2120 -0.3106 -0.1749 -0.1043

MASD 0.1236 0.2156 0.2085 -0.1226 0.1804 0.1903 0.2081 0.2029 0.2248 0.2263 0.2428 0.2376 0.2183 0.2889 0.2508 0.1421

HAUS95 0.1511 0.1333 0.1409 -0.1972 0.1174 0.1172 0.1315 0.1214 0.1329 0.1331 0.1415 0.1422 0.1320 0.1870 0.1544 0.0981

CENTROID_DIST_EUC 0.0911 0.0540 0.0412 -0.2191 0.0056 0.0394 0.0433 0.0200 0.0342 0.0420 0.0676 0.0640 0.0396 0.1109 0.0718 0.0393

CENTROID_DIST_X -0.0107 -0.1009 -0.0658 0.0098 -0.0474 -0.0751 -0.0623 -0.0659 -0.0672 -0.0534 -0.0590 -0.0547 -0.0478 -0.0903 -0.1012 0.0091

CENTROID_DIST_Y 0.0813 -0.0802 -0.0299 0.0178 -0.0028 -0.0258 -0.0350 -0.0652 -0.0605 -0.0524 -0.0894 -0.0785 -0.0667 -0.0362 -0.0422 -0.0006

CENTROID_DIST_Z 0.1466 0.3310 0.2905 -0.1059 0.2589 0.2694 0.2804 0.2812 0.2688 0.2665 0.3017 0.2903 0.2591 0.2951 0.3369 0.2686

aRVD 0.0765 0.0686 0.0987 0.0014 0.0985 0.0651 0.0626 0.0612 0.0764 0.0690 0.0633 0.0564 0.0474 0.0767 0.1761 0.1139

C 0.0483 0.0665 0.0462 0.0239 0.0641 0.0805 0.0655 0.0553 0.0544 0.0877 0.0529 0.0436 0.0527 0.0806 -0.0054 0.0005

NTCPlkb EUD mean max iso35 iso40 iso50 iso55 iso60 iso65 iso70 iso80 iso85 iso90 DX25 DX50

DSC -0.1733 -0.2920 -0.3080 0.0138 -0.3259 -0.3606 -0.3755 -0.3552 -0.3714 -0.3558 -0.3504 -0.3327 -0.2943 -0.2439 -0.3065 -0.2588

VOLSIM 0.3071 0.4378 0.4530 -0.1911 0.4689 0.5123 0.5168 0.5233 0.4997 0.4844 0.4835 0.4632 0.4538 0.4093 0.5008 0.3711

SENS -0.2737 -0.4192 -0.4220 0.1282 -0.4438 -0.4909 -0.5036 -0.5015 -0.4929 -0.4735 -0.4780 -0.4546 -0.4354 -0.3839 -0.4473 -0.3299

SPEC 0.3343 0.4073 0.4420 -0.1663 0.3997 0.4298 0.4350 0.4294 0.4219 0.3941 0.4116 0.3844 0.4016 0.3881 0.4193 0.3495

MASD 0.0669 0.1754 0.2296 0.0418 0.2953 0.3143 0.3094 0.2721 0.2848 0.2614 0.2502 0.2224 0.1777 0.1291 0.2123 0.2359

HAUS95 -0.0674 0.0003 0.0566 0.0558 0.1620 0.1652 0.1356 0.0928 0.1005 0.0739 0.0637 0.0328 -0.0123 -0.0580 0.0357 0.0972

CENTROID_DIST_EUC 0.0319 0.1016 0.1250 0.0335 0.1983 0.2095 0.2032 0.1816 0.1896 0.1704 0.1619 0.1451 0.0984 0.0531 0.1306 0.1141

CENTROID_DIST_X 0.0434 0.0945 0.0523 0.0782 0.1432 0.1739 0.1553 0.1534 0.1308 0.1350 0.1128 0.0920 0.0722 0.0846 0.1067 0.0333

CENTROID_DIST_Y -0.0144 -0.1660 -0.3253 0.1793 -0.2284 -0.2414 -0.2816 -0.2903 -0.2671 -0.2658 -0.2662 -0.2361 -0.2011 -0.1157 -0.1978 -0.3115

CENTROID_DIST_Z 0.1473 0.0205 -0.2361 0.1559 -0.1233 -0.1167 -0.1473 -0.1390 -0.1392 -0.1381 -0.1426 -0.1196 -0.0869 0.0224 -0.1102 -0.2577

aRVD 0.2584 0.3835 0.4626 -0.0233 0.4092 0.4392 0.4658 0.4459 0.4610 0.4478 0.4366 0.4223 0.3932 0.3360 0.3886 0.4356

C -0.2730 -0.4185 -0.4221 0.1282 -0.4442 -0.4913 -0.5036 -0.5014 -0.4926 -0.4731 -0.4775 -0.4539 -0.4347 -0.3830 -0.4469 -0.3301

p > 0.00026

p < 0.00026

Chapter

3.Prostate

contouringmetric

correlations87

TCPpossionLQ TCPlogit EUD min mean max iso90 iso95 iso100 DX90 DX95 dosehomo

DSC 0.0565 0.0734 0.0705 0.0597 0.0677 0.0271 0.0500 -0.0138 0.0788 0.0821 0.0771 -0.1094

VOLSIM 0.0438 0.0400 0.0453 0.0378 0.0397 0.0463 0.0611 0.0723 -0.0116 0.0511 0.0643 0.0656

SENS -0.0335 -0.0229 -0.0287 -0.0442 -0.0249 -0.0565 0.0241 -0.0475 0.0278 -0.0312 -0.0389 -0.0603

SPEC 0.0013 -0.0104 -0.0054 0.0386 -0.0116 -0.0292 0.0266 0.1267 0.0032 0.0217 0.0690 0.0023

MASD -0.0312 -0.0393 -0.0372 -0.0906 -0.0324 0.0143 -0.0305 -0.0109 -0.0699 -0.0656 -0.0693 0.1661

HAUS95 -0.0420 -0.0403 -0.0394 -0.0983 -0.0314 0.0771 -0.0350 -0.0139 -0.0816 -0.0819 -0.0664 0.2104

CENTROID_DIST_EUC -0.0381 -0.0187 -0.0188 -0.0346 -0.0121 0.0151 -0.0979 -0.0236 -0.0723 -0.0714 -0.0839 0.1696

CENTROID_DIST_X -0.0072 -0.0227 -0.0116 0.0248 -0.0247 -0.0329 0.0944 0.1276 -0.0121 -0.0382 0.0152 0.0640

CENTROID_DIST_Y 0.2182 0.2055 0.2080 -0.0303 0.2013 0.1003 0.0791 -0.0127 0.1922 0.1473 0.1523 -0.0665

CENTROID_DIST_Z 0.1702 0.2025 0.2061 -0.0455 0.2053 0.0289 0.1777 0.0169 -0.0878 0.1196 0.0452 0.1753

aRVD -0.0534 -0.0661 -0.0642 -0.0400 -0.0602 -0.0556 -0.0128 0.0644 -0.0641 -0.0806 -0.0721 0.1333

C -0.0320 -0.0217 -0.0274 -0.0429 -0.0237 -0.0564 0.0241 -0.0460 0.0292 -0.0297 -0.0375 -0.0611

p > 0.00035

p < 0.00035

Table 6: CTV contouring similarity and radiobiological metric correlations, Trial 1: STAPLE gold standard


DSC -0.0419 0.1080 0.1060 -0.1994 0.1185 0.0545 0.0222 0.0380 0.0518 0.0581 0.1239 -0.0903

VOLSIM 0.5741 0.3876 0.4329 0.6722 0.3107 0.0010 0.4750 0.5162 0.5449 0.5086 0.5211 -0.5202

SENS -0.4930 -0.2978 -0.3347 -0.6115 -0.2353 -0.0063 -0.4097 -0.4179 -0.4483 -0.4106 -0.4230 0.4301

SPEC 0.5395 0.3227 0.3657 0.6404 0.2485 -0.0578 0.4125 0.4697 0.4956 0.4500 0.5115 -0.5266

MASD 0.0594 -0.0867 -0.0894 0.2137 -0.0962 -0.0441 -0.0007 -0.0239 -0.0249 -0.0316 -0.1042 0.0756

HAUS95 -0.0202 -0.1194 -0.1304 0.1191 -0.1178 0.0044 -0.0990 -0.1108 -0.0749 -0.0513 -0.1720 0.1600

CENTROID_DIST_EUC -0.0045 -0.0586 -0.0687 0.1047 -0.0553 0.0253 -0.0205 -0.0748 -0.0290 -0.0321 -0.1172 0.1144

CENTROID_DIST_X 0.3293 0.1664 0.2117 0.4527 0.1132 0.0048 0.3616 0.3670 0.2405 0.2132 0.2529 -0.2182

CENTROID_DIST_Y 0.0777 0.0828 0.1112 0.0551 0.0701 0.0895 0.1171 0.0908 -0.0081 0.0224 0.0479 0.0005

CENTROID_DIST_Z 0.0503 0.1591 0.1392 0.1248 0.1658 0.0052 0.0392 0.0128 0.0922 0.1366 0.0500 0.0066

aRVD 0.0411 -0.1051 -0.0989 0.2103 -0.1186 -0.1060 -0.0455 -0.0437 -0.0671 -0.0735 -0.1429 0.1054

C -0.4920 -0.2967 -0.3339 -0.6105 -0.2342 -0.0053 -0.4093 -0.4174 -0.4479 -0.4099 -0.4223 0.4303

Table 7: PTV contouring similarity and radiobiological metric correlations, Trial 1: STAPLE gold standard

Chapter

3.Prostate

contouringmetric

correlations88

To see whether the choice of STAPLE as a gold standard reference volume impacted the results of the study, additional reference volumes were

investigated. Each set of observer contours were iteratively designated as the gold standard volume, and contouring similarity and radiobiological metrics

for the remaining observer contours were derived in an analogous manner to those in the paper. Significant correlations are shown in Tables 8 - 11, where

once again Volume Similarity, Sensitivity, and Specificity displayed the strongest correlations with multiple radiobiological metrics. This suggests that the

choice of gold standard did not affect the observed correlations within this study.



DSC -0.0166 0.0029 -0.0002 0.0212 -0.0007 -0.0002 -0.0014 -0.0007 -0.0002 0.0010 0.0010 0.0013 0.0006 0.0019 -0.0022 -0.0008

VOLSIM -0.1103 -0.2538 -0.2297 -0.0210 -0.2316 -0.2454 -0.2501 -0.2446 -0.2410 -0.2465 -0.2445 -0.2411 -0.2390 -0.2734 -0.2206 -0.1359

SENS 0.1022 0.2274 0.2067 0.0114 0.2065 0.2170 0.2212 0.2173 0.2154 0.2216 0.2194 0.2167 0.2131 0.2492 0.1927 0.1229

SPEC -0.1174 -0.2091 -0.1839 -0.0167 -0.1820 -0.1915 -0.1958 -0.1945 -0.1926 -0.1971 -0.1973 -0.1959 -0.1935 -0.2368 -0.1815 -0.1115

MASD 0.0295 -0.0020 -0.0010 -0.0238 -0.0018 -0.0025 -0.0018 -0.0013 -0.0022 -0.0026 -0.0013 -0.0004 -0.0008 0.0015 0.0007 0.0015

HAUS95 0.0551 0.0039 -0.0007 -0.0069 -0.0026 -0.0020 -0.0031 -0.0024 -0.0036 -0.0036 -0.0005 -0.0002 -0.0014 0.0051 -0.0048 0.0011

CENTROID_DIST_EUC 0.0439 -0.0010 0.0014 0.0097 0.0001 -0.0010 -0.0003 0.0005 -0.0001 -0.0001 0.0006 0.0022 0.0016 0.0085 0.0032 0.0026

CENTROID_DIST_X -0.1436 0.0072 -0.0091 0.1412 -0.0019 0.0103 0.0120 0.0095 0.0019 0.0024 0.0172 0.0129 -0.0014 -0.0006 -0.0055 0.0212

CENTROID_DIST_Y 0.0518 -0.0001 -0.0434 0.0924 -0.0436 -0.0226 -0.0227 -0.0278 -0.0263 -0.0194 -0.0049 0.0020 -0.0095 0.0001 -0.0372 -0.0560

CENTROID_DIST_Z 0.1720 0.1979 0.1932 -0.0849 0.1820 0.1672 0.1731 0.1746 0.1645 0.1637 0.1629 0.1617 0.1612 0.1988 0.2284 0.2299

C 0.1023 0.2274 0.2065 0.0118 0.2064 0.2169 0.2211 0.2173 0.2153 0.2216 0.2194 0.2167 0.2131 0.2491 0.1928 0.1226

p > 0.00026

p < 0.00026

Table 8: Bladder contouring similarity and radiobiological metric correlations, Trial 1: observer gold standard

Table 9: Rectum contouring similarity and radiobiological metric correlations, Trial 1: observer gold standard


DSC -0.0212 -0.0200 -0.0208 -0.0044 -0.0483 -0.0483 -0.0363 -0.0309 -0.0288 -0.0234 -0.0248 -0.0131 -0.0158 -0.0082 -0.0342 -0.0423

VOLSIM 0.3471 0.5045 0.5679 -0.2704 0.5923 0.5937 0.5831 0.5786 0.5699 0.5645 0.5640 0.5453 0.5423 0.4702 0.5757 0.5029

SENS -0.3277 -0.4643 -0.5211 0.2056 -0.5543 -0.5572 -0.5439 -0.5369 -0.5299 -0.5227 -0.5226 -0.5009 -0.4998 -0.4297 -0.5385 -0.4646

SPEC 0.3139 0.4315 0.4712 -0.2539 0.4870 0.4884 0.4903 0.4862 0.4814 0.4789 0.4788 0.4649 0.4617 0.4124 0.4801 0.4191

MASD 0.0182 0.0250 0.0157 0.0130 0.0439 0.0462 0.0347 0.0308 0.0288 0.0243 0.0252 0.0157 0.0186 0.0121 0.0354 0.0480

HAUS95 0.0155 0.0302 0.0195 0.0052 0.0444 0.0494 0.0419 0.0375 0.0364 0.0334 0.0330 0.0254 0.0259 0.0216 0.0456 0.0493

CENTROID_DIST_EUC 0.0086 0.0177 0.0148 0.0088 0.0401 0.0428 0.0339 0.0286 0.0256 0.0222 0.0220 0.0128 0.0159 0.0046 0.0323 0.0323

CENTROID_DIST_X -0.0321 -0.0079 -0.0122 -0.0219 0.0639 0.0694 0.0511 0.0414 0.0158 0.0170 0.0086 -0.0094 -0.0158 0.0306 0.0102 -0.0195

CENTROID_DIST_Y -0.0336 -0.1954 -0.3443 0.0886 -0.3148 -0.3049 -0.3049 -0.3002 -0.2888 -0.2919 -0.2884 -0.2666 -0.2361 -0.1323 -0.2770 -0.3649

CENTROID_DIST_Z 0.0750 -0.0713 -0.2705 0.1075 -0.1937 -0.1837 -0.1941 -0.1875 -0.1964 -0.1951 -0.2024 -0.1887 -0.1620 -0.0539 -0.1893 -0.3273

C -0.3277 -0.4644 -0.5213 0.2054 -0.5545 -0.5574 -0.5441 -0.5371 -0.5300 -0.5228 -0.5228 -0.5010 -0.4999 -0.4297 -0.5387 -0.4648

Chapter

3.Prostate

contouringmetric

correlations90


DSC 0.1715 0.0454 0.0693 0.2690 0.0251 -0.0562 0.3807 0.3310 0.1090 0.0174 0.0670 -0.1263

VOLSIM 0.2097 0.1554 0.1792 0.2966 0.1328 -0.0668 0.5039 0.2998 0.1827 0.1436 0.1892 -0.1134

SENS -0.1870 -0.1371 -0.1589 -0.2557 -0.1160 0.0730 -0.4569 -0.2625 -0.1958 -0.1379 -0.1928 0.1220

SPEC 0.2011 0.1400 0.1615 0.3082 0.1193 -0.0716 0.4927 0.3179 0.1994 0.1432 0.1950 -0.1388

MASD -0.1795 -0.0568 -0.0798 -0.2797 -0.0369 0.0539 -0.4016 -0.3375 -0.1237 -0.0311 -0.0841 0.1389

HAUS95 -0.1865 -0.0639 -0.0865 -0.2920 -0.0431 0.0351 -0.4242 -0.3442 -0.1222 -0.0320 -0.0827 0.1382


CENTROID_DIST_X 0.0810 0.0385 0.0517 0.1441 0.0275 -0.1074 0.2647 0.1192 0.1663 0.0709 0.1582 -0.1571

CENTROID_DIST_Y 0.1626 0.1634 0.1669 0.0009 0.1545 0.0242 0.0630 0.0142 0.1662 0.1125 0.1193 -0.0404

CENTROID_DIST_Z 0.2444 0.2009 0.2199 0.0555 0.1848 -0.0194 0.3224 0.1187 0.1379 0.2293 0.2072 0.0514

C -0.1863 -0.1367 -0.1585 -0.2552 -0.1157 0.0733 -0.4558 -0.2618 -0.1956 -0.1377 -0.1928 0.1222

p > 0.00035

p < 0.00035

Table 10: CTV contouring similarity and radiobiological metric correlations, Trial 1: observer gold standard


DSC 0.2254 0.3338 0.3063 0.1789 0.3525 0.0264 0.3231 0.3933 0.3015 0.3156 0.3707 -0.3640

VOLSIM 0.7649 0.7475 0.7621 0.8065 0.6840 0.0094 0.7822 0.7509 0.7716 0.7701 0.7621 -0.7420

SENS -0.6865 -0.6594 -0.6738 -0.7323 -0.5992 0.0342 -0.6969 -0.6595 -0.7008 -0.6915 -0.6727 0.6596

SPEC 0.8216 0.7644 0.7906 0.8380 0.6841 -0.0248 0.8340 0.7990 0.7818 0.7772 0.8019 -0.7834

MASD -0.2335 -0.3241 -0.3014 -0.1895 -0.3397 -0.0293 -0.3231 -0.3919 -0.2884 -0.2989 -0.3614 0.3575

HAUS95 -0.2470 -0.3051 -0.2937 -0.2062 -0.3144 -0.0298 -0.3264 -0.3812 -0.2820 -0.2948 -0.3636 0.3625

CENTROID_DIST_EUC -0.2736 -0.2776 -0.2740 -0.2161 -0.2710 -0.0336 -0.3387 -0.3609 -0.2607 -0.2669 -0.3524 0.3458

CENTROID_DIST_X 0.4876 0.4089 0.4382 0.4866 0.3412 -0.0339 0.5020 0.4590 0.4081 0.4134 0.4533 -0.4376

CENTROID_DIST_Y 0.0029 -0.0445 -0.0275 0.0316 -0.0577 0.0185 0.0085 -0.0323 -0.0845 -0.0769 -0.0225 0.0249

CENTROID_DIST_Z 0.2478 0.3095 0.3055 0.3175 0.3010 -0.0030 0.2888 0.2613 0.3154 0.3335 0.2899 -0.3003

C -0.6857 -0.6581 -0.6725 -0.7312 -0.5977 0.0336 -0.6958 -0.6583 -0.6995 -0.6902 -0.6715 0.6585Table 11: PTV contouring similarity and radiobiological metric correlations, Trial 1: observer gold standard

Chapter

3.Prostate

contouringmetric

correlations91


1 Mean 0.88 0.91 0.94 0.83

SD 0.03 0.02 0.01 0.04

COV 0.03 0.02 0.01 0.05

2 Mean 0.91 0.93 0.95 0.87

SD 0.03 0.02 0.01 0.06

COV 0.04 0.03 0.01 0.07

3 Mean 0.89 0.92 0.98 0.89

SD 0.02 0.01 0.01 0.04

COV 0.02 0.02 0.01 0.04

4 Mean 0.86 0.90 0.95 0.88

SD 0.06 0.04 0.03 0.07

COV 0.07 0.05 0.03 0.08

5 Mean 0.89 0.91 0.95 0.90

SD 0.02 0.02 0.01 0.03

COV 0.03 0.02 0.01 0.03

Table 12: Trial 2 DSC statistics



1 Mean (voxels) 4.65 4.98 2.88 8.65

SD (voxels) 1.07 0.91 0.72 6.59

COV 0.23 0.18 0.25 0.76

2 Mean (voxels) 3.49 3.51 2.45 11.29

SD (voxels) 0.94 0.81 0.11 13.58

COV 0.27 0.23 0.05 1.20

3 Mean (voxels) 4.02 4.46 2.08 7.60

SD (voxels) 0.89 0.88 0.43 6.34

COV 0.22 0.20 0.21 0.83

4 Mean (voxels) 4.60 4.81 2.89 9.01

SD (voxels) 1.87 1.75 0.93 8.66

COV 0.41 0.36 0.32 0.96

5 Mean (voxels) 4.86 5.02 2.57 6.06

SD (voxels) 1.44 1.45 0.52 3.32

COV 0.30 0.29 0.20 0.55

Table 13: Trial 2 95% Hausdorff statistics



DSC 0.0099 0.3073 0.0980 0.6243 0.0437 0.0770 0.0954 0.1139 0.1233 0.1256 0.1122 0.0913 0.0703 0.3787 0.1637 0.0546

VOLSIM -0.1841 -0.1284 -0.0691 -0.1506 -0.0279 -0.0971 -0.1162 -0.1168 -0.1211 -0.1557 -0.1572 -0.1530 -0.1554 -0.0780 -0.0764 -0.0780

SENS 0.0635 0.1809 0.0018 0.3827 -0.0651 0.0031 0.0270 0.0329 0.0427 0.0801 0.0854 0.0732 0.0674 0.1952 0.0474 -0.0208

SPEC -0.0637 -0.0473 0.0203 0.0428 0.0350 -0.0167 -0.0187 -0.0063 -0.0055 -0.0318 -0.0433 -0.0608 -0.0660 -0.0428 0.0285 0.0119

MASD -0.0962 -0.2350 -0.1798 -0.2896 -0.1745 -0.1843 -0.1906 -0.1944 -0.1954 -0.1785 -0.1537 -0.1243 -0.1219 -0.2919 -0.2129 -0.1568

HAUS95 0.0926 -0.1938 0.0029 -0.4696 0.0169 0.0119 0.0103 0.0033 0.0060 0.0043 0.0185 0.0379 0.0512 -0.3298 -0.0353 0.0448

CENTROID_DIST_EUC -0.0038 -0.2315 -0.1130 -0.4925 -0.0704 -0.0653 -0.0726 -0.0965 -0.1081 -0.1000 -0.1049 -0.0793 -0.0502 -0.2500 -0.1516 -0.0754

CENTROID_DIST_X -0.0253 -0.0876 -0.0491 -0.0966 -0.0667 -0.0778 -0.0847 -0.0791 -0.0795 -0.0549 -0.0445 -0.0314 -0.0201 -0.1328 -0.0933 -0.0427

CENTROID_DIST_Y -0.5113 -0.4268 -0.4982 -0.1540 -0.4321 -0.4803 -0.4754 -0.4936 -0.4995 -0.5204 -0.5410 -0.5487 -0.5378 -0.2702 -0.4581 -0.4791

CENTROID_DIST_Z 0.2949 0.2826 0.3122 0.0918 0.3011 0.3166 0.3226 0.3199 0.3211 0.3288 0.3296 0.3149 0.3181 0.2041 0.3148 0.3131

aRVD -0.1242 -0.4148 -0.2103 -0.5437 -0.1788 -0.2236 -0.2424 -0.2570 -0.2538 -0.2523 -0.2323 -0.2211 -0.2049 -0.4856 -0.2650 -0.1635

C 0.0635 0.1809 0.0018 0.3827 -0.0651 0.0031 0.0270 0.0329 0.0427 0.0801 0.0854 0.0732 0.0674 0.1952 0.0474 -0.0208

Table 14: Bladder contouring similarity and radiobiological metric correlations, Trial 2: majority vote gold standard

p > 0.00026

p < 0.00026


DSC -0.2452 -0.12449 0.12541 -0.03286 0.2048 0.18571 0.1351 0.10235 0.05959 0.04857 -0.00622 -0.05786 -0.10163 -0.15306 0.09724 0.12071

VOLSIM 0.13265 0.11878 0.19918 0.23092 0.22031 0.17612 0.15418 0.16245 0.16286 0.1599 0.17286 0.13286 0.13367 0.11051 0.15714 0.2102

SENS -0.29041 -0.20918 -0.11449 -0.19893 -0.05204 -0.03337 -0.06847 -0.09837 -0.12633 -0.13633 -0.1849 -0.18857 -0.21316 -0.22449 -0.1051 -0.06071

SPEC -0.16848 -0.05657 0.34274 0.18089 0.43275 0.34851 0.25615 0.2201 0.16741 0.14275 0.09741 0.00965 -0.02042 -0.09828 0.18278 0.43198

MASD 0.25194 0.14939 -0.07765 0.0325 -0.12398 -0.09959 -0.05969 -0.03745 -0.00418 0.00643 0.04276 0.08796 0.11184 0.15163 -0.04163 -0.05847

HAUS95 0.23502 0.15344 -0.01633 0.02423 -0.0323 -0.00696 0.0088 0.01853 0.0392 0.0499 0.06418 0.09842 0.10856 0.14924 0.00798 0.02759

CENTROID_DIST_EUC 0.24673 0.15296 -0.05776 -0.03133 -0.10092 -0.07459 -0.0501 -0.03918 -0.01408 -0.00194 0.03388 0.08276 0.10776 0.14459 -0.03041 -0.0252

CENTROID_DIST_X 0.30173 0.25092 -0.03469 0.15322 -0.0898 -0.04439 0.04765 0.11163 0.14418 0.16061 0.21316 0.23531 0.2598 0.28265 0.10847 -0.17827

CENTROID_DIST_Y 0.21408 0.14735 0.34235 0.135 0.39 0.32847 0.24602 0.20122 0.16367 0.1499 0.11888 0.12245 0.10735 0.10918 0.18316 0.43153

CENTROID_DIST_Z 0.17643 0.09367 0.26888 0.24149 0.30286 0.24265 0.16653 0.1249 0.10286 0.09592 0.06949 0.07673 0.0598 0.0651 0.12235 0.34602

aRVD -0.02653 -0.05745 -0.11449 0.14812 -0.07724 -0.09949 -0.11796 -0.09735 -0.08724 -0.09316 -0.06551 -0.07418 -0.04245 -0.05571 -0.12051 -0.02122

C -0.29041 -0.20918 -0.11449 -0.19893 -0.05204 -0.03337 -0.06847 -0.09837 -0.12633 -0.13633 -0.1849 -0.18857 -0.21316 -0.22449 -0.1051 -0.06071

Table 15: Rectum contouring similarity and radiobiological metric correlations, Trial 2: majority vote gold standard

Chapter

3.Prostate

contouringmetric

correlations94

p > 0.00035

p < 0.00035


DSC -0.0502 -0.0428 -0.0463 0.2218 -0.0376 -0.0124 NaN -0.0082 -0.0778 -0.1219 -0.1633 0.1011

VOLSIM 0.1468 0.1385 0.1272 -0.0857 0.1408 0.1479 NaN -0.0220 0.0833 0.1630 0.1369 0.0787

SENS -0.1609 -0.1774 -0.1677 0.2126 -0.1788 -0.1674 NaN 0.0367 -0.0515 -0.1983 -0.1722 -0.0974

SPEC 0.3614 0.3670 0.3539 0.1226 0.3712 0.2700 NaN 0.1321 0.0701 0.2570 0.1798 0.2036

MASD 0.0090 0.0262 0.0286 -0.2786 0.0263 0.1157 NaN 0.0140 0.0455 0.0827 0.1396 0.0024

HAUS95 0.0265 0.0181 0.0211 -0.2057 0.0184 0.0731 NaN 0.0078 0.0777 0.1130 0.1561 -0.0665

CENTROID_DIST_EUC -0.0521 -0.0162 -0.0184 -0.2000 -0.0197 0.0720 NaN 0.1690 0.1319 0.0456 0.1114 -0.1806

CENTROID_DIST_X -0.0188 -0.0839 -0.0752 0.2892 -0.0945 -0.2533 NaN 0.1708 0.3427 0.0833 0.1550 -0.3732

CENTROID_DIST_Y -0.1528 -0.1470 -0.1467 0.0270 -0.1496 -0.0301 NaN 0.0039 0.0281 -0.0700 -0.0135 -0.0811

CENTROID_DIST_Z -0.0133 0.0058 -0.0063 0.0285 0.0011 0.0838 NaN -0.0480 -0.1358 -0.1633 -0.1761 0.2082

aRVD 0.0512 -0.0196 -0.0219 0.1389 -0.0298 0.0022 NaN 0.2068 0.2290 0.0561 0.1416 -0.1414

C -0.1588 -0.1741 -0.1643 0.2096 -0.1753 -0.1638 NaN 0.0328 -0.0537 -0.1965 -0.1728 -0.0932

Table 16: CTV contouring similarity and radiobiological metric correlations, Trial 2: majority vote gold standard. Note: correlations for iso90 weren’t possible due to every observer patient having 100% coverage of the CTV at the 90% isodose level, exactly matching the majority vote plans for each patient.


DSC 0.1941 0.1647 0.1718 0.1957 0.1471 -0.0084 0.2564 0.2055 0.0107 0.0706 0.1636 -0.2082

VOLSIM 0.6490 0.7007 0.6685 0.6117 0.6869 0.1683 0.6840 0.7585 0.7705 0.7966 0.7895 -0.7605

SENS -0.4837 -0.5657 -0.5235 -0.4486 -0.5703 -0.1943 -0.5193 -0.6014 -0.6760 -0.6731 -0.6372 0.5977

SPEC 0.8078 0.8699 0.8461 0.7709 0.8422 0.3631 0.8653 0.8719 0.7105 0.7823 0.8774 -0.8175

MASD -0.1729 -0.1419 -0.1408 -0.1572 -0.1221 0.0232 -0.2305 -0.1919 -0.0207 -0.0768 -0.1648 0.2218

HAUS95 -0.0807 -0.0457 -0.0599 -0.0453 -0.0228 -0.0225 -0.1396 -0.0389 0.1355 0.0751 -0.0100 0.0430


CENTROID_DIST_X -0.0782 -0.2126 -0.1938 -0.0830 -0.2643 -0.4759 -0.1186 -0.0663 -0.0628 -0.0629 -0.0501 -0.0743

CENTROID_DIST_Y 0.2333 0.1234 0.1672 0.2898 0.0646 0.1520 0.1995 0.2045 0.0042 0.0168 0.1211 -0.1463

CENTROID_DIST_Z 0.0769 0.1498 0.0927 -0.0173 0.1649 0.0851 0.0947 0.0866 0.1042 0.1424 0.0991 -0.0736

aRVD 0.2798 0.2882 0.2950 0.2642 0.2779 0.0089 0.2339 0.3132 0.3911 0.4017 0.3715 -0.3474

C -0.4814 -0.5642 -0.5215 -0.4464 -0.5695 -0.1922 -0.5170 -0.5999 -0.6773 -0.6735 -0.6358 0.5960

Table 17: PTV contouring similarity and radiobiological metric correlations, Trial 2: majority vote gold standard

Chapter

3.Prostate

contouringmetric

correlations95


DSC -0.0754 0.3174 0.0445 0.6160 0.0006 0.0118 0.0317 0.0445 0.0588 0.0611 0.0336 -0.0031 -0.0160 0.4328 0.1129 -0.0011

VOLSIM -0.3403 -0.4042 -0.3084 -0.3204 -0.2599 -0.3168 -0.3258 -0.3294 -0.3485 -0.3877 -0.3969 -0.3857 -0.3936 -0.3373 -0.3619 -0.3258

SENS 0.1787 0.4448 0.1927 0.5283 0.1238 0.1787 0.1944 0.1961 0.2171 0.2588 0.2725 0.2504 0.2532 0.4908 0.2563 0.1742

SPEC -0.1126 -0.2464 -0.1233 0.0216 -0.1093 -0.1495 -0.1239 -0.1142 -0.1194 -0.1481 -0.1778 -0.2074 -0.2047 -0.2376 -0.1454 -0.1216

MASD -0.0616 -0.1364 -0.1426 -0.2510 -0.1608 -0.1426 -0.1535 -0.1479 -0.1490 -0.1370 -0.0941 -0.0398 -0.0476 -0.2275 -0.1521 -0.1459

HAUS95 0.2344 -0.1417 0.1175 -0.5275 0.1105 0.1358 0.1314 0.1298 0.1359 0.1331 0.1577 0.1866 0.1961 -0.2856 0.0627 0.1593

CENTROID_DIST_EUC 0.1090 -0.1062 0.0664 -0.4331 0.0723 0.0913 0.0913 0.0835 0.0622 0.0762 0.0860 0.1216 0.1375 -0.2193 0.0300 0.0700

CENTROID_DIST_X -0.0359 -0.0978 -0.0664 -0.0325 -0.1048 -0.0964 -0.1056 -0.0978 -0.1011 -0.0667 -0.0569 -0.0347 -0.0221 -0.2087 -0.0896 -0.0639

CENTROID_DIST_Y -0.5546 -0.4381 -0.5157 -0.1339 -0.4336 -0.4947 -0.4849 -0.5095 -0.5176 -0.5384 -0.5650 -0.5793 -0.5695 -0.2160 -0.4896 -0.4838

CENTROID_DIST_Z 0.4821 0.3395 0.4838 0.0616 0.4762 0.4894 0.4978 0.4933 0.4891 0.5014 0.5025 0.4843 0.4885 0.1661 0.4280 0.4997

aRVD 0.0490 -0.3854 -0.0913 -0.5569 -0.0759 -0.0994 -0.1160 -0.1252 -0.1300 -0.1331 -0.1031 -0.0840 -0.0602 -0.5067 -0.2020 -0.0445

C 0.1787 0.4448 0.1927 0.5283 0.1238 0.1787 0.1944 0.1961 0.2171 0.2588 0.2725 0.2504 0.2532 0.4908 0.2563 0.1742

Table 18: Bladder contouring similarity and radiobiological metric correlations, Trial 2: radiation oncologist and radiotherapist contours only


DSC -0.1599 0.0185 0.1443 -0.2266 0.2129 0.2162 0.2017 0.1815 0.1434 0.1401 0.1193 0.0913 0.0563 -0.0325 0.1723 0.0462

VOLSIM 0.0773 0.0417 -0.0011 0.1627 -0.0168 -0.0501 -0.0311 0.0003 0.0241 0.0193 0.0697 0.0535 0.0689 0.0863 0.0028 0.0020

SENS -0.0888 0.0179 0.1008 -0.2269 0.1403 0.1728 0.1504 0.1182 0.0863 0.0913 0.0403 0.0350 0.0112 -0.0370 0.1115 0.0770

SPEC -0.1460 0.0216 0.3461 -0.0177 0.3863 0.3107 0.2585 0.2410 0.1971 0.1711 0.1773 0.1136 0.1008 0.0216 0.1917 0.3865

MASD 0.2504 0.0798 -0.0630 0.2252 -0.1095 -0.1014 -0.0902 -0.0860 -0.0521 -0.0415 -0.0403 -0.0076 0.0062 0.0840 -0.0725 0.0345

HAUS95 0.3241 0.1842 0.0629 0.1681 0.0462 0.0543 0.0553 0.0472 0.0736 0.0850 0.0767 0.1034 0.1020 0.1746 0.0542 0.1630

CENTROID_DIST_EUC 0.2479 0.0840 -0.0115 0.1826 -0.0403 -0.0370 -0.0574 -0.0683 -0.0543 -0.0443 -0.0476 -0.0204 -0.0039 0.0608 -0.0448 0.1106

CENTROID_DIST_X 0.2202 0.1966 0.0190 0.2031 0.0249 0.0585 0.1115 0.1773 0.2042 0.2048 0.2501 0.2249 0.2154 0.2171 0.1871 -0.0896

CENTROID_DIST_Y 0.4403 0.3129 0.3499 0.0426 0.3277 0.2826 0.2445 0.2185 0.1782 0.1765 0.1709 0.2258 0.2347 0.2810 0.2036 0.3894

CENTROID_DIST_Z 0.4759 0.3165 0.3594 0.2515 0.3300 0.2849 0.2496 0.2151 0.1964 0.1916 0.1770 0.2314 0.2314 0.2846 0.2045 0.3919

aRVD -0.0104 -0.0647 -0.1616 0.2092 -0.2165 -0.2104 -0.1770 -0.1272 -0.0891 -0.0899 -0.0457 -0.0639 -0.0342 -0.0084 -0.1457 -0.0849

C -0.0888 0.0179 0.1008 -0.2269 0.1403 0.1728 0.1504 0.1182 0.0863 0.0913 0.0403 0.0350 0.0112 -0.0370 0.1115 0.0770

p > 0.00026

p < 0.00026

Table 19: Rectum contouring similarity and radiobiological metric correlations, Trial 2: radiation oncologist and radiotherapist contours only

Chapter

3.Prostate

contouringmetric

correlations96


DSC -0.1176 -0.0846 -0.0891 0.1448 -0.0801 -0.0896 NaN 0.0840 -0.1426 -0.1667 -0.1683 0.0339

VOLSIM 0.3258 0.2863 0.2798 0.0146 0.2882 0.1333 NaN -0.0060 0.1151 0.2471 0.2280 0.1392

SENS -0.3339 -0.2863 -0.2812 0.0580 -0.2868 -0.1246 NaN 0.0397 -0.1162 -0.2737 -0.2420 -0.1255

SPEC 0.5819 0.5565 0.5506 0.1774 0.5564 0.2301 NaN 0.2437 0.2211 0.4838 0.4161 0.1205

MASD 0.0796 0.0496 0.0549 -0.1899 0.0549 0.2104 NaN -0.0879 0.1333 0.1143 0.1563 0.0493

HAUS95 0.1070 0.0737 0.0780 -0.0654 0.0776 0.1589 NaN -0.0898 0.1283 0.1248 0.1322 0.0622

CENTROID_DIST_EUC -0.0375 -0.0924 -0.0908 0.0120 -0.0938 0.0199 NaN 0.1309 0.1810 -0.0266 0.0549 -0.2132

CENTROID_DIST_X -0.0235 -0.1263 -0.1249 0.3230 -0.1443 -0.2036 NaN 0.1928 0.4073 0.1056 0.2370 -0.5073

CENTROID_DIST_Y -0.1952 -0.2244 -0.2235 0.1874 -0.2311 -0.1605 NaN -0.0610 0.1468 -0.1118 -0.0714 -0.1473

CENTROID_DIST_Z 0.0333 0.0594 0.0513 -0.1101 0.0529 0.0062 NaN 0.0013 -0.1619 -0.0627 -0.0669 0.0919

aRVD 0.2143 0.1384 0.1398 0.2395 0.1193 0.0768 NaN 0.2034 0.3964 0.1958 0.3095 -0.1269

C -0.3300 -0.2798 -0.2748 0.0527 -0.2801 -0.1171 NaN 0.0333 -0.1202 -0.2700 -0.2429 -0.1168

Table 20: CTV contouring similarity and radiobiological metric correlations, Trial 2: radiation oncologist and radiotherapist contours only. Note: correlations for iso90 weren’t possible due to every observer patient having 100% coverage of the CTV at the 90% isodose level, exactly matching the majority vote plans for each patient.

p > 0.00035

p < 0.00035


DSC -0.0254 -0.1361 -0.1129 0.0129 -0.1630 0.0560 -0.0231 -0.0415 -0.2588 -0.1964 -0.0689 -0.0165

VOLSIM 0.7297 0.7359 0.7398 0.6560 0.6927 0.1605 0.7169 0.7406 0.7653 0.8115 0.7782 -0.7230

SENS -0.5646 -0.6314 -0.6190 -0.4927 -0.6171 -0.1389 -0.5656 -0.5961 -0.6994 -0.7129 -0.6420 0.5683

SPEC 0.8609 0.8848 0.8852 0.7925 0.8409 0.3616 0.8765 0.8681 0.7107 0.7731 0.8900 -0.8073

MASD -0.0144 0.1034 0.0801 -0.0176 0.1308 -0.0300 -0.0123 0.0120 0.2190 0.1627 0.0286 0.0678

HAUS95 0.0863 0.1558 0.1324 0.1257 0.1623 -0.1379 0.0471 0.0976 0.3242 0.2628 0.1313 -0.0784

CENTROID_DIST_EUC -0.4065 -0.2686 -0.2994 -0.3577 -0.2118 -0.0762 -0.3864 -0.4090 -0.2328 -0.2759 -0.4129 0.5255

CENTROID_DIST_X 0.0146 -0.1826 -0.1331 0.0078 -0.2613 -0.4479 -0.0493 -0.0036 -0.0919 -0.0560 0.0059 -0.1711

CENTROID_DIST_Y 0.3362 0.1496 0.1849 0.3980 0.0756 0.0717 0.2978 0.2543 0.0555 0.0454 0.1465 -0.2171

CENTROID_DIST_Z -0.0805 0.0406 -0.0429 -0.2255 0.0818 0.1403 -0.0714 0.0095 0.1409 0.1557 0.0401 -0.0303

aRVD 0.5028 0.5356 0.5426 0.4401 0.5140 -0.0389 0.4751 0.5120 0.6720 0.6846 0.5793 -0.5283

C -0.5646 -0.6314 -0.6190 -0.4927 -0.6171 -0.1389 -0.5656 -0.5961 -0.6994 -0.7129 -0.6420 0.5683

Table 21: PTV contouring similarity and radiobiological metric correlations, Trial 2: radiation oncologist and radiotherapist contours only

Chapter

3.Prostate

contouringmetric

correlations97

CHAPTER 4

Multi-observer contouring of male pelvic

anatomy: Highly variable agreement

across conventional and emerging

structures of interest

98

RADIATION ONCOLOGY—ORIGINAL ARTICLE

Multi-observer contouring of male pelvic anatomy: Highlyvariable agreement across conventional and emergingstructures of interestDale Roach,1,2 Lois C Holloway,1,2,3,4 Michael G Jameson,1,2,3,4 Jason A Dowling,1,3,5,6

Angel Kennedy,7 Peter B Greer,6,8 Michele Krawiec,7 Robba Rai,1,2,4 Jim Denham,9 JeremiahDe Leon,10 Karen Lim,4,11 Megan E Berry,4,11 Rohen T White,7 Sean A Bydder,7 Hendrick T Tan,7

Jeremy D Croker,12 Alycea McGrath,7 John Matthews,13 Robert J Smeenk14 and Martin A Ebert3,7,15

1 Faculty of Medicine, University of New South Wales, Sydney, New South Wales, Australia

2 Ingham Institute for Applied Medical Research, Sydney, New South Wales, Australia

3 Centre for Medical Radiation Physics, University of Wollongong, Wollongong, New South Wales, Australia

4 Department of Radiation Oncology, Liverpool and Macarthur Cancer Therapy Centres, New South Wales, Australia

5 Australian e-Health Research Centre, CSIRO, Royal Brisbane Hospital, Brisbane, Queensland, Australia

6 School of Mathematical and Physical Sciences, University of Newcastle, Newcastle, New South Wales, Australia

7 Radiation Oncology, Sir Charles Gairdner Hospital, Perth, Western Australia, Australia

8 Calvary Mater Newcastle Hospital, Newcastle, New South Wales, Australia

9 School of Medicine and Population Health, University of Newcastle, Newcastle, New South Wales, Australia

10 Illawarra Cancer Care Centre, Wollongong, New South Wales, Australia

11 South Western Sydney Clinical School, University of New South Wales, Sydney, New South Wales, Australia

12 Fiona Stanley Hospital, Perth, Western Australia, Australia

13 Radiation Oncology, Auckland City Hospital, Auckland, New Zealand

14 Department of Radiation Oncology, Radboud University Medical Centre, Nijmegen, The Netherlands

15 School of Physics and Astrophysics, Faculty of Science, University of Western Australia, Perth, Western Australia, Australia

D Roach MSc; LC Holloway PhD; MG

Jameson PhD; JA Dowling PhD; A Kennedy

BSc; PB Greer PhD; M Krawiec BSc; R Rai

MHlthSc; J Denham MD, FRANZCR; J De Leon

MBBS, FRANZCR; K Lim MBBS, FRANZCR; ME

Berry MBBS, FRANZCR; RT White MBBS,

FRANZCR; SA Bydder MBA, MPH, FRANZCR;

HT Tan MBBS; JD Croker MBBS, FRANZCR;

A McGrath MBBS; J Matthews MBChB,

FRANZCR; RJ Smeenk MD, PhD; MA Ebert

PhD.

Correspondence

Mr. Dale Roach, Ingham Institute, Liverpool

Hospital, Locked Bag 7103, Liverpool, NSW

1871, Australia.

Email: [email protected]

Conflict of interest: None.

Submitted 5 June 2018; accepted 27

November 2018.

doi:10.1111/1754-9485.12844

Abstract

Introduction: This study quantified inter-observer contouring variations formultiple male pelvic structures, many of which are of emerging relevance forprostate cancer radiotherapy progression and toxicity response studies.Methods: Five prostate cancer patient datasets (CT and T2-weighted MR) weredistributed to 13 observers for contouring. CT structures contoured included theclinical target volume (CTV), seminal vesicles, rectum, colon, bowel bag, bladderand peri-rectal space (PRS). MR contours included CTV, trigone, membranousurethra, penile bulb, neurovascular bundle and multiple pelvic floor muscles.Contouring variations were assessed using the intraclass correlation coefficient(ICC), Dice similarity coefficient (DSC), and multiple additional metrics.Results: Clinical target volume (CT and MR), bladder, rectum and PRS contoursshowed excellent inter-observer agreement (median ICC = 0.97; 0.99; 1.00;0.95; 0.90, DSC = 0.83 � 0.05; 0.88 � 0.05; 0.93 � 0.03; 0.81 � 0.07; 0.80 �0.06, respectively). Seminal vesicle contours were more variable (ICC = 0.75,DSC = 0.73 � 0.14), while colon and bowel bag contoured volumes were consis-tent (ICC = 0.97; 0.97), but displayed poor overlap (DSC = 0.58 � 0.22;0.67 � 0.21). Smaller MR structures showed significant inter-observer variations,with poor overlap for trigone, membranous urethra, penile bulb, and left and rightneurovascular bundles (DSC = 0.44 � 0.22; 0.41 � 0.21; 0.66 � 0.21; 0.16 �0.17; 0.15 � 0.15). Pelvic floor muscles recorded moderate to strong inter-obser-ver agreement (ICC = 0.50–0.97), although large outlier variations were observed.Conclusions: Inter-observer contouring variation was significant for multiplepelvic structures contoured on MR.

Key words: contouring; delineation; inter/intra-observer variability; prostate.

© 2019 The Royal Australian and New Zealand College of Radiologists

Journal of Medical Imaging and Radiation Oncology 63 (2019) 264–271

264

Jour

nal o

f Med

ical

Imag

ing

and

Radi

atio

n O

ncol

ogy

Chapter 4. Contouring variability of male pelvic structures 99

Introduction

Contouring uncertainties are regarded as one of the lar-gest contributors to poor quality radiotherapy.1 Qualityassurance investigations during large clinical trials haverevealed contouring quality to be poor, with amendmentsrequired in over 80% of audited cases.2,3 Non-compliantcontouring has been shown to impact patient treatmentoutcomes during clinical trials.4,5

Inter-observer contouring variations for prostate, seminalvesicles, rectum and bladder have previously been investi-gated,6 with guidelines and training shown to significantlyreduce contouring variability.7–9 Additionally, clinical targetvolumes (CTVs) contoured on magnetic resonance (MR)imaging show less variation between observers than CTVscontoured on computed tomography (CT), with reducedvolumes arising from the better soft tissue definition of theprostatic apex on MR.10,11 Overlap metrics are often used toquantify contouring variations, with Dice similarity coeffi-cient (DSC) values exceeding 0.7 considered clinicallyacceptable for prostate, bladder and rectum contours.12,13

Growing numbers of studies have begun investigatingdose distributions delivered to neighbouring prostaticstructures as potential sources for toxicities followingradiotherapy, as well as becoming important organs-at-risk for modern treatment modalities.14 Structures inves-tigated include penile bulb and neurovascular bundle,15,16

trigone,17 peri-rectal space18 and various pelvic floormuscles.19 However, within these toxicity studies con-touring was performed either by a single expert, definedby simple geometric borders, or no information on thecontouring process was provided.

While consensus contouring guidelines for male pelvicstructures have been published,20 they do not cover therange of structures currently being investigated, nor dothey provide information for consensus contouring onMR. As poor-quality contouring could potentially maskfindings from clinical trials,4 and radiotherapy treatmentplanning relies on dose–volume information which is pro-gressively refined for more diverse anatomy, it is impor-tant to assess inter-observer contouring variations forthese emerging structures. This study aimed to documentand quantify contouring variations observed across multi-ple male pelvic structures on both CT and MR. In addition,it is intended that the derived multi-observer segmentedcohort will be used as a reference atlas for machine learn-ing based propagation of segmentations and their uncer-tainties to accrued and accruing clinical trial data.

Methods

Patient selection

Five prostate cancer patients were selected based on priorclustering work for a larger retrospective atlas segmenta-tion study.21,22 Using deformable registration and hierar-chical clustering, these five patients were determined to

best represent the diversity in anatomical information pre-sent within a large multi-centre clinical trial dataset.23 Eachpatient dataset contained co-registered whole pelvis plan-ning-CT (slice thickness: 2–2.5 mm) and a small field-of-view T2-weighted MR scan (slice thickness: 2 mm).24

Patient datasets were distributed to 13 observers (nineradiation oncologists, two medical physicists, one radiogra-pher, one radiation therapist (also known as planner ordosimetrist)) across five centres. Clinical experience levelranged from highly experienced to specialist trainees.

Contouring

Contoured structures for this study are listed in Table 1.The imaging modality utilized for contouring each struc-ture was pre-determined based on expert genitourinaryradiation oncologist advice for concurrent atlas-basedinvestigations. Study guidelines containing outsourcedexpert defined contours upon a sample patient were dis-tributed to all observers (Supporting Information). Addi-tionally, this sample patient along with associated expertcontours was distributed in DICOM format for additionalreference. Contouring was completed on treatment plan-ning systems available within the observer’s institution.Following contouring, post-processing for multiple struc-tures was required due to some observer’s failure toadhere to specific guideline requirements. Namely, allgenitourinary structures were removed from bowel bagcontours, while distal colon and bowel bag contours wereclipped 2 cm superior to rectum.

Interobserver variability assessment

Collated DICOM structure files were converted to NifTIformat for analysis within MilxView, an open-sourcedimage manipulation and processing platform developedby the Commonwealth Scientific and Industrial ResearchOrganisation (CSIRO).25 Intraclass correlation coeffi-cients (ICC) for contoured volumes were calculated forall structures. The ICC provides a statistical measure ofobserver reliability during contouring studies, and isdependent on the variance both within and betweenobserver contour volumes.26 An ICC greater than 0.75 is

Table 1. Contoured structures

CT Structures MR Structures

Prostate (CTV) Prostate (CTV)

Seminal vesicles (SV) Trigone

Rectum Membranous urethra (Memb. urethra)

Distal colon Penile bulb

Bowel bag Neurovascular bundle left (NVB left)

Bladder Neurovascular bundle right (NVB right)

Peri-rectal space (PRS) Internal anal sphincter (IAS)

External anal sphincter (EAS)

Puborectalis muscle (PRM)

Levator ani muscles (LAM)


Contouring variation of male pelvis

265


typically considered excellent agreement amongst obser-vers.27 Additionally, pairwise assessment of inter-obser-ver contouring variations was performed usingcontouring variation metrics DSC, volume similarity, sen-sitivity, specificity, 95th percentile Hausdorff distance(95_HAUS), mean absolute surface distance (MASD) andEuclidean centroid distance (Table e1).28

Statistics

Differences between CTV volumes contoured on CT andMR were assessed using two-sided Wilcoxon rank sumtesting (P < 0.05 considered significant).

Results

Mean, standard deviation and coefficient of variation (COV)of volumes for all structures are listed in Table 2 (COV = s-tandard deviation/mean). COV provides a normalized mea-sure of inter-observer variation that allows for comparisonbetween structures of differing volume. Pairwise DSC aregraphed in Figure 1, and tabulated in Table e2. Volume simi-larity, sensitivity, specificity, 95_HAUS, MASD and Euclideancentroid distance values are included in the SupportingInformation (Figs e2–e6, Tables e3–e8).

Clinical target volume

Observer CTV volumes showed excellent agreement onboth CT (ICC = 0.97, 0.90–1.00) and MR (ICC = 0.99,0.98–1.00). Median CT volumes were larger than MR(Table 2), with Wilcoxon analysis indicating significant dif-ferences across all five patients (Fig. 2). Mean 95_HAUSfor MR contours were smaller than CT (Table e5, Fig. e4),excluding patient 5 (MR: 10.39 mm, CT: 9.48 mm). Meanpairwise DSC across all patients for CT and MR were0.83 � 0.05 and 0.88 � 0.05, respectively.

Seminal vesicles

Seminal vesicle volume agreement was moderate(ICC = 0.75, 0.36–0.97), with reduced ICC attributed toa single observer contouring significantly larger volumesfor patients 2, 3 and 4 (COV = 0.58, 0,41, 0.86, respec-tively). These contours correspond to the outliersobserved in DSC (Fig. 1), volume similarity, sensitivity,95_HAUS and MASD (Figs e1, e2, e3, e5). Mean DSCacross all patients was 0.73 � 0.14.

Rectum

Excellent observer agreement in contoured rectum vol-umes and overlap was observed (ICC = 0.95, 0.85–0.99;DSC = 0.81 � 0.07). However, substantial variation in95_HAUS existed, with a maximum of 59.88 mmrecorded for patient 2 (Fig. e4, Table e6). Figure 3a illus-trates this example, where significantly different slices

for the superior rectum contour boundary were selectedby the two observers. Significant variation amongstobservers across all patients was observed at this bound-ary (Fig. 3b).

Distal colon and bowel bag

Distal colon and bowel bag were contoured 2 cm superi-orly to the rectum (Supporting Information). Due tolarge variations in rectum boundary contours, poor over-lap between observer contours was observed (Figs 1, 3a,Table e2). Mean DSC for distal colon and bowel bag were0.58 � 0.22 and 0.67 � 0.21, respectively. Large varia-tions in Euclidean centroid distances were observed(Fig. e6). ICC calculations revealed high conformity ofcontoured volumes for distal colon (ICC = 0.97, 0.92–1.00) and bowel bag (ICC = 0.97, 0.75–1.00).

Bladder

Bladder contours showed excellent agreement across vol-ume (ICC = 1.00, 1.00–1.00), overlap (DSC = 0.93 �0.03), and MASD (MASD = 0.76 mm–1.44 mm). Volumevaried significantly between patients (69.27 cm3–

400.13 cm3). Observer 95_HAUS distances revealed sig-nificant outliers for patient 4 (Fig. e4, Table e6), where asingle observer recorded pairwise 95_HAUS distances of72.20 mm–75.65 mm. Further inspection revealed a singleslice where a thin structure had been erroneously con-toured as bladder by this single observer.

Peri-rectal space

Excellent observer agreement (ICC = 0.90, 0.69–0.99) withacceptable overlap (DSC = 0.80 � 0.06) was recorded.Being a larger structure (mean volume = 299.54 cm3–

431.37 cm3), boundary metrics were more sensitive tointer-observer contouring variations than overlap metrics.While MASD was less than 10 mm for all but one pair of con-tours (Fig. e5), 95_HAUS conversely revealed all but onecontour pair having at least 10 mm separation (Fig. e4).

Trigone

Trigone ICC calculations returned a negative value(ICC = �0.29, �2.51 to 0.84), due to inter-observer con-touring variation per patient exceeding the volume differ-ences amongst patients. This is shown as the large spread involume similarity (Fig. e1, Table e3), where a median vol-ume similarity of 0.37 equates to a 45% increase betweencontoured volumes. Patient 5 had the highest agreement inobserver volumes (Table 2). Localization of the trigone waspoor, with DSC and sensitivity revealing zero overlapbetween some contours. Mean DSC was 0.44 � 0.22, withpatient 5 the most conformal. This patient also had signifi-cantly reduced 95_HAUS (9.25 � 5.47 mm) and MASD(2.13 � 1.88 mm).


D Roach et al.

266


Membranous urethra

Poor agreement between observer volumes was observed(ICC = 0.52, 0.09–0.92), with Figure 1 revealing overlap

between contours ranged from none to excellent (DSC = 0–0.89). Mean DSC was 0.41 � 0.21. All patients had poorlyreproducible contours, with COV ranging between 0.49and 0.74 (Table 2). Due to the small mean volumes

Table 2. Mean, standard deviation (SD) and coefficient of variation (COV) of observer contour volumes

Patient 1 2 3 4 5

CTV (CT) Mean (cm3) 54.44 43.38 52.64 33.98 74.02

SD (cm3) 6.58 6.13 7.96 7.12 11.14

COV 0.12 0.14 0.15 0.21 0.15

Seminal vesicles Mean (cm3) 15.91 17.81 25.52 12.39 18.09

SD (cm3) 2.26 10.28 10.52 10.65 3.26

COV 0.14 0.58 0.41 0.86 0.18

Rectum Mean (cm3) 51.13 37.69 38.61 54.79 58.94

SD (cm3) 6.20 6.81 2.78 9.95 6.82

COV 0.12 0.18 0.07 0.18 0.12

Distal colon Mean (cm3) 33.34 16.51 14.18 26.05 25.01

SD (cm3) 4.79 3.54 4.02 3.79 1.82

COV 0.14 0.21 0.28 0.15 0.07

Bowel bag Mean (cm3) 418.35 500.47 294.71 591.69 656.08

SD (cm3) 64.85 94.17 68.09 79.02 51.04

COV 0.16 0.19 0.23 0.13 0.08

Bladder Mean (cm3) 69.27 144.81 400.13 142.32 84.91

SD (cm3) 4.13 9.03 7.38 11.72 3.82

COV 0.06 0.06 0.02 0.08 0.05

Peri-rectal space Mean (cm3) 359.59 300.78 299.54 344.32 431.37

SD (cm3) 61.08 59.72 38.91 58.10 59.15

COV 0.17 0.20 0.13 0.17 0.14

CTV (MR) Mean (cm3) 43.83 31.04 42.34 23.41 62.32

SD (cm3) 3.81 4.31 3.77 4.92 5.39

COV 0.09 0.14 0.09 0.21 0.09

Trigone Mean (cm3) 2.72 3.17 2.95 2.70 3.23

SD (cm3) 0.82 1.29 1.26 1.01 0.81

COV 0.30 0.41 0.43 0.38 0.25

Membranous urethra Mean (cm3) 0.83 0.56 0.74 0.64 0.92

SD (cm3) 0.42 0.34 0.36 0.31 0.68

COV 0.51 0.61 0.49 0.49 0.74

Penile bulb Mean (cm3) 4.10 4.99 2.58 7.45 3.38

SD (cm3) 0.99 1.36 0.55 1.69 1.07

COV 0.24 0.27 0.21 0.23 0.32

Neurovascular bundle (Left) Mean (cm3) 0.42 0.75 0.80 0.47 0.83

SD (cm3) 0.24 0.72 0.68 0.35 0.88

COV 0.57 0.97 0.84 0.75 1.06

Neurovascular bundle (Right) Mean (cm3) 0.66 0.70 0.79 0.39 0.89

SD (cm3) 0.72 0.83 0.69 0.39 0.86

COV 1.10 1.19 0.88 0.99 0.97

Internal anal sphincter Mean (cm3) 9.02 7.51 8.04 10.85 8.92

SD (cm3) 2.82 3.67 2.00 3.45 5.97

COV 0.31 0.49 0.25 0.32 0.67

External anal sphincter Mean (cm3) 7.07 8.55 8.28 11.62 12.63

SD (cm3) 1.68 3.32 3.39 4.40 4.73

COV 0.24 0.39 0.41 0.38 0.37

Puborectalis muscle Mean (cm3) 13.93 12.80 8.95 8.57 23.58

SD (cm3) 3.03 4.43 2.21 2.42 5.79

COV 0.22 0.35 0.25 0.28 0.25

Levator ani muscles Mean (cm3) 15.04 19.87 26.56 27.63 20.68

SD (cm3) 4.72 9.04 12.26 10.31 7.47

COV 0.31 0.45 0.46 0.37 0.36



267


(0.56 cm3–0.92 cm3), contouring variations resulted in asignificant spread in volume similarity (Fig. e1).

Penile bulb

Penile bulb contours exhibited the highest reproducibilityamongst observers for MR pelvic structures, with excel-lent volume agreement (ICC = 0.96, 0.88–1.00). Still, aspread in contour overlap was observed (Fig. 1), withmean DSC = 0.66 � 0.21. Two patients recorded meanDSC exceeding 0.7 (Table e1).

Neurovascular bundles

The left and right NVBs were the least consistently con-toured structures in the study (ICC = 0.48, 0.03–0.91;0.30, �0.22 to 0.88, respectively), with significant varia-tion across most metrics. Overlap was exceptionally poor(Fig. 1), with mean DSC = 0.16 � 0.17 and 0.15 � 0.15,respectively. Significant volume differences arose, withmedian volume similarity for left and right NVBs of 0.69and 0.66 equating to approximately 100% volume differ-ences between contours. Patient 3 recorded the highest

Patien

t 1 -

CT

Patien

t 1 -

MR

Patien

t 2 -

CT

Patien

t 2 -

MR

Patien

t 3 -

CT

Patien

t 3 -

MR

Patien

t 4 -

CT

Patien

t 4 -

MR

Patien

t 5 -

CT

Patien

t 5 -

MR

20

30

40

50

60

70

80

90

Vol

ume

(cm

3 )

Fig. 2. Observer volume variations for CT and MR defined CTV. Significant differences between median CT and MR volume were observed for all patients

(P < 0.05).

CTV (CT)

SV

Rectu

m

Distal

Colon

Bowel

Bag

Bladde

rPRS

CTV (MR)

Trigon

e

Mem

b. U

reth

ra

Penile

Bulb

NVB (Lef

t)

NVB (Righ

t)IA

SEAS

PRMLA

M0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

DS

C

Fig. 1. Pairwise DSC for CT (blue) and MR (green) structures (see Table 1 for acronym definitions).


D Roach et al.

268


mean overlap (DSC = 0.28 � 0.15, 0.24 � 0.14), how-ever, this improved overlap had no significant differenceon volume (VOLSIM = 0.62 � 0.42, 0.59 � 0.45, respec-tively).

Pelvic floor muscles

External anal sphincter, PRM and LAM recorded ICC >

0.7 (ICC = 0.86, 0.51–0.98; 0.97, 0.88–1.00; 0.79,0.44–0.97, respectively), while IAS displayed poor vol-ume agreement (ICC = 0.50, �0.01 to 0.92). Largevariations in volume similarity for all pelvic floor mus-cles were observed (Table e2, Fig. e1). Overlap for IASand PRM (DSC = 0.61 � 0.20, 0.60 � 0.19) exceededthose observed for EAS and LAM (DSC = 0.47 � 0.24;

0.48 � 0.17). DSC greater than 0.7 was observedfor patients 1 and 3 for IAS (DSC = 0.71 � 0.12, 0.72 �0.10). LAM displayed poorest specificity amongst allstructures (Fig. e3), a consequence of large volumes(15.04 cm3–27.63 cm3) coupled with poor overlap(Fig. 1).

Discussion

The aim of this study was to document the range ofinter-observer contouring variations for multiple malepelvic structures. These structures have established rele-vance to the response to radiotherapy for prostate can-cer, including the response of the cancer and the toxicitycaused by it, are of emerging interest in such studies,and may be utilized in guiding radiotherapy planningoptimization.14 Additionally, automated contouring tech-niques (atlases, shape models, machine learning, etc.)for these structures all require high quality data inputduring development to ensure accurate automated con-touring.29,30 To the authors’ knowledge, no other studyhad investigated inter-observer contouring variationsacross this range of structures, therefore an assessmentwas performed.

Previous studies have shown specific contouring met-rics correlate with dosimetry,28,31 so multiple metricswere utilized during analysis. Contouring variation met-rics for CT structures (particularly CTV, bladder, and rec-tum) matched values commonly cited within theliterature.12 Additionally, significant reductions in MRdefined CTV volumes compared to CT (Fig. 2) supportfindings from prior studies investigating the utilization ofMR for prostate cancer radiotherapy.10,11

Neighbouring structures contoured on MR displayed sig-nificant inter-observer variation. As many MR structureswere anatomically small in volume (Table 2), small

Fig. 4. Variations in contouring for left and right NVB for Patient 1. For

many observers, no contour overlap was evident.

(a) (b)

Fig. 3. (a) Sagittal view of patient 2, where variations in superior boundary for rectum contours between two observers (filled contours) resulted in poor

overlap of bowel bag contours (outlined). (b) Density distribution of all observer rectum contours for patient 2, ranging from 100% observer agreement (pur-

ple) to only a single observer (red).



269


contouring variations led to significant differences instructure localization (Fig. 4). Additionally, observers par-ticipating in this study came from multiple institutions,preventing the implementation of teaching seminars priorto commencement of the study to reduce contouring vari-ations.7 Subsequently, while explicit guidelines and a sam-ple patient had been distributed for reference, contouringexperience levels were found to be a concern, promptingus to collate the experience levels of the observers in con-touring these structures (Tables e9–e10).

Assessment of experience levels revealed that non-radiation oncologists (non-ROs) often had no prior expe-rience contouring the multiple MR structures comparedto radiation oncologists (ROs). However, further Wilcoxonsigned-rank analysis revealed no statistically significantdifferences in contoured volumes between these twoobserver groups (Table e11). Consequently, the utiliza-tion of inexperienced non-RO contours was consideredrepresentative of the spread in contouring variation thatwould be encountered clinically for these structures.

Other studies have investigated inter-observer con-touring variations for some of these structures.32,33 Inthese studies a single structure’s contour was investi-gated, which Cassidy et al.32 noted would produce differ-ing inter-observer contouring variations than if multiplestructures were contoured simultaneously. In their study,they observed significantly improved overlap for NVBthan was recorded in this study (ICC = 0.89,DSC = 0.72 � 0.09). Differences in their study designinclude the implementation of training exercises prior tocontouring, utilization of an expert-defined contour for‘gold-standard’ analysis (opposed to pairwise analysis,which results in a larger spread in metric values), anddifferent MR scan sequences employed.

Additionally, in Cassidy et al.’s32 study screening ofpatients was performed, with patients excluded if extrapro-static extension was observed by a radiologist. Within thisstudy no screening was employed, as patient selection wasbased on clustering work completed for an ongoing atlasanalysis.22 As only 50% of men have NVBs in standard pos-tero-lateral positions forming distinct bundles,34 it is likelythat multiple patients within this study had no identifiableNVBs. Observers consequently contoured NVBs using clini-cal knowledge, rather than image intensity informationderived from the patient scans, leading to the large varia-tions in localization of the NVBs (Fig. 4).

This study highlights that greater efforts are required toreduce inter-observer contouring variations for these struc-tures. As outcomes of clinical trials can be masked by poorquality contouring,4 it is imperative that studies investigat-ing these emerging structures have some assessment ofcontouring quality to ensure that statistically rigorous con-clusions can be derived. This study provides a benchmarkfor the extent of inter-observer contouring variations thatcan be expected for these structures. Additionally, thisstudy emphasizes the need for continual and adequateobserver training, as well as utilization of contouring

guidelines, to reduce the inter-observer variability in con-touring these structures on MR.7–9

In conclusion, inter-observer contouring variations forCT-defined male pelvic structures matched the variationsrecorded within the literature. Smaller MR-defined struc-tures, not regularly encountered by observers, exhibitedsignificantly greater contouring variability. It is thereforeessential that studies include contouring variationassessments amongst observers before analysis of thesestructures can be undertaken.

Acknowledgements

The project was funded by NHMRC project grant number1077788.

References

1. Weiss E, Hess CF. The impact of gross tumor volume(GTV) and clinical target volume (CTV) definition onthe total accuracy in radiotherapy. Strahlenther Onkol2003; 179: 21–30.

2. Kachnic LA, Winter K, Myerson RJ et al. RTOG 0529: aphase 2 evaluation of dose-painted intensitymodulated radiation therapy in combination with 5-fluorouracil and mitomycin-C for the reduction of acutemorbidity in carcinoma of the anal canal. Int J RadiatOncol Biol Phys 2013; 86: 27–33.

3. Fairchild A, Weber DC, Bar-Deroma R et al. Qualityassurance in the EORTC 22033–26033/CE5 phase IIIrandomized trial for low grade glioma: the digitalindividual case review. Radiother Oncol 2012; 103:287–92.

4. Peters LJ, O’Sullivan B, Giralt J et al. Critical impact ofradiotherapy protocol compliance and quality in thetreatment of advanced head and neck cancer: resultsfrom TROG 02.02. J Clin Oncol 2010; 28: 2996–3001.

5. Abrams RA, Winter KA, Regine WF et al. Failure toadhere to protocol specified radiation therapyguidelines was associated with decreased survival inRTOG 9704—a phase III trial of adjuvantchemotherapy and chemoradiotherapy for patientswith resected adenocarcinoma of the pancreas. Int JRadiat Oncol Biol Phys 2012; 82: 809–16.

6. Vinod SK, Jameson MG, Min M, Holloway LC.Uncertainties in volume delineation in radiationoncology: a systematic review and recommendationsfor future studies. Radiother Oncol 2016; 121: 169–79.

7. Vinod SK, Min M, Jameson MG, Holloway LC. A reviewof interventions to reduce inter-observer variability involume delineation in radiation oncology. J MedImaging Radiat Oncol 2016; 60: 393–406.

8. Khoo EL, Schick K, Plank AW et al. Prostate contouringvariation: can it be fixed? Int J Radiat Oncol Biol Phys2012; 82: 1923–9.

9. Mitchell DM, Perry L, Smith S et al. Assessing the effectof a contouring protocol on postprostatectomyradiotherapy clinical target volumes and interphysicianvariation. Int J Radiat Oncol Biol Phys 2009; 75: 990–3.


D Roach et al.

270


10. Debois M, Oyen R, Maes F et al. The contribution ofmagnetic resonance imaging to the three-dimensionaltreatment planning of localized prostate cancer. Int JRadiat Oncol Biol Phys 1999; 45: 857–65.

11. Rasch C, Barillot I, Remeijer P, Touw A, van Herk M,Lebesque JV. Definition of the prostate in CT and MRI:a multi-observer study. Int J Radiat Oncol Biol Phys1999; 43: 57–66.

12. Sharp G, Fritscher KD, Pekar V et al. Vision 20/20:perspectives on automated image segmentation forradiotherapy. Med Phys 2014; 41: 050902.

13. Dice LR. Measures of the amount of ecologic associationbetween species. Ecology 1945; 26: 297–302.

14. Goddard L, Brodin NP, Bodner WR, Garg M, Tom�e WA.Comparing photon and proton based hypofractionedsbrt for prostate cancer accounting for robustness andrealistic treatment deliverability. Brit J Radiol 2018;91: 20180010.

15. Mendenhall WM, Henderson RH, Indelicato DJ, KeoleSR, Mendenhall NP. Erectile dysfunction afterradiotherapy for prostate cancer. Am J Clin Oncol2009; 32: 443–7.

16. van der Wielen GJ, Mulhall JP, Incrocci L. Erectiledysfunction after radiotherapy for prostate cancer andradiation dose to the penile structures: a criticalreview. Radiother Oncol 2007; 84: 107–13.

17. Ghadjar P, Zelefsky MJ, Spratt DE et al. Impact of doseto the bladder trigone on long-term urinary functionafter high-dose intensity modulated radiation therapyfor localized prostate cancer. Int J Radiat Oncol BiolPhys 2014; 88: 339–44.

18. Gulliford S, Ghose S, Ebert M et al. Radiotherapy dose-distribution to the perirectal fat space (PRS) is relatedto gastrointestinal control-related complications. ClinTranslat Rad Oncol 2017; 7: 62–70.

19. Smeenk RJ, Hoffmann AL, Hopman WP, van Lin ENT,Kaanders JH. Dose-effect relationships for individualpelvic floor muscles and anorectal complaints afterprostate radiotherapy. Int J Radiat Oncol Biol Phys2012; 83: 636–44.

20. Gay HA, Barthold HJ, O’Meara E et al. Pelvic normaltissue contouring guidelines for radiation therapy: aRadiation Therapy Oncology Group consensus panelatlas. Int J Radiat Oncol Biol Phys 2012; 83: e353–62.

21. Dowling JA, Sun J, Pichler P et al. Automatic substitutecomputed tomography generation and contouring formagnetic resonance imaging (MRI)-alone externalbeam radiation therapy from standard MRI sequences.Int J Radiat Oncol Biol Phys 2015; 93: 1144–53.

22. Kennedy AM, Dowling JA, Ghose S, Rivest-H�enault D,Ebert MA. Cluster Based selection of CT exemplarsfrom a clinical dataset for inter-patient registration anddose mapping. International Conference on the use ofComputers in Radiation Therapy. London, 2016.

23. Denham JW, Joseph D, Lamb DS, et al. Short-termandrogen suppression and radiotherapy versusintermediate-term androgen suppression andradiotherapy, with or without zoledronic acid, in men

with locally advanced prostate cancer (TROG 03.04RADAR): an open-label, randomised, phase 3 factorial.Lancet Oncol 2014; 15: 1076–89.

24. Dowling JA, Lambert J, Parker J et al. An atlas-basedelectron density mapping method for magneticresonance imaging (MRI)-alone treatment planningand adaptive MRI-based prostate radiation therapy. IntJ Radiat Oncol Biol Phys 2012; 83: e5–11.

25. Burdett N, Fripp J, Bourgeat P, Salvado O. MILXView: AMedical Imaging, Analysis and Visualization Platform.Springer, E-Health, 2010; 177–86.

26. Fotina I, L€utgendorf-Caucig C, Stock M, P€otter R,Georg D. Critical discussion of evaluation parametersfor inter-observer variability in target definition forradiation therapy. Strahlenther Onkol 2012; 188:160–7.

27. Bartko JJ. The intraclass correlation coefficient as ameasure of reliability. Psychol Rep 1966; 19: 3–11.

28. Roach D, Jameson MG, Dowling JA et al. Correlationsbetween contouring similarity metrics and simulatedtreatment outcome for prostate radiotherapy. PhysMed Biol 2018; 63: 035001.

29. El Naqa I, Murphy MJ. What is Machine Learning?Machine Learning in Radiation Oncology. Springer,Cham, 2015; 3–11.

30. Cruz JA, Wishart DS. Applications of machine learningin cancer prediction and prognosis. Cancer Inform2006; 2: 59–77.

31. Jameson MG, Kumar S, Vinod SK, Metcalfe PE, HollowayLC. Correlation of contouring variation with modeledoutcome for conformal non-small cell lung cancerradiotherapy. Radiother Oncol 2014; 112: 332–6.

32. Cassidy RJ, Nour SG, Liu T et al. Reproducibility incontouring the neurovascular bundle for prostate cancerradiotherapy. Pract Rad Oncol 2017; 8: e125–31.

33. Carillo V, Cozzarini C, Perna L et al. Contouringvariability of the penile bulb on CT images: quantitativeassessment using a generalized concordance index. IntJ Radiat Oncol Biol Phys 2012; 84: 841–6.

34. Kiyoshima K, Yokomizo A, Yoshida T et al. Anatomicalfeatures of periprostatic tissue and its surroundings: ahistological analysis of 79 radical retropubicprostatectomy specimens. Jpn J Clin Oncol 2004; 34:463–8.

Supporting Information

Additional Supporting Information may be found in theonline version of this article at the publisher’s web-site:

Appendix S1. Contouring variation metric derivations,additional inter-observer contouring variation analysisusing multiple additional metrics, and assessment ofobserver experience levels and profession on contouringvariations.

Appendix S2. Multi-observer study protocol distributedto all observers prior to contouring.



271


1

Treatment planning system biases

Additional analysis investigated biases in contouring arising from treatment planning software used,

whether Pinnacle3® (Philips Healthcare, Best, Netherlands) or Eclipse™ (Varian Medical Systems, Palo

Alto, CA), both across individual patients and the complete patient cohort. Multiple hypothesis

testing required Bonferroni corrections of p < 0.05/102 = 0.00049 for significance. Wilcoxon testing

revealed no significant differences in volume contoured for all structures (p < 0.00049), across

individual patients as well as the entire patient dataset.



2

Table e1: Contouring variation metrics

Metric Equations and Derivations

Volume Similarity VOLSIM =

Y-X

(X+Y)/2

Dice Similarity Coefficient (DSC) DSC =

2|X∩Y|

|X|+|Y|

True Positives Number of voxels contoured by observer X that also

lie within observer Y contour

False Positives Number of voxels contoured by observer X that do

not lie within observer Y contour

True Negatives Number of voxels not contoured by observer X that

also do not lie within observer Y contour

False Negatives Number of voxels not contoured by observer X that

lie within observer Y contour

Sensitivity p =

True Positives

True Positives + False Negatives

Specificity q =

True Negatives

True Negatives + False Positives

95th percentile Hausdorff distance

(95_HAUS)

HDasym(XS,YS) = 95th percentilex∈XS

( miny∈YS

d(x,y))

HD(XS,YS) =max(HDasym(XS,YS),HDasym(YS,XS))

Mean absolute surface distance (MASD) MASD =

1

NXS+NYS

( ∑ miny∈YS

d(x,y)

x∈XS

+∑ minx∈XS

d(y,x)

y∈YS

)

Euclidean centroid distance Straight line distance between centre-of-mass of

observer contours

X: Number of voxels within observer X contour

Y: Number of voxels within observer Y contour

XS: Surface points of observer X contour

YS: Surface points of observer Y contour


3

NXS: Number of surface points of observer X contour

NYS: Number of surface points of observer Y contour

d(x,y): Euclidean distance from point x to point y


4

Table e2: DSC distribution

Patient 1 2 3 4 5

CTV (CT) Mean 0.83 0.86 0.84 0.79 0.83

S.D. 0.03 0.05 0.03 0.07 0.04

C.O.V. 0.04 0.05 0.04 0.09 0.05

Seminal Vesicles Mean 0.72 0.72 0.79 0.66 0.73

S.D. 0.06 0.16 0.13 0.20 0.07

C.O.V. 0.09 0.22 0.16 0.30 0.09

Rectum Mean 0.76 0.80 0.83 0.80 0.85

S.D. 0.07 0.07 0.05 0.09 0.03

C.O.V. 0.09 0.08 0.06 0.11 0.04

Distal Colon Mean 0.65 0.60 0.45 0.49 0.71

S.D. 0.14 0.19 0.23 0.29 0.13

C.O.V. 0.21 0.31 0.52 0.58 0.19

Bowel Bag Mean 0.77 0.74 0.55 0.57 0.73

S.D. 0.11 0.14 0.24 0.26 0.15

C.O.V. 0.14 0.19 0.43 0.45 0.21

Bladder Mean 0.90 0.92 0.97 0.91 0.92

S.D. 0.02 0.02 0.00 0.03 0.01

C.O.V. 0.02 0.02 0.00 0.03 0.02

Peri-Rectal Space Mean 0.77 0.79 0.82 0.81 0.82

S.D. 0.07 0.07 0.04 0.05 0.05

C.O.V. 0.09 0.09 0.05 0.06 0.06

CTV (MR) Mean 0.90 0.87 0.90 0.83 0.87

S.D. 0.02 0.04 0.02 0.06 0.05

C.O.V. 0.02 0.04 0.03 0.07 0.05

Mean 0.46 0.42 0.43 0.29 0.58


5

Trigone S.D. 0.23 0.17 0.21 0.24 0.15

C.O.V. 0.50 0.40 0.48 0.82 0.25

Membranous Urethra Mean 0.47 0.45 0.40 0.52 0.27

S.D. 0.20 0.17 0.20 0.13 0.23

C.O.V. 0.41 0.37 0.50 0.25 0.87

Penile Bulb Mean 0.78 0.59 0.68 0.71 0.55

S.D. 0.16 0.24 0.12 0.18 0.23

C.O.V. 0.20 0.41 0.18 0.26 0.43

Neurovascular Bundle (Left) Mean 0.07 0.16 0.28 0.17 0.13

S.D. 0.12 0.16 0.15 0.17 0.16

C.O.V. 1.70 0.99 0.52 0.98 1.21

Neurovascular Bundle (Right) Mean 0.07 0.09 0.24 0.12 0.21

S.D. 0.09 0.14 0.14 0.15 0.15

C.O.V. 1.32 1.45 0.59 1.19 0.71

Internal Anal Sphincter Mean 0.71 0.58 0.72 0.67 0.43

S.D. 0.12 0.21 0.10 0.13 0.22

C.O.V. 0.17 0.35 0.13 0.20 0.50

External Anal Sphincter Mean 0.54 0.43 0.53 0.48 0.37

S.D. 0.24 0.26 0.13 0.18 0.28

C.O.V. 0.45 0.59 0.25 0.37 0.76

Puborectalis Muscle Mean 0.66 0.51 0.61 0.53 0.69

S.D. 0.16 0.16 0.20 0.15 0.19

C.O.V. 0.24 0.32 0.33 0.28 0.27

Levator Ani Muscles Mean 0.44 0.44 0.49 0.58 0.46

S.D. 0.18 0.17 0.13 0.15 0.15

C.O.V. 0.41 0.37 0.27 0.26 0.33


6

Figure e1: Pairwise Volume Similarity for CT (blue) and MR (green) structures. As no distinction between which observer contour would be treated as the

gold standard volume, absolute values of volume similarity are reported here.

Chapter

4.Contouring

variabilityofm

alepelvic

structures112

7

Table e3: Absolute Volume Similarity Distribution

Patient 1 2 3 4 5

CTV (CT) Mean 0.16 0.17 0.17 0.25 0.18

S.D. 0.10 0.14 0.14 0.20 0.12

C.O.V. 0.63 0.85 0.80 0.80 0.65


S.D. 0.11 0.38 0.31 0.47 0.14

C.O.V. 0.67 1.22 1.29 1.17 0.65

Rectum Mean 0.14 0.20 0.09 0.21 0.13

S.D. 0.11 0.14 0.06 0.19 0.09

C.O.V. 0.75 0.69 0.66 0.89 0.70


S.D. 0.12 0.16 0.22 0.12 0.06

C.O.V. 0.68 0.60 0.65 0.65 0.65

Bowel Bag Mean 0.19 0.23 0.27 0.15 0.09

S.D. 0.12 0.15 0.19 0.10 0.06

C.O.V. 0.66 0.64 0.69 0.66 0.65

Bladder Mean 0.07 0.07 0.02 0.09 0.05

S.D. 0.06 0.05 0.02 0.08 0.04

C.O.V. 0.81 0.75 0.78 0.83 0.66


S.D. 0.14 0.16 0.10 0.13 0.10

C.O.V. 0.76 0.70 0.64 0.65 0.67

CTV (MR) Mean 0.10 0.14 0.10 0.17 0.10

S.D. 0.07 0.11 0.08 0.18 0.08

C.O.V. 0.66 0.79 0.75 1.08 0.77

Mean 0.38 0.48 0.52 0.43 0.31


8

Trigone S.D. 0.26 0.30 0.32 0.27 0.24

C.O.V. 0.69 0.61 0.62 0.63 0.77


S.D. 0.38 0.35 0.40 0.35 0.55

C.O.V. 0.60 0.77 0.65 0.65 0.58

Penile Bulb Mean 0.29 0.33 0.26 0.27 0.42

S.D. 0.31 0.29 0.19 0.31 0.46

C.O.V. 1.07 0.86 0.72 1.17 1.11


S.D. 0.38 0.50 0.42 0.41 0.59

C.O.V. 0.61 0.53 0.67 0.86 0.56


S.D. 0.49 0.61 0.45 0.51 0.51

C.O.V. 0.56 0.72 0.76 0.58 0.74


S.D. 0.32 0.51 0.19 0.29 0.49

C.O.V. 0.84 0.81 0.64 0.71 0.61


S.D. 0.19 0.43 0.35 0.29 0.30

C.O.V. 0.68 0.83 0.73 0.63 0.73


S.D. 0.20 0.28 0.20 0.21 0.30

C.O.V. 0.76 0.66 0.67 0.63 0.99


S.D. 0.25 0.36 0.37 0.32 0.30

C.O.V. 0.65 0.65 0.64 0.70 0.68


9

Figure e2: Pairwise sensitivity scores for CT (blue) and MR (green) structures

Chapter

4.Contouring

variabilityofm

alepelvic

structures115

10

Table e4: Sensitivity Distribution

Patient 1 2 3 4 5

CTV (CT) Mean 0.84 0.83 0.85 0.79 0.82

S.D. 0.08 0.11 0.11 0.16 0.11

C.O.V. 0.10 0.13 0.13 0.20 0.13


S.D. 0.09 0.20 0.18 0.22 0.10

C.O.V. 0.14 0.28 0.23 0.34 0.14

Rectum Mean 0.75 0.79 0.82 0.80 0.85

S.D. 0.10 0.11 0.07 0.13 0.07

C.O.V. 0.13 0.14 0.08 0.17 0.08


S.D. 0.14 0.20 0.24 0.28 0.14

C.O.V. 0.22 0.35 0.51 0.58 0.20

Bowel Bag Mean 0.73 0.73 0.57 0.57 0.74

S.D. 0.13 0.19 0.24 0.27 0.15

C.O.V. 0.18 0.26 0.42 0.47 0.21

Bladder Mean 0.92 0.91 0.96 0.90 0.92

S.D. 0.04 0.05 0.01 0.07 0.03

C.O.V. 0.04 0.05 0.01 0.08 0.03


S.D. 0.11 0.12 0.09 0.10 0.08

C.O.V. 0.14 0.15 0.11 0.12 0.10

CTV (MR) Mean 0.92 0.85 0.91 0.82 0.86

S.D. 0.05 0.09 0.05 0.12 0.07

C.O.V. 0.05 0.10 0.06 0.14 0.08

Mean 0.45 0.45 0.47 0.34 0.64


11

Trigone S.D. 0.24 0.19 0.26 0.29 0.20

C.O.V. 0.53 0.43 0.56 0.86 0.30


S.D. 0.21 0.23 0.25 0.20 0.26

C.O.V. 0.47 0.49 0.62 0.38 0.86

Penile Bulb Mean 0.80 0.64 0.70 0.77 0.61

S.D. 0.14 0.29 0.15 0.17 0.26

C.O.V. 0.17 0.45 0.22 0.23 0.42


S.D. 0.12 0.15 0.18 0.17 0.19

C.O.V. 1.80 1.09 0.65 1.02 1.28


S.D. 0.11 0.13 0.19 0.19 0.19

C.O.V. 1.57 1.41 0.77 1.31 0.78


S.D. 0.16 0.29 0.16 0.21 0.27

C.O.V. 0.22 0.47 0.21 0.30 0.52


S.D. 0.23 0.28 0.19 0.19 0.29

C.O.V. 0.45 0.66 0.34 0.43 0.79


S.D. 0.18 0.18 0.22 0.20 0.21

C.O.V. 0.29 0.40 0.37 0.37 0.32


S.D. 0.24 0.27 0.22 0.23 0.19

C.O.V. 0.51 0.54 0.45 0.40 0.36


12

Figure e3: Pairwise specificity scores for CT (blue) and MR (green) structures

Chapter

4.Contouring

variabilityofm

alepelvic

structures118

13

Table e5: Specificity Distribution

Patient 1 2 3 4 5

CTV (CT) Mean 0.9998 1.0000 0.9999 0.9999 0.9999

S.D. 0.0001 0.0000 0.0000 0.0000 0.0001

C.O.V. 0.0001 0.0000 0.0000 0.0000 0.0001


S.D. 0.0000 0.0001 0.0001 0.0001 0.0000

C.O.V. 0.0000 0.0001 0.0001 0.0001 0.0000

Rectum Mean 0.9998 0.9999 1.0000 0.9999 0.9999

S.D. 0.0001 0.0001 0.0000 0.0001 0.0001

C.O.V. 0.0001 0.0001 0.0000 0.0001 0.0001


S.D. 0.0001 0.0000 0.0000 0.0001 0.0000

C.O.V. 0.0001 0.0000 0.0000 0.0001 0.0000

Bowel Bag Mean 0.9988 0.9989 0.9989 0.9980 0.9979

S.D. 0.0009 0.0006 0.0007 0.0013 0.0012

C.O.V. 0.0009 0.0006 0.0007 0.0013 0.0012

Bladder Mean 0.9999 0.9999 0.9999 0.9999 0.9999

S.D. 0.0001 0.0000 0.0000 0.0001 0.0000

C.O.V. 0.0001 0.0000 0.0000 0.0001 0.0000


S.D. 0.0009 0.0005 0.0002 0.0004 0.0006

C.O.V. 0.0009 0.0005 0.0002 0.0004 0.0006

CTV (MR) Mean 0.9989 0.9994 0.9991 0.9993 0.9986

S.D. 0.0007 0.0006 0.0006 0.0007 0.0010

C.O.V. 0.0007 0.0006 0.0006 0.0007 0.0010

Mean 0.9997 0.9996 0.9997 0.9995 0.9997


14

Trigone S.D. 0.0002 0.0003 0.0002 0.0002 0.0001

C.O.V. 0.0002 0.0003 0.0002 0.0002 0.0001


S.D. 0.0000 0.0000 0.0000 0.0001 0.0001

C.O.V. 0.0000 0.0000 0.0000 0.0001 0.0001

Penile Bulb Mean 0.9998 0.9996 0.9998 0.9995 0.9997

S.D. 0.0002 0.0003 0.0001 0.0003 0.0002

C.O.V. 0.0002 0.0003 0.0001 0.0003 0.0002


S.D. 0.0000 0.0001 0.0001 0.0001 0.0002

C.O.V. 0.0000 0.0001 0.0001 0.0001 0.0002


S.D. 0.0001 0.0001 0.0001 0.0001 0.0002

C.O.V. 0.0001 0.0001 0.0001 0.0001 0.0002


S.D. 0.0005 0.0006 0.0003 0.0006 0.0011

C.O.V. 0.0005 0.0006 0.0003 0.0006 0.0011


S.D. 0.0004 0.0006 0.0006 0.0006 0.0011

C.O.V. 0.0004 0.0006 0.0006 0.0006 0.0011


S.D. 0.0004 0.0005 0.0004 0.0003 0.0006

C.O.V. 0.0004 0.0005 0.0004 0.0003 0.0006


S.D. 0.0009 0.0014 0.0016 0.0012 0.0014

C.O.V. 0.0009 0.0014 0.0016 0.0012 0.0014


15

Figure e4: Pairwise Hausdorff Distance scores for CT (blue) and MR (green) structures

Chapter

4.Contouring

variabilityofm

alepelvic

structures121

16

Table e6: Hausdorff Distance Distribution

Patient 1 2 3 4 5

CTV (CT) Mean (mm) 10.31 7.41 8.61 9.48 9.84

S.D. (mm) 2.31 1.54 1.53 2.16 2.03

C.O.V. 0.22 0.21 0.18 0.23 0.21

Seminal Vesicles Mean (mm) 11.15 12.61 11.99 12.83 11.06

S.D. (mm) 3.00 11.70 10.20 9.73 2.96

C.O.V. 0.27 0.93 0.85 0.76 0.27

Rectum Mean (mm) 17.59 23.55 18.49 21.84 17.41

S.D. (mm) 9.57 17.72 10.36 12.32 9.34

C.O.V. 0.54 0.75 0.56 0.56 0.54

Distal Colon Mean (mm) 24.70 20.97 29.51 30.29 15.39

S.D. (mm) 9.20 7.85 14.41 15.55 7.22

C.O.V. 0.37 0.37 0.49 0.51 0.47

Bowel Bag Mean (mm) 19.58 27.61 24.30 26.91 18.08

S.D. (mm) 7.06 11.18 9.26 8.89 4.75

C.O.V. 0.36 0.40 0.38 0.33 0.26

Bladder Mean (mm) 8.16 6.07 6.25 20.57 6.28

S.D. (mm) 3.06 1.65 1.46 27.24 1.27

C.O.V. 0.37 0.27 0.23 1.32 0.20

Peri-Rectal Space Mean (mm) 25.05 22.33 25.11 24.76 32.92

S.D. (mm) 6.42 5.00 6.95 7.51 11.49

C.O.V. 0.26 0.22 0.28 0.30 0.35

CTV (MR) Mean (mm) 6.04 6.83 6.11 7.43 10.39

S.D. (mm) 1.35 1.82 1.42 2.46 4.30

C.O.V. 0.22 0.27 0.23 0.33 0.41

Mean (mm) 14.49 14.16 17.54 19.74 9.25


17

Trigone S.D. (mm) 9.62 5.99 7.90 8.60 5.47

C.O.V. 0.66 0.42 0.45 0.44 0.59

Membranous Urethra Mean (mm) 6.72 5.33 7.75 5.32 10.17

S.D. (mm) 3.61 1.93 3.60 1.99 5.13

C.O.V. 0.54 0.36 0.46 0.37 0.50

Penile Bulb Mean (mm) 6.65 12.22 7.96 12.24 9.82

S.D. (mm) 2.84 6.13 3.85 6.16 3.92

C.O.V. 0.43 0.50 0.48 0.50 0.40

Neurovascular Bundle (Left) Mean (mm) 20.24 14.83 16.88 11.58 20.30

S.D. (mm) 7.46 6.00 8.85 4.03 7.40

C.O.V. 0.37 0.40 0.52 0.35 0.36

Neurovascular Bundle (Right) Mean (mm) 20.34 17.13 17.12 14.05 14.67

S.D. (mm) 6.64 7.46 8.51 5.19 6.38

C.O.V. 0.33 0.44 0.50 0.37 0.43

Internal Anal Sphincter Mean (mm) 8.55 9.64 8.50 10.79 18.01

S.D. (mm) 3.51 4.11 3.45 4.46 7.96

C.O.V. 0.41 0.43 0.41 0.41 0.44

External Anal Sphincter Mean (mm) 9.07 10.19 10.52 13.51 19.33

S.D. (mm) 6.03 4.19 3.52 5.75 11.23

C.O.V. 0.67 0.41 0.33 0.43 0.58

Puborectalis Muscle Mean (mm) 10.46 13.45 8.15 12.33 12.45

S.D. (mm) 5.10 6.94 3.88 5.58 10.73

C.O.V. 0.49 0.52 0.48 0.45 0.86

Levator Ani Muscles Mean (mm) 25.84 24.47 22.64 18.53 28.26

S.D. (mm) 11.23 7.26 8.21 9.77 9.62

C.O.V. 0.43 0.30 0.36 0.53 0.34


18

Figure e5: Pairwise Mean Absolute Surface Distance scores for CT (blue) and MR (green) structures

Chapter

4.Contouring

variabilityofm

alepelvic

structures124

19

Table e7: Mean Absolute Surface Distance Distribution

Patient 1 2 3 4 5

CTV (CT) Mean (mm) 2.12 1.56 1.86 2.18 2.24

S.D. (mm) 0.49 0.57 0.45 0.91 0.56

C.O.V. 0.23 0.36 0.24 0.42 0.25


S.D. (mm) 0.59 3.17 2.42 3.19 0.85

C.O.V. 0.35 1.23 1.20 1.26 0.39

Rectum Mean (mm) 2.92 4.14 2.07 3.21 1.97

S.D. (mm) 1.59 3.86 1.46 2.48 0.94

C.O.V. 0.54 0.93 0.71 0.77 0.48


S.D. (mm) 2.52 3.27 8.19 5.79 1.72

C.O.V. 0.57 0.69 0.97 0.76 0.69

Bowel Bag Mean (mm) 3.84 4.95 6.65 7.53 4.79

S.D. (mm) 1.82 2.56 4.72 4.13 2.65

C.O.V. 0.47 0.52 0.71 0.55 0.55

Bladder Mean (mm) 0.98 0.88 0.76 1.44 0.89

S.D. (mm) 0.23 0.20 0.10 0.76 0.21

C.O.V. 0.23 0.23 0.14 0.53 0.24


S.D. (mm) 1.45 1.70 1.35 1.18 2.05

C.O.V. 0.33 0.41 0.35 0.29 0.43

CTV (MR) Mean (mm) 1.11 1.28 1.07 1.61 1.68

S.D. (mm) 0.33 0.60 0.34 0.81 0.79

C.O.V. 0.30 0.47 0.32 0.50 0.47

Mean (mm) 4.98 3.55 5.22 7.45 2.13


20

Trigone S.D. (mm) 6.95 2.58 4.89 6.48 1.88

C.O.V. 1.40 0.73 0.94 0.87 0.88


S.D. (mm) 1.69 1.00 1.57 0.81 3.31

C.O.V. 0.67 0.51 0.60 0.45 0.69


S.D. (mm) 1.03 1.75 0.59 1.41 2.11

C.O.V. 0.81 0.60 0.39 0.62 0.65


S.D. (mm) 5.03 3.05 2.52 2.63 5.29

C.O.V. 0.50 0.53 0.53 0.55 0.58


S.D. (mm) 4.69 3.34 2.36 3.91 2.97

C.O.V. 0.47 0.51 0.48 0.55 0.57


S.D. (mm) 1.40 1.96 1.17 1.89 4.08

C.O.V. 0.52 0.59 0.51 0.60 0.64


S.D. (mm) 1.84 1.70 1.37 2.33 8.66

C.O.V. 0.82 0.60 0.50 0.64 1.06


S.D. (mm) 2.06 2.13 1.73 1.88 3.10

C.O.V. 0.87 0.69 0.85 0.65 1.15


S.D. (mm) 5.02 2.80 2.29 2.73 3.33

C.O.V. 0.91 0.67 0.61 0.95 0.64


21

Figure e6: Pairwise Euclidean Centroid Distance scores for CT (blue) and MR (green) structures

Chapter

4.Contouring

variabilityofm

alepelvic

structures127

22

Table e8: Euclidean Centroid Distance Distribution

Patient 1 2 3 4 5

CTV (CT) Mean (mm) 2.63 1.41 1.98 2.27 2.76

S.D. (mm) 1.24 0.55 0.87 1.34 1.09

C.O.V. 0.47 0.39 0.44 0.59 0.40


S.D. (mm) 1.39 2.22 2.44 2.48 1.39

C.O.V. 0.43 0.85 0.94 0.70 0.46

Rectum Mean (mm) 9.50 8.55 6.50 8.33 4.59

S.D. (mm) 5.65 7.16 4.37 6.58 2.65

C.O.V. 0.59 0.84 0.67 0.79 0.58


S.D. (mm) 8.11 6.20 13.17 15.91 11.35

C.O.V. 0.64 0.66 0.67 0.70 0.72

Bowel Bag Mean (mm) 7.34 8.52 12.93 12.40 9.07

S.D. (mm) 3.99 4.89 9.03 7.22 5.86

C.O.V. 0.54 0.57 0.70 0.58 0.65

Bladder Mean (mm) 1.87 0.97 0.66 1.62 1.17

S.D. (mm) 1.09 0.36 0.29 1.13 0.72

C.O.V. 0.58 0.37 0.44 0.70 0.61


S.D. (mm) 3.26 2.56 2.57 1.83 3.22

C.O.V. 0.49 0.57 0.56 0.53 0.58

CTV (MR) Mean (mm) 0.89 1.49 1.25 1.86 2.20

S.D. (mm) 0.39 0.93 0.52 1.06 1.43

C.O.V. 0.44 0.63 0.41 0.57 0.65

Mean (mm) 8.31 7.33 8.22 12.81 3.77


23

Trigone S.D. (mm) 8.78 4.53 6.32 8.67 3.49

C.O.V. 1.06 0.62 0.77 0.68 0.93


S.D. (mm) 3.13 1.75 2.66 1.38 3.86

C.O.V. 0.78 0.53 0.64 0.54 0.63


S.D. (mm) 1.44 2.92 1.38 3.19 3.66

C.O.V. 0.74 0.63 0.47 0.73 0.72


S.D. (mm) 5.96 4.41 4.20 3.63 6.25

C.O.V. 0.48 0.57 0.54 0.49 0.55


S.D. (mm) 5.87 5.26 3.98 4.68 4.38

C.O.V. 0.45 0.56 0.52 0.51 0.57


S.D. (mm) 1.71 2.84 2.90 4.47 6.32

C.O.V. 0.54 0.61 0.71 0.71 0.64


S.D. (mm) 2.26 2.91 2.09 5.42 11.03

C.O.V. 0.71 0.72 0.52 0.83 1.06


S.D. (mm) 3.06 2.80 3.35 3.00 4.35

C.O.V. 0.78 0.53 1.14 0.59 0.97


S.D. (mm) 12.44 7.05 8.13 6.43 9.51

C.O.V. 0.77 0.60 0.74 0.69 0.61


24

Table e9: Contouring experience levels of CT structures for all volunteers prior to commencement of this investigation. Volunteers have been stratified as Radiation Oncologists (RO) or non-Radiation Oncologists (Non-RO). Median experience level for each structure is bolded red, with the lower bound bolded if the median lies between two columns.

0 Patients 1 – 10 Patients 11 – 30 Patients 30+ Patients

Prostate/CTV RO 0 0 2 7

Non-RO 3 0 0 1

Seminal Vesicles RO 0 0 3 6

Non-RO 3 0 0 1

Rectum RO 0 0 1 8

Non-RO 2 0 0 2

Distal Colon RO 0 1 2 6

Non-RO 2 0 2 0

Bowel Bag RO 1 2 3 3

Non-RO 2 2 0 0

Bladder RO 0 0 1 8

Non-RO 2 0 0 2

Peri-rectal Space RO 2 4 1 2

Non-RO 3 0 1 0


25

Table e10: Contouring experience levels of CT structures for all volunteers prior to commencement of this investigation. Volunteers have been stratified as Radiation Oncologists (RO) or non-Radiation Oncologists (Non-RO). Median experience level for each structure is bolded red, with the lower bound bolded if the median lies between two columns.

0 Patients 1 – 10 Patients 11 – 30 Patients 30+ Patients

Prostate/CTV RO 0 1 2 6

Non-RO 3 0 0 0

Trigone RO 4 4 1 0

Non-RO 4 0 0 0

Membranous Urethra

RO 0 6 1 2

Non-RO 3 1 0 0

Penile Bulb RO 0 4 4 1

Non-RO 3 1 0 0

Neurovascular Bundles

RO 3 4 2 0

Non-RO 4 0 0 0

Internal Anal Sphincter

RO 3 4 1 0

Non-RO 4 0 0 0

External Anal Sphincter

RO 3 5 1 0

Non-RO 4 0 0 0

Puborectalis Muscle

RO 4 4 1 0

Non-RO 4 0 0 0

Levator Ani Muscles

RO 4 3 2 0

Non-RO 4 0 0 0


26

Table e11: Wilcoxon rank sum testing of contoured volumes between Radiation Oncologists (RO) and non-Radiation Oncologists (Non-RO). Median volumes for each structure for each patient are given in cm3. P-values indicate differences between median volumes, with p-values < 0.05 bolded red. However, Bonferroni corrections must be made due to the large number (85) of statistical tests undertaken. This requires p < 0.0006 for statistical significance, which no comparison reached.

Patient 1 2 3 4 5

CTV (CT) RO (cm3) 53.30 45.89 55.42 34.21 76.10

Non-RO (cm3) 56.91 41.88 50.16 33.53 71.96

p-value 0.762 0.476 0.171 0.914 0.914

Seminal Vesicles RO (cm3) 15.41 14.39 23.38 8.16 18.79

Non-RO (cm3) 15.23 15.33 20.91 10.82 16.66

p-value 1.000 0.352 0.352 0.067 0.610

Rectum RO (cm3) 54.93 36.35 38.05 58.86 55.38

Non-RO (cm3) 47.13 37.41 40.26 40.69 59.30

p-value 0.038 0.914 0.352 0.381 0.610

Distal Colon RO (cm3) 31.80 17.51 16.24 28.48 24.37

Non-RO (cm3) 33.66 15.07 10.83 25.14 25.13

p-value 0.610 0.486 0.352 0.286 0.762

Bowel Bag RO (cm3) 477.40 530.18 309.55 607.21 697.09

Non-RO (cm3) 368.34 433.06 281.41 530.67 635.74

p-value 0.190 0.190 0.476 0.190 0.190

Bladder RO (cm3) 70.18 142.68 400.88 144.23 82.17

Non-RO (cm3) 70.72 152.33 401.30 140.18 88.40

p-value 0.610 0.257 0.914 0.762 0.067

Peri-Rectal Space RO (cm3) 364.79 308.39 315.70 345.72 439.66

Non-RO (cm3) 334.06 293.69 282.50 326.78 401.15

p-value 0.067 0.610 0.762 0.352 0.352

CTV (MR) RO (cm3) 43.98 29.55 43.47 22.64 63.59

Non-RO (cm3) 42.68 29.71 41.97 20.62 58.04

p-value 0.527 0.808 0.808 0.230 0.570

Trigone RO (cm3) 2.72 2.91 2.68 2.29 3.20

Non-RO (cm3) 2.76 3.39 2.94 2.71 3.73

p-value 1.000 0.927 0.788 0.927 0.283

Membranous Urethra RO (cm3) 0.87 0.46 0.80 0.65 0.73

Non-RO (cm3) 0.72 0.50 0.59 0.51 0.83

p-value 0.505 0.830 0.570 0.927 0.808

Penile Bulb RO (cm3) 4.27 5.55 2.84 8.09 3.75

Non-RO (cm3) 4.20 4.22 2.10 7.22 2.98

p-value 0.610 0.315 0.012 0.412 0.073

Neurovascular Bundle (Left) RO (cm3) 0.47 0.55 0.53 0.39 0.59

Non-RO (cm3) 0.29 0.47 0.71 0.31 0.61

p-value 0.927 1.000 0.315 0.762 0.933

Neurovascular Bundle (Right) RO (cm3) 0.53 0.47 0.59 0.35 0.68

Non-RO (cm3) 0.27 0.34 0.58 0.27 0.76

p-value 0.610 0.905 0.927 0.867 0.412

Internal Anal Sphincter RO (cm3) 8.98 7.99 6.92 9.85 5.91

Non-RO (cm3) 9.24 9.99 9.66 13.36 11.95

p-value 0.927 0.164 0.230 0.352 0.283


27

External Anal Sphincter RO (cm3) 6.59 9.72 7.70 12.42 10.22

Non-RO (cm3) 7.82 7.53 9.49 9.60 12.98

p-value 0.315 0.412 0.412 0.762 0.214

Puborectalis Muscle RO (cm3) 16.02 11.89 8.61 6.44 25.00

Non-RO (cm3) 12.61 14.93 9.67 9.54 24.10

p-value 0.109 0.648 0.788 0.230 1.000

Levator Ani Muscles RO (cm3) 17.30 18.20 25.03 28.18 20.30

Non-RO (cm3) 14.78 21.95 31.10 27.54 20.47

p-value 0.927 1.000 0.788 0.927 0.683


Multi-Observer Segmentation Study – Male Pelvis - Organ-specific Instructions

Naming convention, Structure_ID_Initials (e.g., Seminal_Vesc_XX)


4.2 Supplementary Material B

CT CONTOURING CHECKLIST

Prostate

☐ Contoured on CT (‘CTV_CT’)

Seminal Vesicles

☐ Contoured on CT (‘Seminal_Vesc’)

Rectum

☐ Contoured on CT (‘Rectum’)

☐ Most inferior contour level with ischial tuberosities

☐ Most superior contour at sigmoid colon

Distal Colon

☐ Contoured on CT (‘Colon’)

☐ Most superior contour 3 cm above rectum

Bowel Bag

☐ Contoured on CT (‘Bowel_Bag’)

☐ Most inferior contour small/large bowel loop?

☐ If yes, rectum is included.

☐ Most superior contour 3 cm above rectum

☐ Bladder included in Bowel Bag

Bladder

☐ Contoured on CT (‘Bladder’)

☐ Bladder contour includes bladder neck

Peri-Rectal Space

☐ Contoured on CT (‘PRS’)

☐ Most superior contour 2 cm above CTV

☐ Most inferior contour 2 cm below CTV

Chapter

4.Contouring

variabilityofm

alepelvic

structures135

MR CONTOURING CHECKLIST

Prostate

☐ Contoured on MR (‘CTV_MR’)

Trigone

☐ Contoured on MR (‘Trigone’)

Membranous Urethra

☐ Contoured on MR (‘Memb_Urethra’)

☐ Most superior contour starts below prostatic apex

☐ Most inferior contour finishes at penile bulb

Penile Bulb

☐ Contoured on MR (‘Penile_Bulb’)

☐ Located below membranous urethra and urogenital diaphragm

Neurovascular Bundle

☐ Contoured on MR (‘Neuro_Bundle’)

Pelvic Floor Muscles

☐ Contoured on MR

☐ Internal anal sphincter (‘IAS’)

☐ External anal sphincter (‘EAS’)

☐ Puborectalis muscle (‘PRM’)

☐ Levator ani muscles (‘LAM’)

Chapter

4.Contouring

variabilityofm

alepelvic

structures136

CT STRUCTURES

Sagittal Slice Coronal Slice

Bladder

Colon

Rectum

Seminal_Vesc

CTV_CT

Bowel_Bag

Bladder

CTV_CT

Colon

Bowel_Bag

Chapter

4.Contouring

variabilityofm

alepelvic

structures137

STRUCTURE ID DESCRIPTION

Prostate on CT CTV_CT Prostate (CTV) on CT images. No specific instructions. Clinician-interpretation.

Seminal Vesicles Seminal_Vesc Entire seminal vesicles including those slices that also have prostate identified • ~ 4-5 cm or more longer and largely above the prostate, behind the bladder and in front of the rectum

Seminal Vesicles: Axial Slice

Bladder

CTV_CT

Seminal_Vesc

Rectum

Bladder

CTV_CT

Seminal_Vesc

Rectum

Chapter

4.Contouring

variabilityofm

alepelvic

structures138

• Inferiorly the proximal seminal vesicles are directly behind the upper posterior prostate and enter the prostate as the ejaculatory ducts ~ 5-10 mm below the prostatic base (depending on extent of any benign prostatic enlargement

Seminal Vesicles: Sagittal Slice

• The SVs are readily seen on CT above the prostate but are difficult to separate from the posterior prostate at the level of the prostatic base • On MRI the seminal vesicles are light in colour and readily distinguishable from the darker colour of the prostate

Rectum Rectum Based on RTOG guidelines1. Outline on CT axial slices. The external wall of the rectum, inferiorly from the lowest level of the ischial tuberosities (right or left) incorporating the external anal canal, superiorly before the rectum loses its round shape in the axial plane and connects anteriorly with the sigmoid (where the rectum turns horizontally into the sigmoid colon, usually at the caudal border of the sacroiliac joint).

1https://www.rtog.org/CoreLab/ContouringAtlases/MaleRTOGNormalPelvisAtlas.aspx

Chapter

4.Contouring

variabilityofm

alepelvic

structures139

Rectum: Inferior Slice Rectum: Superior Slice

Rectum: Sagittal Slice

First slice ischial tuberosities are visible

Last slice before rectum loses round shape

Chapter

4.Contouring

variabilityofm

alepelvic

structures140

Distal Colon Colon Outline on CT axial slices. Large bowel, from the superior extent of the rectum contour, continuing 3 cm superiorly.

Colon: Inferior Slice Colon: Superior Slice

Colon: Sagittal Slice

Colon

Colon

3 cm

Chapter

4.Contouring

variabilityofm

alepelvic

structures141

Bowel Bag Bowel_Bag The Bowel_Bag is a simpler way of contouring the bowel, and will encompass the small-bowel and colon. Contour the abdominal contents, excluding muscle and bones. Contour every other slice when the contour is not changing rapidly – interpolate and edit as needed. Finally, subtract any overlapping non-GI normal structures. Stop contouring superiorly 3 cm above rectum, and inferiorly at the level of whichever is most inferior of the rectum or the most inferior small or large bowel loop.

Bowel Bag: Inferior Slice Bowel Bag: Superior Slice

Bowel_Bag

Chapter

4.Contouring

variabilityofm

alepelvic

structures142

Bowel Bag: Sagittal Slice

Bowel_Bag and Colon outlines contoured up to same superior slice

Chapter

4.Contouring

variabilityofm

alepelvic

structures143

Bladder Bladder Based on RTOG guidelines. Inferiorly from its base (including the bladder neck) and superiorly to the dome.

Bladder: Sagittal Slice

Bladder

Rectum

CTV_CT

Chapter

4.Contouring

variabilityofm

alepelvic

structures144

Peri-rectal Space PRS The perirectal region of interest is the potential space in which the irradiated portion of the rectum could expand within. For practical purposes i.e. DVH estimations it is therefore the rectal volume plus the peri-rectal fat around it, extending to 2 cm caudal to the CTV and 2 cm cranial to the CTV.

Peri-Rectal Space: Sagittal Slice

PRS

2 cm

CTV_CT

2 cm

Chapter

4.Contouring

variabilityofm

alepelvic

structures145

Anteriorly the perirectal fat space is defined by a coronal plane midway through the prostate. Posteriorly the perirectal fat space is defined by the anterior border of sacrum (i.e. the sacral hollow), the anterior surface of the coccyx and, caudal to this, the anterior borders of the gluteal muscles.

The caudal perpendicular line that defines the caudal extent of the peri-rectal fat space may in some cases be so caudally situated that it passes through the pelvic floor musculature.

Peri-Rectal Space: Axial Slice

Chapter

4.Contouring

variabilityofm

alepelvic

structures146

MR STRUCTURES

Sagittal Slice Axial Slice

CTV_MR

Memb_Urethra

Penile_Bulb

CTV_MR

Neuro_Bundle

Trigone

LAM

PRM

EAS

IAS

LAM

Chapter

4.Contouring

variabilityofm

alepelvic

structures147

STRUCTURE ID DESCRIPTION

Prostate on MR CTV_MR Prostate (CTV) on MR images. No specific instructions. Clinician-interpretation.

Trigone Trigone The area of bladder between the 2 ureteric orifices and the proximal prostatic urethra.

Trigone: Superior slice

Trigone

Trigone

Chapter

4.Contouring

variabilityofm

alepelvic

structures148

Trigone: Mid slice

Trigone: Sagittal Slice

CTV_MR

Trigone

Chapter

4.Contouring

variabilityofm

alepelvic

structures149

Membranous Urethra

Memb_Urethra • Difficult to see on CT but with dark window and high contrast may be seen as circular structure 5-7 mm diameter with a small amount of dark fat around the outside circumference • On T2 MRI is easily seen • Proximally starts immediately below the last slice of the prostatic apex and passes through the sphincter urogenital diaphragm

Membranous Urethra: Superior Slice

Memb_Urethra

Chapter

4.Contouring

variabilityofm

alepelvic

structures150

• Ends after 5-15 mm length as it enters the penile bulb and becomes penile urethra

Membranous Urethra: Inferior Slice

Chapter

4.Contouring

variabilityofm

alepelvic

structures151

Penile Bulb Penile_Bulb • Best visualised on MRI being white on T2 MRI and lying below membranous urethra and urogenital diaphragm • On CT best seen on axial view, and is anterior to anus and usually separated from it by a small rim of black fat • Best visualised with low brightness and dark contrast. In an axial view it is oval in shape • It is approximately 1 cm proximal to distal and is usually just above the bottom of the ischial tuberosities

Penile Bulb: Superior Slice

Penile_Bulb

Chapter

4.Contouring

variabilityofm

alepelvic

structures152

• Anteriorly it blends with the crus and corpora of the penis and on MRI the penile urethra is seen running anteriorly from it • The penile bulb is small proximally and distally blending with the urogenital diaphragm and fat respectively • The distal membranous urethra becomes the bulbar urethra and distally the bulbar urethra becomes the penile urethra • These transitions are not readily seen on CT but readily visible on MRI T2 sequence

Penile Bulb: Inferior Slice

Chapter

4.Contouring

variabilityofm

alepelvic

structures153

Neurovascular Bundle

Neuro_Bundle • May be seen on MRI as a bundle running down postero lateral aspect of both sides of prostate. However sometimes is more spread out anteriorly and posteriorly and not seen

Neurovascular Bundle: Superior Slice

Neuro_Bundle

CTV_MR

Neuro_Bundle

CTV_MR

Chapter

4.Contouring

variabilityofm

alepelvic

structures154

Neurovascular Bundle: Inferior Slice

Neurovascular Bundle: Sagittal View

Neuro_Bundle

CTV_MR

Chapter

4.Contouring

variabilityofm

alepelvic

structures155

Pelvic Floor Muscles IAS

EAS

PRM

LAM

• Four pelvic floor muscles are considered to be involved in normal faecal continence, which can be delineated separately on MRI images:

Pelvic Floor Muscles: Sagittal Slice

PRM

LAM

EAS

IAS

Chapter

4.Contouring

variabilityofm

alepelvic

structures156

Internal anal sphincter (IAS) - the distal continuation of the smooth muscle layer of the rectum and is usually delineated when contouring the anal canal.

IAS: Inferior slice IAS: Superior slice

IAS

Chapter

4.Contouring

variabilityofm

alepelvic

structures157

External anal sphincter (EAS) - partially encircles the IAS. It is separated from the IAS by the intersphincteric space, which can be distinguished on the axial CT-slices as a thin hypodense separation between both anal sphincters.

EAS: Inferior slice EAS: Superior slice

EAS

Chapter

4.Contouring

variabilityofm

alepelvic

structures158

Puborectalis muscle (PRM) - a U-shaped muscle, which forms a sling around the anorectal junction, thereby creating the anorectal angle; it is inserted to the pubic bone.

PRM: Inferior slice PRM: Superior slice

PRM

Chapter

4.Contouring

variabilityofm

alepelvic

structures159

Levator ani muscles (LAM) - Cranially, form a plate-like continuation of the PRM.

LAM: Inferior slice LAM: Superior slice

LAM

Chapter

4.Contouring

variabilityofm

alepelvic

structures160

Detail of genito-urinary structures

Figure 1 mid bulb on MRI

Figure 2 Mid bulb on CT

Proximal penile bulb

Distal penile bulb

Bulb Crus

corpora

Anus

Bulb

Bulb Bulb

Bulb

penile urethra

Bulb

bulbar urethra


CHAPTER 5

Spatial analysis of target volume contours

within three international prostate

radiotherapy trials – Clinical impact of

contouring variability on patient outcome

162

Chapter 5. Clinical impact of contouring variability on patient outcome 163

Spatial analysis of target volume contours withinthree international prostate radiotherapy trials -Clinical impact of contouring variability on patientoutcomeD. Roach1,2,3, J. W. Denham4, D. J. Joseph5,6,7, S. L. Gulliford8, D. P. Dearnaley9, M. R.

Sydes10, E.Hall11, A. Steigler4, J. A. Dowling1,12,13,14, A. Kennedy15, A. Haworth16, M. G.

Jameson1,2,3,12, M. Marcello5,15, P. Greer14,17, L. C. Holloway1,2,3,12,16, M. A. Ebert5,6,12,15

1Faculty of Medicine, University of New South Wales, New South Wales, Australia2Ingham Institute for Applied Medical Research, New South Wales, Australia3Department of Radiation Oncology, Liverpool and Macarthur Cancer Therapy Centres,

New South Wales, Australia4School of Medicine and Public Health, University of Newcastle, New South Wales, Aus-

tralia5University of Western Australia, Western Australia, Australia65D Clinics, Claremont, Western Australia, Australia7GenesisCare, Western Australia, Australia8Radiotherapy Department, University College London Hospitals NHS Foundation Trust,

London, United Kingdom9Academic UroOncology Unit, The Institute of Cancer Research and Royal Marsden Hos-

pital, London, United Kingdom10MRC Clinical Trials Unit, Medical Research Council, United Kingdom11Clinical Trials and Statistics Unit, The Institute of Cancer Research, London, United

Kingdom12Centre for Medical Radiation Physics, University of Wollongong, New South Wales, Aus-

tralia13Australian e-Health Research Centre, CSIRO, Royal Brisbane Hospital, Queensland,

Australia14School of Mathematical and Physical Sciences, University of Newcastle, New South

Wales, Australia15Department of Radiation Oncology, Sir Charles Gairdner Hospital, Western Australia,


Australia16School of Physics, Institute of Medical Physics, University of Sydney, New South Wales,

Australia17Calvary Mater Newcastle Hospital, New South Wales, Australia

5.1 Abstract

5.1.1 Background and Purpose

Non-compliant contouring of target volumes can lead to poorer patient outcome and mask

potential findings within clinical trials. This study retrospectively investigated correlations

between contouring variations of the clinical target volume (CTV) and recorded patient

outcomes within the RADAR, RT01, and CHHiP randomised prostate radiotherapy clin-

ical trials.

5.1.2 Materials and Methods

Prostate-only CTVAtlas were automatically contoured upon the RADAR, RT01, and

CHHiP clinical trial datasets using a previously validated male pelvic atlas. Variations

between the original contour CTVManual and CTVAtlas were evaluated using Distance-

to-Agreement (DtA) vector mappings, which incorporated spatial information during as-

sessment. Correlations between DtA vectors and recorded patient outcome measures were

assessed using Cox Proportional Hazards Modelling.

5.1.3 Results

Significant variations between CTVManual and CTVAtlas were observed at the bladder

and rectum boundaries. Insufficient contouring of CTVManual relative to CTVAtlas in

the vicinity of the bladder significantly correlated with an increased risk of patient death


(median HR = 0.57, 0.36, 0.70 for RADAR, RT01 and CHHiP) and PSA progression (me-

dian HR = 0.37, 0.56 for RT01 and CHHiP). PSA progression correlated with insufficient

contouring of CTVManual around the rectum for RADAR patients (median HR = 0.55).

Increased risk of local progression correlated with insufficient contouring at the anterior-

inferior boundary of CTVManual (median HR = 0.37, 0.16, 0.29 for RADAR, RT01 and

CHHiP).

5.1.4 Conclusion

Incorporation of DtA vector mappings allowed for a comprehensive spatial analysis of

contouring variations. Contouring variability of the CTV near the bladder and rectum

significantly increased the risk of poorer patient outcome following treatment.

5.2 Introduction

Accuracy in contouring remains one of the largest sources of geometric uncertainty in

radiotherapy [24, 228]. Deviations from protocol in target volume contouring have been

shown to reduce overall patient outcome [11, 23, 221, 222, 389], with potential findings

from clinical trials obfuscated [217, 390]. The need for high levels of quality assurance

(QA) policies for contouring during clinical trials have been recently re-emphasised [391],

with benchmarking cases highlighting the efficacy of these processes [392].

Uncertainties associated with contouring in radiotherapy have been thoroughly investi-

gated [393], however a lack of consistency regarding how contouring variations are quan-

tified makes comparisons between studies problematic [25, 26]. It has been shown that

commonly utilised contouring metrics often do not correlate with one another [28]. To

account for this, current recommendations state contouring studies should incorporate

multiple metrics during analysis [25, 27]. However, it has been shown that commonly

utilised metrics correlate poorly with dosimetry [29, 30, 394]. Furthermore, these metrics

provide no spatial information during analysis, failing to identify the specific regions where

contouring variations are most prevalent.


There has been an ongoing desire for automatic contouring methods to be implemented

within clinical practice to reduce treatment planning times [28]. Additionally, utilisation

of automatic contouring allows for a consistent definition of the contour boundary, in-

dependent of uncertainties associated with manual contouring. Not only could clinical

QA procedures assess manually-derived contours against these automatically generated

contours [395, 396], but quality of contouring in clinical trials could be retrospectively

analysed. An assessment of contouring consistency between original and automatically

generated contours could be correlated with known clinical trial endpoints, providing a

link between contouring variations and recorded patient outcomes.

The aim of this manuscript is to provide a comprehensive analysis of the clinical impact

of contouring variations of the clinical target volume (CTV) for intact prostate radiother-

apy. Three large clinical trial datasets were investigated, with recorded patient outcomes

correlated with a spatial analysis of contouring uncertainties.

5.3 Materials and Methods

5.3.1 Clinical Trial Datasets

The Trans-Tasman Radiation Oncology Group (TROG) 03.04 Randomised Androgen De-

privation and Radiotherapy (RADAR) trial investigated the clinical benefit of short- versus

intermediate-term androgen deprivation therapy combined with conformal radiotherapy

for intermediate- and high-risk prostate cancer patients [278, 280]. Comprehensive QA

practices initiated for the RADAR trial included a set-up accuracy study, planning bench-

marking exercise, and periodic auditing of target and normal tissue contours [283, 285].

In total, 1071 patients were recruited between 2003 and 2007, with 814 receiving external

beam radiotherapy.

The Medical Research Council (MRC) RT01 phase 3 clinical trial investigated the efficacy

of dose escalation during conformal radiotherapy for localised prostate cancer [293, 294].

Between 1998 and 2002 843 patients were recruited, and randomly assigned to receive

either standard (64 Gy, 32 fractions), or dose escalated (74 Gy, 37 fractions) treatment.

Chapter

5.Clinicalim

pactofcontouringvariability

onpatientoutcom

e167

RADAR RT01 CHHiP

Clinical Trial DatasetsTrial Location Australia and New Zealand United Kingdom United KingdomAccrual Period 2003 - 2007 1998 - 2001 2002 - 2011Trial Registration ISRCTN90298520 ISRCTN47772397 ISRCTN97182923Ethics approval Hunter New England Human Re-

search Ethics Committee TrialID 03/06/11/3.02

North Thames Multi-centre Re-search Ethics Committee numberMREC/97/2/16

London Multi-centre ResearchEthics Committee number04/MRE02/10

Total Patients 1071 388 253Prostate-only CTVs 513 (47.9%) 82 (21.1%) 247 (97.6%)

Baseline VariablesPrescribed dose 66 Gy: 47 (9.16%) 64 Gy: 34 (41.46%) 57 Gy: 87 (35.22%)

70 Gy: 301 (58.67%) 74 Gy: 48 (58.54%) 60 Gy: 82 (33.20%)74 Gy: 165 (32.16%) 74 Gy: 78 (31.58%)

Disease risk group Gleason score ≤ 7: 374 (72.90%) T1b, T1c, T2a WITH PSA +[Gleason score - 6]*10 < 15: 44(53.66%)

T1b, T1c, T2a WITH PSA ≤10 AND Gleason score ≤ 6: 59(23.89%)

Gleason score > 7: 139 (27.10%) T1b, T1c, T2a WITH PSA +[Gleason score - 6]*10 ≥ 15; orT2b, T3a: 38 (46.34%)

Cancer Stage ≥ T2b, 10 < PSA≤ 20, or Gleason score > 6: 188(76.11%)

Cancer Stage T2: 395 (77.00%) T1, T2a: 60 (73.17%) T1, T2a: 182 (73.68%)T3/T4: 118 (23.00%) T2b, T3a: 22 (26.83%) T2b, T2c, T3a: 65 (26.32%)

Androgen deprivation 6 months: 261 (50.88%) 3 - 6 months: 82 (100.00%) Median 24 weeks: 247 (100.00%)therapy duration 18 months: 252 (49.12%)

Patient OutcomePatient Death 116 (22.6%) 20 (24.4%) 40 (16.2%)Local Progression 30 (5.9%) 12 (14.6%) 24(9.7%)PSA Progression 182 (35.5%) 38 (46.3%) 71 (28.7%)

Table 5.1: Patient numbers, baseline variables, and recorded patient outcome across each clinical trial. Patients were dichotomised basedon disease-risk groups and cancer stages. Baseline variable and patient outcome numbers are with respect to patients analysed.


Finally, the ‘Conventional or Hypofractionated High-dose Intensity Modulated Radiother-

apy in Prostate Cancer’ (CHHiP) trial was a large multicentre randomised phase 3 trial

investigating the non-inferiority of hypofractionated radiotherapy with standard treatment

for localised prostate cancer [308, 312]. 3216 patients were recruited between 2002 and

2011, with patients randomly assigned to either a single conventional radiotherapy arm (74

Gy, 37 fractions), or one of two hypofractionated radiotherapy arms (60 Gy, 20 fractions;

57 Gy, 19 fractions). Analysis in this manuscript utilises an early cohort of 253 CHHiP

participants for whom DICOM plan data was available prior to final CHHiP outcomes

reporting.

Patient numbers, baseline variables, and 5 to 10 year patient outcome data from these

clinical trials are shown in Table 5.1. Only patients where a prostate-only CTV contour

was identified were included during this investigation.

5.3.2 Atlas-Based Contouring

The methodology of developing and validating the male pelvic atlas utilised by this study

has been outlined in previous studies [397, 398]. In brief, 13 observers contoured prostate-

only CTV (with no additional margin) across 5 exemplar patient CT scans, defining the

atlas. Exemplar patients were rigidly and deformably registered to each clinical trial

patient investigated, with atlas CTVs propagated onto each clinical trial patient. Further

details regarding construction of the atlas can be found in Kennedy et al. [397].

Local-weighted voting, following the method of Dowling et al. [345], provided weight-

ings to all atlas CTVs based on localised image similarity between the exemplar patients

and the clinical trial patient. Weighted atlas CTVs were combined to produce a proba-

bility mapping of atlas-defined CTV agreement, whereupon thresholding this probability

mapping at 50% agreement defined CTVAtlas.


Figure 5.1: Schematic of Distance-to-Agreement (DtA) vector construction be-tween CTVAtlas (blue) and CTVManual (red). A vector propagating from thecentre-of-mass of CTVManual intersects CTVManual and CTVAtlas. DtA betweenCTVs along this vector is measured (red arrow), with positive DtA correspondingto CTVManual extending further outwards than CTVAtlas. Vectors were sampledevery 6°, in both azimuthal and elevation, providing a complete spatial mapping

of contouring variation.

5.3.3 Vector Mappings

Three-dimensional vector mappings, similar to the methodology proposed by Remeijer

et al., were utilised to assess contouring variations in this study [253]. A vector prop-

agating radially outwards from the centre-of-mass of the original trial CTV (hereon de-

fined CTVManual) intersects both CTVAtlas and CTVManual contours (Figure 5.1). The

distance-to-agreement (DtA) between CTVs along this vector is always measured from

CTVManual to CTVAtlas. Thus, positive DtA values correspond to CTVManual protruding

further outwards along the radial vector than CTVAtlas (i.e. over-contouring, Figure 5.1),

while negative DtA values correspond to CTVManual lying within the CTVAtlas contour

(i.e. under-contouring).

By defining the centre-of-mass of CTVManual as the origin, all clinical trial CTVs are

rigidly registered to this coordinate system. Consequently, all vectors will be oriented


identically for all patients, allowing for direct comparisons of contouring variations between

patients along each vector. These vectors were then constructed every 6°, in both azimuthal

and elevation coordinates, producing a three-dimensional mapping of contouring variation

between CTVManual and CTVAtlas. Three-dimensional mappings of DtA were projected

into two-dimensions, whereby spatial regions corresponding to neighbouring organs-at-risk

could be identified (Figure 5.2).

5.3.4 Statistical Analysis

Significant differences between CTVManual and CTVAtlas along each vector for each trial

were assessed using Wilcoxon signed rank tests. For each patient outcome listed in Table

5.1, along each vector a DtA cut-point (DCP) was determined that resulted in the greatest

statistical difference in outcome for patients dichotomised at this point. This was calcu-

lated using max-rank sum statistics, computed using the maxstat R package [399]. Hazard

ratios for patients dichotomised by the DCP were calculated for each outcome using Cox

Proportional Hazards modelling. Baseline variables listed in Table 5.1 were included as

potential control variables during analysis, following the method employed during previous

studies investigating these datasets [290].

This process was repeated for every vector, with a unique DCP calculated each time.

Vectors with a hazard ratio (HR) larger than 1 indicated that patients with DtA larger than

DCP had an increased risk of developing the investigated patient outcome. Conversely,

vectors with HR less than 1 indicated that it was patients with DtA less than DCP had

an increased risk of experiencing the recorded outcome.

Associated p-value mappings for hazard ratios were calculated, with p < 0.05 consid-

ered statistically significant. Being an exploratory investigation, unadjusted p-values were

utilised for analysis [400]. To investigate the impact of multiple testing on HR mappings,

Free Step-Down Resampling based on the methodology of Westfall and Young was utilised

to calculate adjusted p-value mappings [17].


Figure 5.2: Median DtA between CTVManual and CTVAtlas for RADAR (top-left), RT01 (top-right), and CHHiP (bottom-left) datasets following vector anal-ysis. Shaded areas within each figure correspond to spatial regions where nostatistically significant difference existed between CTVAtlas and CTVManual (p >0.05 following Wilcoxon Analysis). Positive DtA values (yellow shaded regions)correspond to CTVManual extending further compared to CTVAtlas. NegativeDtA values (blue shaded regions) correspond to the converse. Spatial regions cor-responding to median bladder and rectum position lying within 5mm of CTVAtlas

are shown in the bottom-right figure.

5.4 Results

Following the removal of patients without prostate-only CTV contours, a total of 513, 82,

and 247 patients from the RADAR, RT01, and CHHiP trials respectively were available

for analysis.


Figure 5.3: Hazard Ratio mappings with corresponding significance levels foroverall patient survival across RADAR, RT01, and CHHiP trials.


Figure 5.4: Hazard Ratio mappings with corresponding significance levels forlocal progression across RADAR, RT01, and CHHiP trials.


Figure 5.5: Hazard Ratio mappings with corresponding significance levels forPSA progression across RADAR, RT01, and CHHiP trials.


Median DtA mappings for RADAR, RT01, and CHHiP patients are shown in Figure 5.2.

Regions where DtA between CTVManual and CTVAtlas were not statistically significant are

shaded red. Spatial regions in close proximity of bladder and rectum (defined as median

OAR located within 5mm of CTVAtlas) are shown in the bottom right subfigure in Figure

5.2. CTVManual contours within the RADAR trial extended further inferiorly and posteri-

orly compared to CTVAtlas. Significant differences were observed in the superior-anterior

region, oriented towards the bladder, where CTVManual insufficiently covered CTVAtlas.

This was similarly observed for RT01 patients, however differences between CTVManual

and CTVAtlas were not statistically significant. CTVManual extended significantly further

in the inferior-anterior direction (towards membranous urethra and penile bulb), however

inadequately covered CTVAtlas in the vicinity of the rectum. CHHiP CTVs also displayed

a statistically significant region oriented towards the rectum where CTVManual inade-

quately covered CTVAtlas. However, coverage of CTVManual exceeded CTVAtlas towards

the bladder, as well as inferiorly towards the membranous urethra and penile bulb.

HR mappings for overall patient death across RADAR, RT01 and CHHiP are shown in

Figure 5.3, with associated DCP mappings found in the Supplementary Material (Figure

5.6). Near the bladder large spatial regions with statistically significant hazard ratios were

observed for RADAR (median HR = 0.57 for vectors with p < 0.05), RT01 (median HR

= 0.36), and CHHiP patients (median HR = 0.70). Adjacent to the rectum, correlations

were observed for RT01 (median HR = 0.29), however were less pronounced for RADAR

and CHHiP. Complete summaries of HR for the bladder and rectum spatial regions are

found in Tables 5.2, 5.3, and 5.4.

HR mappings for local progression across all trials are shown in Figure 5.4. No significant

results were observed near the bladder, while significant results adjacent the rectum were

only observed for RADAR (median HR = 0.37). RT01 and CHHiP patients displayed

regions of significance in the anterior-inferior orientation, while for RADAR these regions

were observed inferiorly, but slightly laterally. HRs for all three trials within these regions

were less than 1 (RADAR median HR = 0.37, RT01 median HR = 0.16, CHHiP median

HR = 0.29).

Figure 5.5 displays HR mappings for all three trials with respect to PSA progression, where


results between trials were variable with respect to spatial regions of significance. Within

the RADAR trial the rectum (median HR = 0.55) and anterior-inferior region (median

HR = 0.61) all displayed significant regions with HR less than 1. RT01 patients exhibited

a significant region anterior-superior to the CTV within the bladder (median HR = 0.37),

as well as directly inferior (median HR = 0.37). The anterior-superior region within the

bladder was also statistically significant for CHHiP patients (median HR = 0.56).

Additionally, a spatial region posterior-superior to the CTV (within the bladder) signif-

icantly correlated with PSA progression for RADAR patients, with HR greater than 1

(median HR = 1.51). While HR values for RT01 patients also exceeded 1 in this region,

the results were not statistically significant. CHHiP patients recorded two distinct statis-

tically significant spatial regions within the rectum, alternating between HR greater than

and less than 1 (median HR = 1.80).

5.5 Discussion

Significant relationships between CTV contouring variations and recorded patient out-

come have been demonstrated through the retrospective analysis of multiple clinical trial

datasets within this investigation. While this result may be implied by contouring studies

throughout the literature [25], verification of this relationship with recorded outcomes has

proven to be a more difficult exercise. Cox et al. identified five clinical trial datasets

where retrospective analysis of contour quality was correlated with recorded patient out-

come [391]. Within these studies contour quality was typically assessed as either compliant

or non-compliant. An analysis correlating contouring variations at specific spatial regions

has, to the author’s knowledge, not yet been presented in the literature. This study

therefore not only shows that contouring variations in specific spatial regions correlate

significantly with patient outcome, but also provides a framework for adapting contouring

QA practices to provide the best possible outcomes for patients during treatment.

Across all three trials patients with CTVManual contours with DtA greater than the cut-

point in the anterior-superior orientation displayed a significantly reduced risk of PSA

progression and death (median HR less than 1). Therefore, CTVManual contours with


DtA less than the cut-point in this orientation conversely had a significantly increased

risk of PSA progression and death. This region corresponds to the prostate and bladder

boundary, which can be difficult to distinguish on CT without the addition of a contrast

agent [252, 265]. Often this spatial region of the prostate receives less focus in contouring

guidelines, with recommendations stating that the boundary is easy to identify with the

addition of a contrast agent. Consequently, primary attention in these guidelines is given

towards the interface between prostate and rectum, as well as the prostatic apex [145]. If no

contrast agent is used, contouring in this region will become quite susceptible to potentially

significant variations. This was clearly seen for RADAR trial patients (Figure 5.2), where

median CTVManual was approximately 4 mm less than CTVAtlas in this region. Inadequate

contouring of the CTV in this region would result in an insufficient dose coverage, which

could lead to the poorer patient outcomes observed. A similar trend was observed for RT01

patients, however differences between CTVManual and CTVAtlas were not statistically

significant.

Significant differences in median DtA between CTVManual and CTVAtlas contours in the

posterior-inferior orientation (i.e., near the rectum) were observed for RT01 and CHHiP

trial patients. During prostate radiotherapy considerable effort is made to ensure large

doses are not delivered to the rectum to reduce potential rectal toxicities developing. This

could result in a conservative approach when contouring the CTV, additionally compro-

mising the dose coverage in this region. Interestingly, neither trial observed significant

correlations between contouring variation and patient outcome in this region. It should

be noted that median DtA between CTVManual and CTVAtlas for these trials was approx-

imately 1 mm. Additional margins applied to the manual CTVs would therefore generate

planning target volumes that would likely account for any potential insufficiency of the

CTVManual contour in this region.

Meanwhile, despite the median RADAR CTVManual exceeding the corresponding median

CTVAtlas in the vicinity of the rectum, hazard ratios for both local and PSA progression

were found to be statistically significant. While a hazard ratio less than unity corresponded

to an increased risk of progression for insufficiently contoured RADAR CTVManual con-

tours in this region, it is important to consider the DCP value utilised when calculating

these correlations (Figures 5.7 and 5.8). DCP of approximately 6 mm were calculated for


outcomes in this region for RADAR patients. These results may therefore be interpreted

as excessive contouring of CTVManual in this region significantly reducing local and PSA

progression, as opposed to inferior contouring resulting in poorer patient outcome.

Contours with DtA below the cut-point in the anterior-inferior orientation were found to

significantly correlate with an increased risk of local progression across all three trials.

Anatomically, this is where the urethra leaves the prostate as the membranous urethra

and passes through to the penile bulb. Interestingly, cut-points for local progression were

also positive in this region (Figure 5.7), although they were less than the median DtA for

each trial.

A consideration when undertaking this investigation is the interplay between CTV con-

touring variations and delivered dose distributions. This is particularly important when

the treatment delivery varied between trials. Conformal radiotherapy was utilised for

both RADAR and RT01 trials, while CHHiP patients received intensity-modulated ra-

diotherapy (IMRT) for treatment. IMRT treatment plans produce steep dose gradients

surrounding the target volume when compared to conformal radiotherapy, so the impact

of contouring variations could result in more substantial dose discrepancies. To ensure

adequate dose coverage of the CTV, particularly in the vicinity of the bladder, treatment

margins for CTV were modified in this region for the CHHiP clinical trial when compared

to RT01 [401].

While this study focussed on assessments of contouring variability for prostate-only CTV,

many patient datasets included an additional CTV that incorporated prostate and seminal

vesicles. The resulting dose distributions for these patients would obfuscate results in

the spatial region bordering the seminal vesicles, which was reflected in no statistically

significant correlations being observed in this region within this study. However, it should

be stressed that the remaining contour boundary would be invariant to the inclusion or

exclusion of the seminal vesicles when defining the CTV. Consequently, the results from

this study would not be impacted by the inclusion or exclusion of seminal vesicles during

treatment.

The multiple testing problem is a major concern for investigations with numerous measure-

ments, whereby the large number of tests performed resulting in a significant increase in


the likelihood of observing false positive results. The noise within the p-value mappings of

Figures 5.3, 5.4, and 5.5 could be interpreted as such, with statistically significant vectors

appearing at random throughout the mappings. Accounting for multiple testing in this

study is difficult, as the nature of vector construction means that tests between vectors are

not truly independent. Adjusted p-value mappings following the bootstrapping method of

Westfall and Young [17] are provided in the Supplementary Material for reference, how-

ever unadjusted p-value mappings were utilised during analysis. The authors deemed this

appropriate due to the highly conservative nature of the bootstrapping method, with many

significant results potentially obfuscated, as well as due to the exploratory nature of the

investigation [400].

To account for the potential of false positives, only statistically significant results in the

vicinity of neighbouring organs were reported in this study. Results could therefore be

explained by the uncertainty in contouring in the vicinity of these structures. While

significant results were not consistent between trials, it should be noted that the statistical

power of each trial differed substantially due to number of patients analysed (Table 5.1).

Across the three clinical trials data related to the clinical endpoints of overall survival,

local progression, and progression free survival were captured. The consistent recording

of data allowed for a unique investigation where direct comparisons between contouring

variations and outcome could be made, irrespective of the clinical pathways associated

with each endpoint. As this was an exploratory investigation, the primary purpose of

the study was to assess whether correlations could be observed for any endpoint. Future

studies could build upon the methodology developed in this study, overcoming the multiple

testing problem by investigating only a single clinically relevant endpoint such as overall

survival.

Another point of contention is the validity of the CTVAtlas contours. It had previously

been shown that the utilised male pelvic atlas displayed excellent localisation when defin-

ing the CTV, while having accuracy comparable to inter-observer variability [397, 398].

Utilisation of an automated approach allows for a reproducible method for contouring the

large number of patients across all trials. Consequently, while this trial lacked a true gold

standard, it did incorporate a consistently defined CTV across all patients from which DCP


mappings could be derived. Future investigations could utilise deep learning pipelines to

automatically contour the CTV, a process that has become increasingly well established

within the literature [331, 335–337].

Finally, while an incorrectly defined CTVAtlas could alter the distribution of DCP for each

vector, a more probable outcome would be a systematic shift in DCP. For example, if

a correctly defined CTVAtlas resulted in a DCP of 2 mm along a single vector, then an

erroneously defined CTVAtlas shifted by 1 mm would produce a new DCP of 3 mm. While

this would alter the values within the DCP mappings, the Hazard Ratio and associated p-

value mappings would remain invariant of these. For this reason utilisation of an atlas for

automatic contouring of the CTV on all datasets can be justified over the use of more recent

methods such as deep learning. Future work could incorporate calculated DCP across all

patient outcome measures in conjunction with either an atlas or deep learning contour

to redefine define the CTV such that pathological information, as well as anatomical, is

incorporated when defining the contour boundary.

5.6 Conclusions

Retrospective analysis of contouring variations between CTVManual and CTVAtlas were

correlated with patient outcome across the RADAR, RT01, and CHHiP clinical trial

datasets. Incorporation of a vector mapping allowed for a spatial analysis of contouring

variations, where it was found that variability in the vicinity of the bladder, membranous

urethra, and rectum resulted in an increased risk of patient death, PSA progression, and

local progression. DtA cut-point mappings derived from patient outcomes could be used to

refine atlas-defined contours, providing a method for defining the CTV contour boundary

using both pathological and anatomical information.

5.7 Acknowledgements

We acknowledge funding from the Australian National Health and Medical Research Coun-

cil (grants 300705, 455521, 1006447, 1077788), the Hunter Medical Research Institute, the


Health Research Council (New Zealand), the University of Newcastle, the Calvary Mater

Newcastle, Abbott Laboratories and Novartis Pharmaceuticals. We gratefully acknowl-

edge the support of the Trans-Tasman Radiation Oncology Group, the Medical Research

Council Clinical Trials Unit at University College London, and the Institute of Cancer

Research UK. We also acknowledge NHS funding to the National Institute for Health Re-

search Biomedical Research Centre at the Royal Marsden NHS Foundation Trust and The

Institute of Cancer Research, London.

Chapter

5.Clinicalim


onpatientoutcom

e182


Figure 5.6: DtA Cut-points (DCP) across RADAR, RT01, and CHHiP patients, based on recorded patient death. Each vectorcut-point was calculated using max-rank sum statistics, whereby patients dichotomised by DCP result in the most significant

difference in patient outcome calculated by Wilcoxon rank sum test.

Chapter

5.Clinicalim


onpatientoutcom

e183

Figure 5.7: DtA Cut-points (DCP) across RADAR, RT01, and CHHiP patients, based on local progression.

Chapter

5.Clinicalim


onpatientoutcom

e184

Figure 5.8: DtA Cut-points (DCP) across RADAR, RT01, and CHHiP patients, based on PSA progression.

Chapter

5.Clinicalim


onpatientoutcom

e185

Figure 5.9: Adjusted p-value mappings for recorded patient death across RADAR, RT01, and CHHiP patients. p-values wereadjusted using the Free Step-Down Resampling method as described by Westfall and Young (Algorithm 2.8, pages 66 – 67)

[17].

Chapter

5.Clinicalim


onpatientoutcom

e186

Figure 5.10: Adjusted p-value mappings for local progression across RADAR, RT01, and CHHiP patients.

Chapter

5.Clinicalim


onpatientoutcom

e187

Figure 5.11: Adjusted p-value mappings for PSA progression across RADAR, RT01, and CHHiP patients.

Chapter

5.Clinicalim


onpatientoutcom

e188

Patient DeathRADAR RT01 CHHiP

Bladder Rectum Bladder Rectum Bladder Rectum

ALL p<0.05 ALL p<0.05 ALL p<0.05 ALL p<0.05 ALL p<0.05 ALL p<0.05

Mean 0.6269 0.5351 0.8558 0.8958 > 100 0.4175 > 100 0.4331 0.5537 0.3689 0.8949 1.1748

min 0.3306 0.3306 0.4327 0.4327 0 0.103 0 0.1693 0 0.1604 0.2781 0.2781

25% 0.4974 0.4731 0.6353 0.5993 0.2627 0.2047 0.3368 0.2486 0.3855 0.2987 0.5741 0.3803

Med. 0.5679 0.5284 0.7053 0.6376 0.3627 0.2491 0.4690 0.2923 0.5476 0.3923 0.6957 0.4370

75% 0.6470 0.5736 0.8220 1.5092 1.806 0.3051 1.0193 0.3423 0.7121 0.4392 0.9306 2.3022

Max. 1.6796 1.6796 2.0210 2.0210 > 100 4.8217 > 100 4.4466 3.0477 0.5092 2.7329 2.7329

Count 588 382 269 58 588 226 269 77 588 151 269 33

Table 5.2: Hazard Ratios for patient death within bladder and rectum spatial regions for RADAR, RT01, and CHHiP patients.

Chapter

5.Clinicalim


onpatientoutcom

e189

Local ProgressionRADAR RT01 CHHiP



Mean > 100 2.6031 0.5734 0.3312 > 100 1.3734 > 100 0.2199 > 100 2.2424 > 100 2.661

min 0.1963 0.1963 0 0.0987 0 0.0691 0 0.107 0 0.353 0 0.2689

25% 1.7017 2.3273 0.2847 0.2454 0.2880 0.1568 0.2376 0.1839 0.5729 0.3952 1.4566 2.5433

Med. 2.0227 2.4891 0.4142 0.3715 0.4505 0.1822 0.3146 0.2226 0.7534 2.7452 2.0544 2.9757

75% 2.5508 2.7237 0.5141 0.4126 0.9479 0.2176 0.4816 0.2498 2.0172 3.7714 3.0719 3.5602

Max. > 100 4.3482 4.4418 0.4689 > 100 20.3086 > 100 0.3096 > 100 5.2741 > 100 5.0952

Count 588 55 269 71 588 37 269 82 588 32 269 38

Table 5.3: Hazard Ratios for local progression within bladder and rectum spatial regions for RADAR, RT01, and CHHiP patients.

Chapter

5.Clinicalim


onpatientoutcom

e190

PSA ProgressionRADAR RT01 CHHiP



Mean 1.2306 1.4069 0.7997 0.6048 > 100 0.6873 > 100 1.0677 0.7117 0.5673 1.3710 1.5338

min 0.4712 0.4712 0.3643 0.3643 0.1006 0.1006 0.2164 0.2952 0.0897 0.0897 0.3030 0.3030

25% 1.1310 1.3952 0.6034 0.4810 0.4552 0.3262 0.5068 0.3954 0.5847 0.4834 0.6250 0.4992

Med. 1.2404 1.5056 0.7354 0.5451 1.1121 0.3707 1.2113 0.4331 0.6855 0.5590 1.2734 1.8015

75% 1.4004 1.6029 0.8436 0.6610 1.6829 0.4140 2.2886 0.4726 0.7821 0.5879 1.9277 2.1599

Max. 2.1492 2.1492 1.5462 1.4697 > 100 9.1175 > 100 7.4682 3.5215 3.5215 4.2678 4.2678

Count 588 138 269 113 588 134 269 48 588 107 269 83

Table 5.4: Hazard Ratios for PSA progression within bladder and rectum spatial regions for RADAR, RT01, and CHHiP patients.

CHAPTER 6

Adapting automated treatment planning

configurations across international

centres for prostate radiotherapy

191

Contents lists available at ScienceDirect

Physics and Imaging in Radiation Oncology

journal homepage: www.elsevier.com/locate/phro

Original Research Article

Adapting automated treatment planning configurations across internationalcentres for prostate radiotherapyDale Roacha,b,⁎, Geert Wortelc, Cesar Ochoad, Henrik R. Jensene, Eugene Damenc, Philip Vialb,d,f,Tomas Janssenc, Christian Rønn Hansene,f,ga Faculty of Medicine, University of New South Wales, Sydney, Australiab Ingham Institute for Applied Medical Research, Sydney, Australiac Department of Radiation Oncology, The Netherlands Cancer Institute, Amsterdam, The NetherlandsdDepartment of Radiation Oncology, Liverpool and Macarthur Cancer Therapy Centres, NSW, Australiae Laboratory of Radiation Physics, Odense University Hospital, Odense, Denmarkf Institute of Medical Physics, School of Physics, University of Sydney, Sydney, Australiag Institute of Clinical Research, University of Southern Denmark, Denmark

A R T I C L E I N F O

Keywords:ProstateAutomaticVMATTreatment planningPinnacleMulti-centre

A B S T R A C T

Background and purpose: Automated configurations are increasingly utilised for radiotherapy treatment plan-ning. This study investigates whether automated treatment planning configurations are adaptable across clinicswith different treatment planning protocols for prostate radiotherapy.Material and methods: The study comprised three participating centres, each with pre-existing locally developedprostate AutoPlanning configurations using the Pinnacle3® treatment planning system. Using a three-patienttraining dataset circulated from each centre, centres modified local prostate configurations to generate protocolcompliant treatment plans for the other two centres. Each centre applied modified configurations on validationdatasets distributed from each centre (10 patients from 3 centres). Plan quality was assessed through DVHanalysis and protocol compliance.Results: All treatment plans were clinically acceptable, based off relevant treatment protocol. Automatedplanning configurations from Centre’s A and B recorded 2 and 18 constraint and high priority deviations re-spectively. Centre C configurations recorded no high priority deviations. Centre A configurations producedtreatment plans with superior dose conformity across all patient PTVs (mean=1.14) compared with Centre’s Band C (mean= 1.24 and 1.22). Dose homogeneity was consistent between all centre’s configurations(mean=0.083, 0.077, and 0.083 respectively).Conclusions: This study demonstrates that automated treatment planning configurations can be shared andimplemented across multiple centres with simple adaptations to local protocols.

1. Introduction

Conventional treatment planning for Intensity ModulatedRadiotherapy (IMRT) and Volumetric Modulated Arc Therapy (VMAT)require many manual processes, with treatment planners iterativelyadjusting optimisation goals within a treatment planning system todevelop clinically acceptable treatment plans. This process is not onlytime-consuming, but treatment plan quality is inherently dependent onthe individual skill of the planner [1–5]. The importance of high-qualitytreatment planning on clinical outcomes has been demonstrated duringclinical trials [6–8].Both automated treatment planning methods and knowledge-based

optimisation engines are now available within most commercial treat-ment planning systems [9–11], and have demonstrated improvementsin planning efficiency and plan quality compared to current practice[12,13]. Multiple institutions have investigated the efficacy of auto-mated treatment planning for head and neck [14–19], oesophageal[20], and prostate cancers [21–24], and found the automatically gen-erated plan to be noninferior and often superior to manual planningquality, while significantly reducing treatment planning times. Ad-ditionally, previously complex and time-consuming stereotactic treat-ments for liver cancer have had automatic treatment plans developed[25], while quantitative tools have been constructed to automaticallyidentify poorer quality treatment plans [26,27].

https://doi.org/10.1016/j.phro.2019.04.007Received 5 October 2018; Received in revised form 10 April 2019; Accepted 14 April 2019

⁎ Corresponding author at: Ingham Institute, Liverpool Hospital, Locked Bag 7103, Liverpool, NSW 1871, Australia.E-mail address: [email protected] (D. Roach).

Physics and Imaging in Radiation Oncology 10 (2019) 7–13

2405-6316/ © 2019 The Authors. Published by Elsevier B.V. on behalf of European Society of Radiotherapy & Oncology. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/).

T

Chapter 6. Adapting automated treatment planning configurations 192

Previous automated treatment planning studies have been con-ducted either within a single institution, or during clinical trials using asingle protocol across multiple centres [28,29]. As most radiotherapypatients are not treated within a clinical trial, they are planned ac-cording to a wide variety of local protocols and practices developed atdifferent centres. Consequently, comparisons in treatment plan qualitybetween developed automated techniques difficult, with no commonbaseline for plan comparison between centres.This study investigated whether locally developed automatic treat-

ment planning configurations for prostate radiotherapy could beadapted to meet distributed protocols shared amongst multiple centres.It provides an explicit example of how protocol sharing between centrescould proceed, allowing smaller clinics to benefit from the work oflarger centres during future clinical trials.

2. Materials and methods

2.1. AP Configuration development

The three participating centres (two European, one Australian, de-nominated A, B, and C) had previously developed automated treatmentplanning techniques utilising the AutoPlanning (AP) module withinPinnacle3® treatment planning system (Philips Radiation Oncology,Madison, WI). All centres distributed local protocol details on pre-scription, contouring and dose constraints for prostate radiotherapy(Tables 1, A.1). Evaluation criteria were specified by the host-centre aseither low, medium, high, or constraint priority. Additionally, threepatient datasets (one easy, one medium, and one difficult to plan) wereprovided by each centre as training datasets. All distributed datasetscontained target volumes and organs-at-risk previously contoured byeach centre.Each centre modified their locally developed clinical AP config-

urations using the training datasets to meet other centre’s distributedprotocols. Only dose objectives and AP optimisation parameters were tobe adjusted during configuration development (i.e. beam configuration,energy, etc. were to remain as used clinically, see Tables A.2(a–c)). Itshould be noted that all centres’ AP configurations incorporated similarVMAT treatment techniques, although the number of arcs utilised(single or dual) varied.Consequently, each centre generated three distinct VMAT AP con-

figurations to meet the dose objectives for each of the three protocols.Naming convention is shown in Fig. 1, with the protocol name corre-sponding to the origin of the patients and planning objectives.

Conversely, the naming of the AP configuration corresponds to thecentre the AP technique was developed.

2.2. Treatment planning

An additional ten prostate patient datasets with target volume andorgan-at-risk contours were distributed by each centre as validationdatasets, with each patient previously planned using the host-centre’slocal AP configuration. Treatment planning system setups and patientselection criteria are found in Tables A.3 and A.4 respectively. Centresapplied modified AP configurations to the corresponding centre’s vali-dation datasets, resulting in three AP plans each for the thirty patients.Post-optimisation following AP was allowed, based on each centre’sstandard clinical practice. All treatment plans were exported as DICOMRT files and uploaded to a single host for analysis.

2.3. AP configuration evaluation and statistics

Quantitative analyses between AP configurations were performedby dose-volume histogram (DVH) analysis, and total protocol devia-tions. Population median DVH for planning target volumes (PTVs) andOARs across all ten patients per centre were generated, utilising pre-viously developed software [16,20]. Conformity and homogeneity in-dices (CI=V95%/VPTV, HI= (D2% – D98%)/DPrescription) were calculatedfor all patient treatment plans. Two-sided Wilcoxon matched-pair signrank probability curves, previously described by Bertelsen et al. [30],were used to illustrate differences between AP configuration DVH dis-tributions. It should be noted that while individual p-values are not asolid statistical test of differences at a given dose level, given theseparameters are highly correlated, they can be used to visualize wheredifferences between the population median DVHs exist.

3. Results

All centres successfully developed AP configurations that met eachcentre’s protocols utilising the validation datasets (Table A.2(a–c),supplementary material). Mean and standard deviations of target vo-lume (Table 2), as well as constraint, high priority (Table 2), andmedium and low priority OAR evaluations (Table A.5, supplementarymaterial) were recorded across all AP configurations.

Table 1Target and high/constraint priority OAR prescriptions and evaluations. Complete lists of evaluation criteria can be found within Table A.1 (supplementary material).

Protocol A Protocol B Protocol C

Target Prescription Target Prescription Target PrescriptionPTV1 35×2Gy PTV1 39×2Gy PTV1 39×2GyPTV2 35×2.2 Gy

Target Evaluation Priority Target Evaluation Priority Target Evaluation Priority

PTV1 V95% > 99% Constraint PTV1 D95% > 100% Medium PTV1imrt D99% > 95% MediumPTV2 V95% > 99% Constraint PTV1 D2% < 105% Medium PTV1imrt D0% < 107% MediumPTV1/PTV2 D1% < 107% Constraint CTV1 D99% > 100% High PTV1imrt D95% > 95% HighPTV1/PTV2 D1% < 105% Medium PTV1 Dmax < 107% Medium–Soft PTV1 D99.9% > 66.5 Gy High

PTV_prostate_imrt D99% > 95% MediumPTV_prostate_imrt D97% > 95% High

OAR Evaluation Priority OAR Evaluation Priority OAR Evaluation Priority

Rectal Wall V75Gy < 10% High Rectum V65Gy < 20% Constraint Rectum V74Gy < 1 cm3 ConstraintRectal Wall V64Gy < 35% High Rectum V70Gy < 10% Constraint Circumference of

rectumDmax < 50 Gy Constraint

Anal Sphincter Dmean < 45 Gy High Rectum V75Gy < 5% High90% − 95% isodose tight around

targetHigh RECT_IN_PTV V79Gy < 1 cm3 High

RECT_78_03_NS V80Gy < 1 cm3 Constraint

D. Roach, et al. Physics and Imaging in Radiation Oncology 10 (2019) 7–13

8


3.1. Protocol A

AP Configurations B and C for protocol A patients recorded superiordose coverage across PTV1 and PTV2, shown by an elevated shoulderregion within the DVH curve (Fig. 2). The PTV coverage was compen-sated by superior rectal wall and anal sphincter sparing by the host APconfiguration (Table 2). A large difference between AP ConfigurationsA and C is evident in the rectum DVH curves between 10 and 60 Gy(Fig. 3), with V30Gy varying by over 20% (30.4% vs. 52.4%). Bothmodified AP configurations produced superior femoral head coverage(Table A.5, Figs. A.1, A.2).

3.2. Protocol B

AP Configurations A and C for protocol B recorded statisticallysignificant improvements in comparison to the original treatment plansfor rectum V70Gy (8.5%, 8.1% vs. 11.0%) and V75Gy (3.6%, 3.5% vs.7.6%). Conversely, rectum V40Gy and V50Gy were superior for the hostAP configuration (Fig. 3). Dose coverage across PTV1 and CTV1 weresuperior for the host configuration treatment plans, however differ-ences were generally not statistically significant. AP Configurations Aand C produced superior femoral head coverage, although AP Config-uration A produced inferior bladder V50Gy and V60Gy (Table A.5, Figs.A.1–A.3).

Fig. 1. Schematic of the study design. Participating centres had a local prostate radiotherapy protocol and associated AP configuration. Each centre created new APconfigurations for the other two protocols through modification of their local AP configuration, based on a training dataset of three patients provided by each centre(pre-contoured CT datasets).

Table 2Target and high/constraint priority OAR mean and standard deviations (S.D.) for all protocols. Conformity and homogeneity indices are shown for all PTVs.Differences in metrics considered significant (p < 0.05) are bolded. Complete list of objectives can be found within Table A.4 (supplementary material).

Centre A Centre B Centre C

Mean S.D. p-value Mean S.D. p-value Mean S.D. p-value

Protocol A/Configuration A (Host) Protocol A/Configuration B Protocol A/Configuration CPTV1: V95% > 99% 99.9% 0.1% 100.0% 0.1% 0.08 100.0% < 0.1% 0.002PTV2: V95% > 99% 99.7% 0.2% 99.9% 0.2% 0.04 100.0% < 0.1% 0.002PTV2: D1% < 107% 103.1% 0.5% 102.6% 0.6% 0.03 102.4% 0.2% 0.010Rectal Wall: V75Gy < 10% 4.2% 2.2% 5.1% 3.0% 0.11 6.6% 2.6% 0.002Rectal Wall: V64Gy < 35% 14.8% 5.8% 16.2% 5.3% 0.002 17.2% 6.0% 0.002Anal Sphincter: Dmean < 45Gy 9.1 Gy 7.7 Gy 10.4 Gy 8.7 Gy 0.004 10.0 Gy 9.4 Gy 0.23

PTV1 Conformity Index 1.19 0.04 1.33 0.06 0.002 1.31 0.06 0.002PTV2 Conformity Index 1.12 0.05 1.29 0.10 0.002 1.29 0.08 0.002PTV1 Homogeneity Index 0.13 0.01 0.10 0.01 0.002 0.09 0.01 0.002PTV2 Homogeneity Index 0.07 < 0.01 0.05 0.01 0.002 0.05 < 0.01 0.002

Protocol B/Configuration A Protocol B/Configuration B (Host) Protocol B/Configuration C

PTV1: D95% > 100% 99.2% 1.4% 0.23 99.6% 0.9% 99.2% 1.6% 0.50PTV1: D2% < 105% 104.0% 0.2% 0.69 104.0% 0.3% 104.4% 0.3% 0.02CTV1: D99% > 100% 101.0% 0.7% 0.83 101.0% 0.5% 101.3% 0.7% 0.32PTV1: Dmax < 107% 105.2% 0.3% 0.13 105.5% 0.7% 105.7% 0.3% 0.70Rectum: V65Gy < 20% 12.6% 2.8% 0.11 13.6% 4.0% 11.7% 3.3% 0.002Rectum: V70Gy < 10% 8.5% 2.1% 0.004 11.0% 3.6% 8.1% 2.4% 0.002Rectum: V75Gy < 5% 3.6% 1.2% 0.002 7.6% 2.7% 3.5% 1.2% 0.002RECT_IN_PTV: V79Gy < 1 cm3 0.1 cm3 0.1 cm3 0.004 0.5 cm3 0.4 cm3 0.1 cm3 0.1 cm3 0.006RECT_78_03_NS: V80Gy < 1 cm3 0.3 cm3 0.2 cm3 0.49 0.3 cm3 0.3 cm3 0.5 cm3 0.3 cm3 0.13

PTV1 Conformity Index 1.32 0.07 0.63 1.33 0.08 1.27 0.05 0.04PTV1 Homogeneity Index 0.07 0.02 0.006 0.06 0.01 0.08 0.03 0.006

Protocol C/Configuration A Protocol C/Configuration B Protocol C/Configuration C (Host)

PTV1imrt: D99% > 95% 93.9% 0.7% 0.006 94.9% 0.9% 0.85 95.0% 0.6%PTV1imrt: D0% < 107% 103.7% 0.5% 0.002 106.1% 0.4% 0.05 105.5% 0.8%PTV1imrt: D95% > 95% 96.2% 0.3% 0.002 97.5% 0.6% 0.04 97.1% 0.4%PTV1: D99.9% > 66.5 Gy 69.1 Gy 0.5 Gy 0.002 69.9 Gy 1.0 Gy 0.004 67.9 Gy 0.4 GyPTV_prostate_imrt: D99% > 95% 94.0% 0.9% 0.02 95.0% 0.9% 1.000 94.9% 0.6%PTV_prostate_imrt: D97% > 95% 95.6% 0.4% 0.004 96.7% 0.7% 0.16 96.3% 0.5%Rectum: V74Gy < 1 cm3 0.5 cm3 0.2 cm3 0.16 0.8 cm3 0.3 cm3 0.38 0.7 cm3 0.3 cm3

PTV1 Conformity Index 0.99 0.02 0.002 1.11 0.02 0.70 1.11 0.03PTV1 Homogeneity Index 0.11 0.01 0.002 0.12 0.01 0.70 0.12 0.01


9


3.3. Protocol C

AP Configuration A modified for protocol C recorded significantlypoorer D95% and D99% for PTV1imrt (96.2%, 93.9%), and D97% andD99% for PTV_prostate_imrt (95.6%, 94.0%) compared to the hostconfiguration treatment plans (97.1%, 95.0%, 96.3%, 94.9% respec-tively). Rectum V70Gy, V65Gy, V60Gy, and V50Gy were consistently hotterfor AP Configuration A (Table A.5). AP Configuration B plans producedsuperior dose coverage for PTV1imrt (D95%=97.5%) and PTV1(D99.9%= 69.9 Gy) compared to the host treatment plans(D95%= 97.1%, D99.9%=67.9 Gy), although produced hotter PTVimrt(D0%=106.09%, 105.51% respectively). High rectal dose sparing wassuperior for the host configuration compared to modified AP Config-uration B (V70Gy= 7.19%, 9.5% respectively), but poorer V50Gy (TableA.5, Fig. 3). Femoral Head maximum doses were significantly higher forthe modified AP Configurations A and B (Table A.5, Figs. A.1, A.2),although no significant differences between AP configurations wasobserved for Bladder V50Gy, V60Gy, or V70Gy.Protocol A and C patients planned with AP Configuration A dis-

played significantly increased conformity compared to APConfigurations B and C, although for Protocol A this was at the cost ofreduced homogeneity (Table 2). Homogeneity and conformity for APConfigurations B and C were consistent across protocols A and C.

Protocol B patients displayed the least conformal plans across all APconfigurations, with the significant improvements in the modified APConfiguration C plans resulting in reduced homogeneity.Total constraint/high priority deviations recorded by AP

Configurations A, B, and C were 2, 18, and 0 respectively (Table 3). Allconstraint and high priority deviations occurred for Protocol B patients.Additionally, AP Configurations B and C recorded no deviations forProtocol A patients, compared to 5 medium priority deviations re-corded by the host centre. AP Configuration A recorded a significantlylarger number of medium priority deviations (35, 15, and 17 respec-tively), while low priority deviation numbers were consistent betweenall AP configurations. Specific deviations are given in Table A.6 (sup-plementary material).

4. Discussion

With only a three-patient training dataset utilised in this study, eachcentre was successfully able to modify clinical AP configurations tomeet distributed prostate radiotherapy protocols. By ensuring thetraining datasets contained a spread with regards to planning difficulty,the modified AP configurations were able to account for the expectedanatomical variety that was encountered in the validation datasets. Theability for developed AP configurations to be modified using a small

Fig. 2. Median PTV DVHs for all patients for AP Configurations A (blue), B (red), and C (green). Note that the scale begins at 60 Gy for clarity. Solid lines correspondto host-centre AP configuration, modified configurations are dashed. Solid and dotted grey p-value curves, indicating significant differences between populationmedian DVHs, are also illustrated. The dashed black line shows p=0.05. (For interpretation of the references to colour in this figure legend, the reader is referred tothe web version of this article.)


10


training dataset has major potential advantages in improving qualityassurance during clinical trials, where poor quality manual treatmentplanning quality can significantly impact clinical trial efficacy [3,6].While this improvement is inherently dependent on centres possessingthe same treatment planning system, it still represents an improvementin reducing treatment plan variation in these trials.Across all patients, only a single constraint and high priority de-

viation were recorded by the modified AP configurations (Table 3).Both deviations occurred for a single patient from centre B, with thispatient exhibiting a large overlap (12.5%) between PTV and rectum. Asthe original clinically accepted treatment plan for this patient also re-corded identical protocol deviations, all treatment plans developedfrom modified AP configurations were considered clinically acceptable.

Treatment plans developed by AP Configuration B delivered higherdoses to PTV than those developed by AP Configurations A and C forProtocol C (Table 2, Fig. 2). This contributed in hotter doses beingdelivered to the rectum (Fig. 3), although differences were often mar-ginal and not significant. Significant differences were seen between APconfigurations for intermediate rectum doses, with host-centre treat-ment plans delivering reduced dose at these volumes for all protocols,except for Protocol C, where AP Configuration B produced lower doses.This is reflective of the different emphasis centres placed on meetingconstraint and high priority deviations, with less importance placed onmoderate and low priority deviations. Similarly, Figs. A.1 and A.2 showsignificantly reduced dose to the femoral heads across all protocols byAP Configuration C. Table A.2 shows that AP Configuration C was the

Fig. 3. Median Rectum DVHs for all patients for AP Configurations A (blue), B (red), and C (green). Solid and dotted grey p-value curves again indicate significantdifferences between population median DVHs. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of thisarticle.)

Table 3Total deviations for each AP configuration. Specific deviations can be found in Table A.5 (supplementary material).

AP Configuration A AP Configuration B AP Configuration C

Protocol A Protocol B Protocol C Protocol A Protocol B Protocol C Protocol A Protocol B Protocol C

Constraint 1 8High 1 10Medium 5 7 23 4 11 4 13Low 1 11 13 11


11


only AP configuration that included the femoral heads as objectives.Multiple constraint and high priority deviations were recorded for

protocol B by the host AP configuration. Further inspection revealed adiscrepancy between the protocol requirements submitted by centre Bfor the study, and what was utilised clinically. ln particular, rectumV65Gy, V70Gy, and V75Gy were assigned higher priorities for this studythan applied clinically. To account for this, the host AP configurationfor protocol B was modified for further analysis to bring it in line withthe distributed protocol. The modified AP configuration is given inTable A.7, with new DVH parameters shown in Table A.8 (supple-mentary material). The modified AP configuration significantly reducedV65Gy, V70Gy, and V75Gy, meeting distributed protocol for all metricsexcept V75Gy, at the expense of poorer dose coverage. It should be notedthat these rectum doses were still significantly warmer than thoseachieved by AP Configurations A and C for this protocol.A concern with this type of analysis, particularly with multiple

metrics with an inherent dependence to one another, is the detection offalse positives from multiple testing. Bonferroni corrections are oftenemployed to account for this, however these corrections risk being tooconservative. Just as with the median DVHs, the results presented hereare illustrative of differences that arise between AP configurations forthe same protocol. As the main aim of this investigation was to assesswhether AP configurations could be adapted to meet these protocols,rather than document the extent that these plans differ, it was decidedby the authors not to proceed with multiple testing procedures.A limitation of this study is the lack of qualitative assessment from

an experienced radiation oncologist to assess the clinical acceptabilityof treatment plans developed from the modified AP configurations. Asthis was an investigation assessing whether AP configurations could beadapted between centres to meet local protocol, quantitative analysisonly was proposed. Future work investigating improvements in treat-ment plan quality through sharing of multiple centre’s AP configura-tions would require such a qualitative assessment.Automated treatment planning for prostate radiotherapy is a high

interest field, with multiple studies investigating implementation ofknowledge-based [22,23,29] and template derived [21] treatmentplanning protocols. Investigations validating the use of RapidPlan(Varian Medical Systems, Palo Alto, USA) have demonstrated the fea-sibility of sharing models across multiple centres [10,11]. This studyprovides further evidence for the viability of sharing clinical protocolsand training datasets to aid template based automatic treatment plan-ning of prostate cancer.Through the distribution of clinical protocols and three patient

training datasets centres were able to adapt local AP configurations tosatisfy other centre’s protocols. This study shows that AP configurationsare readily adaptable to different prostate radiotherapy protocols, andprovides a methodology for the sharing of AP configurations acrosscentres to assist efficient implementation and increased uptake of AP inthe clinic for the benefit of patients. Future work can then investigateadapting and improving local AP configurations based on the colla-boration of multiple centre’s configurations.

Funding

This project was partly funded by NHMRC project grant number1077788, Danish Cancer Society grant, University of Southern Denmarkscholarship and Odense University Hospital scholarship.

Conflict of interest statement

None.

Acknowledgements

The authors would like to thank Lois Holloway and MichaelJameson for their feedback.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.phro.2019.04.007.

References

[1] Batumalai V, Jameson MG, Forstner DF, Vial P, Holloway LC. How important isdosimetrist experience for intensity modulated radiation therapy? A comparativeanalysis of a head and neck case. Pract Radiat Oncol 2013;3:e99–106. https://doi.org/10.1016/j.prro.2012.06.009.

[2] Nelms BE, Robinson G, Markham J, Velasco K, Boyd S, Narayan S, et al. Variation inexternal beam treatment plan quality: an inter-institutional study of planners andplanning systems. Pract Radiat Oncol 2012;2:296–305. https://doi.org/10.1016/j.prro.2011.11.012.

[3] Berry SL, Boczkowski A, Ma R, Mechalakos J, Hunt M. Interobserver variability inradiation therapy plan output: Results of a single-institution study. Pract RadiatOncol 2016;6:442–9. https://doi.org/10.1016/j.prro.2016.04.005.

[4] Moore KL, Brame RS, Low DA, Mutic S. Quantitative metrics for assessing planquality. Semin Radiat Oncol 2012. https://doi.org/10.1016/j.semradonc.2011.09.005.

[5] Hansen C, Sykes J, Barber J, West K, Bromley R, Szymura K, et al. Multicentreknowledge sharing and planning/dose audit on flattening filter free beams for SBRTlung. J Phys Conf Ser 2015. https://doi.org/10.1088/1742-6596/573/1/012018.

[6] Peters LJ, O'Sullivan B, Giralt J, Fitzgerald TJ, Trotti A, Bernier J, et al. Criticalimpact of radiotherapy protocol compliance and quality in the treatment of ad-vanced head and neck cancer: results from TROG 02.02. J Clin Oncol2010;28:2996–3001. https://doi.org/10.1200/JCO.2009.27.4498.

[7] Abrams RA, Winter KA, Regine WF, Safran H, Hoffman JP, Lustig R, et al. Failure toadhere to protocol specified radiation therapy guidelines was associated with de-creased survival in RTOG 9704—a phase III trial of adjuvant chemotherapy andchemoradiotherapy for patients with resected adenocarcinoma of the pancreas. Int JRadiat Oncol Biol Phys 2012;82:809–16. https://doi.org/10.1016/j.ijrobp.2010.11.039.

[8] Dühmke E, Franklin J, Pfreundschuh M, Sehlen S, Willich N, Rühl U, et al. Low-doseradiation is sufficient for the noninvolved extended-field treatment in favorableearly-stage Hodgkin’s disease: long-term results of a randomized trial of radio-therapy alone. J Clin Oncol 2001;19:2905–14. https://doi.org/10.1200/JCO.2001.19.11.2905.

[9] Xhaferllari I, Wong E, Bzdusek K, Lock M, Chen JZ. Automated IMRT planning withregional optimization using planning scripts. J Appl Clin Med Phys2013;14:176–91. https://doi.org/10.1120/jacmp.v14i1.4052.

[10] Fogliata A, Belosi F, Clivio A, Navarria P, Nicolini G, Scorsetti M, et al. On the pre-clinical validation of a commercial model-based optimisation engine: application tovolumetric modulated arc therapy for patients with lung or prostate cancer.Radiother Oncol 2014;113:385–91. https://doi.org/10.1016/j.radonc.2014.11.009.

[11] Schubert C, Waletzko O, Weiss C, Voelzke D, Toperim S, Roeser A, et al. Intercentervalidation of a knowledge based model for automated planning of volumetricmodulated arc therapy for prostate cancer. The experience of the German RapidPlan Consortium. PLoS One 2017;12:e0178034. https://doi.org/10.1371/journal.pone.0178034.

[12] Moore KL, Kagadis GC, McNutt TR, Moiseenko V, Mutic S. Vision 20/20: automa-tion and advanced computing in clinical radiation oncology. Med Phys 2014;41.https://doi.org/10.1118/1.4842515.

[13] Heijmen B, Voet P, Fransen D, Penninkhof J, Milder M, Akhiat H, et al. Fully au-tomated, multi-criterial planning for Volumetric Modulated Arc Therapy–an inter-national multi-center validation for prostate cancer. Radiother Oncol 2018. https://doi.org/10.1016/j.radonc.2018.06.023.

[14] Hazell I, Bzdusek K, Kumar P, Hansen CR, Bertelsen A, Eriksen JG, et al. Automaticplanning of head and neck treatment plans. J Appl Clin Med Phys 2016;17. https://doi.org/10.1120/jacmp.v17i1.5901.

[15] Gintz D, Latifi K, Caudell J, Nelms B, Zhang G, Moros E, et al. Initial evaluation ofautomated treatment planning software. J Appl Clin Med Phys 2016;17. https://doi.org/10.1120/jacmp.v17i3.6167.

[16] Hansen CR, Bertelsen A, Hazell I, Zukauskaite R, Gyldenkerne N, Johansen J, et al.Automatic treatment planning improves the clinical quality of head and neck cancertreatment plans. Clin Transl Radiat Oncol 2016;1:2–8. https://doi.org/10.1016/j.ctro.2016.08.001.

[17] Krayenbuehl J, Norton I, Studer G, Guckenberger M. Evaluation of an automatedknowledge based treatment planning system for head and neck. Radiat Oncol2015;10:226. https://doi.org/10.1186/s13014-015-0533-2.

[18] Kusters J, Bzdusek K, Kumar P, van Kollenburg PM, Kunze-Busch M, Wendling M,et al. Automated IMRT planning in Pinnacle. Strahlenther Onkol 2017;193:1031–8.https://doi.org/10.1007/s00066-017-1187-9.

[19] Tol JP, Delaney AR, Dahele M, Slotman BJ, Verbakel WF. Evaluation of a knowl-edge-based planning solution for head and neck cancer. Int J Radiat Oncol Biol Phys2015;91:612–20. https://doi.org/10.1016/j.ijrobp.2014.11.014.

[20] Hansen CR, Nielsen M, Bertelsen AS, Hazell I, Holtved E, Zukauskaite R, et al.Automatic treatment planning facilitates fast generation of high-quality treatmentplans for esophageal cancer. Acta Oncol 2017;56:1495–500. https://doi.org/10.1080/0284186X.2017.1349928.

[21] Nawa K, Haga A, Nomoto A, Sarmiento RA, Shiraishi K, Yamashita H, et al.Evaluation of a commercial automatic treatment planning system for prostate


12


cancers. Med Dosim 2017. https://doi.org/10.1016/j.meddos.2017.03.004.[22] Nwankwo O, Mekdash H, Sihono DSK, Wenz F, Glatting G. Knowledge-based ra-

diation therapy (KBRT) treatment planning versus planning by experts: validationof a KBRT algorithm for prostate cancer treatment planning. Radiat Oncol2015;10:111. https://doi.org/10.1186/s13014-015-0416-6.

[23] Powis R, Bird A, Brennan M, Hinks S, Newman H, Reed K, et al. Clinical im-plementation of a knowledge based planning tool for prostate VMAT. Radiat Oncol2017;12:81. https://doi.org/10.1186/s13014-017-0814-z.

[24] Chanyavanich V, Das SK, Lee WR, Lo JY. Knowledge-based IMRT treatment plan-ning for prostate cancer. Med Phys 2011;38:2515–22. https://doi.org/10.1118/1.3574874.

[25] Gallio E, Giglioli FR, Girardi A, Guarneri A, Ricardi U, Ropolo R, et al. Evaluation ofa commercial automatic treatment planning system for liver stereotactic body ra-diation therapy treatments. Phys Med 2018;46:153–9. https://doi.org/10.1016/j.ejmp.2018.01.016.

[26] Wang Y, Heijmen BJ, Petit SF. Prospective clinical validation of independent DVHprediction for plan QA in automatic treatment planning for prostate cancer patients.

Radiother Oncol 2017;125:500–6. https://doi.org/10.1016/j.radonc.2017.09.021.[27] Janssen TM, Kusters M, Wang Y, Wortel G, Monshouwer R, Damen E, et al.

Independent knowledge-based treatment planning QA to audit Pinnacle autoplan-ning. Radiother Oncol 2018. https://doi.org/10.1016/j.radonc.2018.10.035.

[28] Wu B, Kusters M, Kunze-busch M, Dijkema T, McNutt T, Sanguineti G, et al. Cross-institutional knowledge-based planning (KBP) implementation and its performancecomparison to Auto-Planning Engine (APE). Radiother Oncol 2017;123:57–62.https://doi.org/10.1016/j.radonc.2017.01.012.

[29] Good D, Lo J, Lee WR, Wu QJ, Yin F-F, Das SK. A knowledge-based approach toimproving and homogenizing intensity modulated radiation therapy planningquality among treatment centers: an example application to prostate cancer plan-ning. Int J Radiat Oncol Biol Phys 2013;87:176–81. https://doi.org/10.1016/j.ijrobp.2013.03.015.

[30] Bertelsen A, Hansen CR, Johansen J, Brink C. Single arc volumetric modulated arctherapy of head and neck cancer. Radiother Oncol 2010;95:142–8. https://doi.org/10.1016/j.radonc.2010.01.011.


13


Table A.1: Target objectives, and OAR evaluations for all three protocols


Target Prescription Target Prescription Target Prescription PTV1 35 x 2 Gy PTV1 39 x 2 Gy PTV1 39 x 2 Gy PTV2 35 x 2.2 Gy Target Evaluation Priority Target Evaluation Priority Target Evaluation Priority PTV1 V95% > 99% Constraint PTV1 D95% > 100% Medium PTV1imrt D99% > 95% Medium PTV2 V95% > 99% Constraint PTV1 D2% < 105% Medium PTV1imrt D0% < 107% Medium PTV1/PTV2 D1% < 107% Constraint CTV1 D99% > 100% High PTV1imrt D95% > 95% High PTV1/PTV2 D1% < 105% Medium PTV1 Dmax < 107% Medium –

Soft PTV1 D99.9% > 66.5

Gy High

PTV_prostate_imrt D99% > 95% Medium PTV_prostate_imrt D97% > 95% High OAR Evaluation Priority OAR Evaluation Priority OAR Evaluation Priority Rectal Wall V75Gy < 10% High Rectum V40Gy < 50% Medium Rectum V74Gy < 1cm3 Constraint Rectal Wall V64Gy < 35% High Rectum V50Gy < 40% Medium Rectum V70Gy < 20% Medium Anal Sphincter Dmean < 45 Gy High Rectum V60Gy < 30% Medium Rectum V70Gy < 10% Soft Femur Joint Area Dmax < 50 Gy Medium Rectum V65Gy < 20% Constraint Rectum V65Gy < 30% Medium 90% - 95% isodose tight around target

High Rectum V70Gy < 10% Constraint Rectum V60Gy < 50% Medium

No rectum hotpots Medium Rectum V75Gy < 5% High Rectum V50Gy < 60% Medium RECT_IN_PTV V79Gy < 1 cm3 High Circumference of

rectum Dmax < 50 Gy Constraint

RECT_78_03_NS V80Gy < 1 cm3 Constraint Femur Cap Dmax < 52 Gy Medium Post rectum Dmax < 39 Gy Medium Femur Cap V50Gy < 50% Medium Femoral Heads V40Gy < 10% Soft Penile Bulb V40Gy < 50% Medium Bladder V50Gy < 40% Soft Bladder V70Gy < 30% Medium Bladder V60Gy < 30% Soft Bladder V70Gy < 20% Soft Peripheral Tissue V50Gy < 1 cm3 Soft Bladder V60Gy < 40% Medium Bladder V60Gy < 25% Soft

Chapter

6.Adapting

automated

treatmentplanning

configurations199


Bladder V50Gy < 60% Medium Bladder V50Gy < 50% Soft Bowel Cavity V50Gy < 50% Medium

Chapter

6.Adapting

automated

treatmentplanning

configurations200

Table A.2(a): AP Technique A adapted across all three protocols

Protocol A

Target Optimisation Goals Dose PTV70min77 70 Gy PTV2 77 Gy Organs at Risk Objective Dose Volume Priority Compromise Rectal Wall Mean Dose 0 Gy High Yes PTVring Max Dose 66.5 Gy High Yes Anal Sphincter Mean Dose 0 Gy High Yes Rectal Wall Max DVH 75 Gy 7% High Yes RECT-PTV+5 Max Dose 0 Gy Medium Yes Tuning Balance 20% Dose Fall-Off 1.5 cm Hot-Spot Maximum Goal 107% Use Cold-Spot ROIs Yes Protocol B

Target Optimisation Goals Dose PTV1MR-CTV_safe 79 Gy CTV_safe 80 Gy Organs at Risk Objective Dose Volume Priority Compromise Rectum Max DVH 65 Gy 17% Medium Yes Rectum Max DVH 75 Gy 2% Medium Yes Rectum Max DVH 40 Gy 47% Medium Yes Rectum Max DVH 70 Gy 7% Medium Yes Tuning Balance 11% Dose Fall-Off 2.6 cm Hot-Spot Maximum Goal 107% Use Cold-Spot ROIs Yes Protocol C

Target Optimisation Goals Dose Level resd_ptv 70 Gy ptv-ctv 78.01 Gy PTV78imrt 78 Gy Organs at Risk Objective Dose Level Volume Priority Compromise Rectum Mean Dose 0 Gy High Yes Rectum Max DVH 74 Gy 1% High Yes PTVring Max Dose 74.1 Gy High Yes Bladder Max DVH 70 Gy 17% Medium Yes


Bladder Max DVH 60 Gy 22% Medium Yes Tuning Balance 20% Dose Fall-Off 1.5 cm Hot-Spot Maximum Goal 107% Use Cold-Spot ROIs Yes


Table A.2(b): AP Technique B adapted across all three protocols

Protocol A

Target Optimisation Goals Dose PTV1 70.5 Gy PTV2 78 Gy Organs at Risk Objective Dose Volume Priority Compromise Rectum Max Dose 74 Gy High Yes Rectum Mean Dose 34 Gy Medium Yes Rectum Max DVH 72 Gy 5% High Yes Rectum Max DVH 60 Gy 40% Medium Yes Rectum Max DVH 38 Gy 50% High Yes RECT_78_03_NS Max Dose 80 Gy High Yes POSTRECT_NS Max DVH 38 Gy 2% Medium Yes Tuning Balance 11% Dose Fall-Off 2.6 cm Hot-Spot Maximum Goal 103% Use Cold-Spot ROIs Yes Protocol B

Target Optimisation Goals Dose PTV78_OPT 78.2 Gy CTV_ANTPROST 80 Gy Organs at Risk Objective Dose Volume Priority Compromise Rectum Max Dose 79.3 Gy High Yes Rectum Mean Dose 34 Gy Medium Yes Rectum Max DVH 73 Gy 10% High Yes Rectum Max DVH 60 Gy 40% Medium Yes Rectum Max DVH 38 Gy 50% High Yes RECT_78_03_NS Max Dose 80 Gy High Yes Bladder Max DVH 50 Gy 50% Medium Yes Bladder Max DVH 60 Gy 40% Medium Yes Bladder Mean Dose 15 Gy Low Yes POSTRECT_NS Max DVH 38 Gy 2% Medium Yes Tuning Balance 11% Dose Fall-Off 2.6 cm Hot-Spot Maximum Goal 103% Use Cold-Spot ROIs Yes Protocol C

Target Optimisation Goals Dose Level PTV78imrt 78 Gy


CTV_ANTPROST 80 Gy PTV1 72 Gy Organs at Risk Objective Dose Level Volume Priority Compromise Rectum Max Dose 73 Gy High Yes Rectum Mean Dose 34 Gy Medium Yes Rectum Max DVH 72.5 Gy 2% High Yes Rectum Max DVH 60 Gy 40% Medium Yes Rectum Max DVH 38 Gy 50% High Yes RECT_78_03_NS Max Dose 72 Gy High Yes Bladder Max DVH 50 Gy 40% Medium Yes Bladder Max DVH 60 Gy 30% Medium Yes Bladder Mean Dose 15 Gy Low Yes POSTRECT_NS Max DVH 38 Gy 2% Medium Yes Tuning Balance 11% Dose Fall-Off 2.6 cm Hot-Spot Maximum Goal 103% Use Cold-Spot ROIs Yes


Table A.2(c): AP Technique C adapted across all three protocols

Protocol A

Target Optimisation Goals Dose PTV1 70 Gy PTV2 77 Gy CTV1 77 Gy Organs at Risk Objective Dose Volume Priority Compromise Rectal Wall Max DVH 75 Gy 10% High Yes Rectal Wall Max DVH 64 Gy 35% High Yes Rectal Wall Mean Dose 50 Gy High Yes Anal Sphincter Mean Dose 45 Gy High Yes Femur Joint Area (Left) Max Dose 50 Gy Medium Yes Femur Joint Area (Right) Max Dose 50 Gy Medium Yes Tuning Balance 11% Dose Fall-Off 2.6 cm Hot-Spot Maximum Goal 107% Use Cold-Spot ROIs Yes Protocol B

Target Optimisation Goals Dose PTV1 78.5 Gy CTV1 80.5 Gy PTVring3mm-Rectum 78.5 Gy Organs at Risk Objective Dose Volume Priority Compromise Rectum Max DVH 40 Gy 50% Medium Yes Rectum Max DVH 50 Gy 40% Medium Yes Rectum Max DVH 60 Gy 30% Medium Yes Rectum Max DVH 65 Gy 20% High Yes Rectum_non_compromise Max DVH 70 Gy 10% High No Rectum Max DVH 75 Gy 5% Medium Yes Rectum Max DVH 79 Gy 1.4% Medium Yes Rectum Mean Dose 45 Gy Medium Yes RECT_78_03_NS Max DVH 80 Gy 5% High Yes Femoral Heads (Left) Max DVH 40 Gy 10% Low Yes Femoral Heads (Right) Max DVH 40 Gy 10% Low Yes Bladder Max DVH 50 Gy 40% Low Yes Bladder Max DVH 60 Gy 30% Low Yes Peripheral Tissue Max DVH 60 Gy 0% Low Yes Tuning Balance 11% Dose Fall-Off 2.6 cm Hot-Spot Maximum Goal 107% Use Cold-Spot ROIs Yes


Protocol C

Target Optimisation Goals Dose Level PTV78imrt 77.5 Gy PTVinRec 68 Gy PTV78LatRectum 77.3 Gy PTV_prostata_imrt 77.5 Gy PTVindreRand 77.4 Gy Organs at Risk Objective Dose Level Volume Priority Compromise RectumVclose Max DVH 74 Gy 25% High No Bladder Max DVH 70 Gy 20% Medium Yes Bladder Mean Dose 30 Gy Medium Yes Bowel Cavity Max DVH 70 Gy 1% High Yes Bowel Cavity Mean Dose 40 Gy High Yes Femur Cap (Left) Max Dose 52 Gy Medium Yes Femur Cap (Right) Max Dose 52 Gy Medium Yes RectumRes Mean Dose 20 Gy High Yes Penile Bulb Mean Dose 40 Gy High Yes RectumPostWall Max Dose 50 Gy High No BladderIMRT Max Dose 74.1 Gy High Yes Tuning Balance 11% Dose Fall-Off 2.6 cm Hot-Spot Maximum Goal 107% Use Cold-Spot ROIs Yes


Table A.3: Treatment planning system and beam setup configurations

Centre A Centre B Centre C

LINAC Versa HD 5 mm MLC at ISO

Versa HD 5 mm MLC at ISO

Versa HD 5 mm MLC at ISO

Slice Thickness 3 mm 2 mm 3 mm Dose Grid 4 mm x 4 mm x 4 mm 2.5 mm x 2.5 mm x 2.5

mm 3 mm x 3 mm x 3 mm

Dose Algorithm Collapsed Cone Adaptive Convolve Collapsed Cone Arcs Dual Arc Single Arc Single Arc Beam Energy 10 MV 6 MV 18 MV Beam Angles 140° - 220° 176° - 184° 180° - 182° Collimator 20° 10° 35°


Table A.4: Patient demographics


Selection Criteria

High Risk (T1-T3a AND PSA >= 20 AND Gleason >=8) OR (T3b – T4a)

High Risk (T1-T3a AND PSA >= 20 AND Gleason >=8) OR (T3b – T4a)

Hormonal Treatment

Yes Yes

Age 58 – 83 (Median: 72) 66.8 – 76.9 (Median: 70.7) PTV Volume PTV1: 84.96 – 313.81 cm3

(Median: 160.91 cm3) PTV2:55.70 – 246.28 cm3 (Median: 113.45 cm3)

PTV1: 110.36 – 373.88 cm3 (Median: 136.07 cm3)

PTV1: 104.65 – 307.92 cm3 (Median: 139.15 cm3)

PTV/Rectum Overlap

PTV1: 2.28 – 11.30% (Median: 6.84%) PTV2: 0.32 - 4.75% (Median: 1.84%)

PTV1: 1.30 – 12.50% (Median: 7.35%)

PTV1: 3.47 – 14.22% (Median: 8.13%)

Planning Difficulty

Easy: 2 Medium: 3 Hard: 5




Table A.5: Target and OAR mean and standard deviations (S.D.) for each centre’s protocols. Conformity and homogeneity indices are also shown for each protocol’s PTV. Differences in metrics considered significant (p < 0.05) are bolded.

AP Technique A AP Technique B AP Technique C

Mean S.D. p-value Mean S.D. p-value Mean S.D. p-value

Protocol A

PTV1: V95% > 99% 99.9% 0.1% 100.0% 0.1% 0.08 100.0% < 0.1% 0.002 PTV2: V95% > 99% 99.7% 0.2% 99.9% 0.2% 0.04 100.0% < 0.1% 0.002 PTV2: D1% < 107% 103.1% 0.5% 102.6% 0.6% 0.03 102.4% 0.3% 0.01 Rectal Wall: V75Gy < 10% 4.2% 2.3% 5.1% 3.0% 0.11 6.6% 2.6% 0.002 Rectal Wall: V64Gy < 35% 14.8% 5.8% 16.2% 5.3% 0.002 17.2% 6.0% 0.002 Anal Sphincter: Dmean < 45 Gy 9.1 Gy 7.7 Gy 10.4 Gy 8.7 Gy 0.004 10.0 Gy 9.4 Gy 0.23 Femur Joint Area (Left): Dmax < 50 Gy 45.9 Gy 6.0 Gy 36.7 Gy 6.6 Gy 0.002 19.8 Gy 9.2 Gy 0.002 Femur Joint Area (Right): Dmax < 50 Gy 43.9 Gy 6.7 Gy 35.9 Gy 7.1 Gy 0.002 19.9 Gy 7.9 Gy 0.002 PTV1 Conformity Index 1.19 0.04 1.33 0.06 0.002 1.31 0.06 0.002 PTV2 Conformity Index 1.12 0.05 1.29 0.10 0.002 1.29 0.08 0.002 PTV1 Homogeneity Index 0.13 0.01 0.10 0.01 0.002 0.09 0.01 0.002 PTV2 Homogeneity Index 0.07 < 0.01 0.05 0.01 0.002 0.05 < 0.01 0.002 Protocol B

PTV1: D95% > 100% 99.2% 1.4% 0.23 99.6% 0.9% 99.2% 1.6% 0.51 PTV1: D2% < 105% 104.0% 0.2% 0.69 104.0% 0.3% 104.4% 0.3% 0.02 CTV1: D99% > 100% 101.0% 0.7% 0.83 101.0% 0.5% 101.3% 0.7% 0.32 PTV1: Dmax < 107% 105.2% 0.3% 0.13 105.5% 0.7% 105.7% 0.3% 0.70 Rectum: V40Gy < 50% 34.8% 6.0% 0.002 25.7% 6.1% 31.9% 7.0% 0.002 Rectum: V50Gy < 40% 24.6% 4.6% 0.002 20.3% 5.1% 22.4% 5.5% 0.02 Rectum: V60Gy < 30% 16.4% 3.4% 0.28 15.8% 4.4% 15.1% 4.0% 0.23 Rectum: V65Gy < 20% 12.6% 2.8% 0.11 13.6% 4.0% 11.7% 3.3% 0.002 Rectum: V70Gy < 10% 8.5% 2.1% 0.004 11.0% 3.6% 8.1% 2.4% 0.002

Chapter

6.Adapting

automated

treatmentplanning

configurations209

Rectum: V75Gy < 5% 3.6% 1.2% 0.002 7.6% 2.7% 3.5% 1.2% 0.002 RECT_IN_PTV: V79Gy < 1 cm3 0.1 cm3 0.1 cm3 0.004 0.5 cm3 0.4 cm3 0.1 cm3 0.1 cm3 0.006 RECT_78_03_NS: V80Gy < 1 cm3 0.3 cm3 0.2 cm3 0.49 0.3 cm3 0.3 cm3 0.5 cm3 0.3 cm3 0.13 Femoral Heads (Left): V40Gy < 10% 1.6% 3.9% 0.69 1.7% 2.1% 0.0% 0.0% 0.02 Femoral Heads (Right): V40Gy < 10% 0.8% 2.0% 0.03 1.8% 2.9% 0.0% 0.0% 0.03 Bladder: V50Gy < 40% 23.4% 7.8% 0.002 18.8% 7.3% 18.5% 6.7% 1.00 Bladder: V60Gy < 30% 17.2% 5.5% 0.002 14.3% 5.3% 13.7% 4.9% 0.32 Peripheral Tissue: V50Gy < 1 cm3 0.0 cm3 0.0 cm3 0.031 0.1 cm3 0.1 cm3 0.0 cm3 0.0 cm3 0.03 PTV1 Conformity Index 1.32 0.07 0.63 1.33 0.08 1.27 0.05 0.04 PTV1 Homogeneity Index 0.07 0.02 0.006 0.06 0.01 0.08 0.03 0.006 Protocol C

PTV1imrt: D99% > 95% 93.9% 0.7% 0.006 94.9% 0.9% 0.85 95.0% 0.6% PTV1imrt: D0% < 107% 103.7% 0.5% 0.002 106.1% 0.4% 0.05 105.5% 0.8% PTV1imrt: D95% > 95% 96.2% 0.3% 0.002 97.5% 0.6% 0.04 97.1% 0.4% PTV1: D99.9% > 66.5 Gy 69.1 Gy 0.5 Gy 0.002 69.9 Gy 1.0 Gy 0.004 67.9 Gy 0.4 Gy PTV_prostate_imrt: D99% > 95% 94.0% 0.9% 0.020 95.0% 0.9% 1.00 94.9% 0.6% PTV_prostate_imrt: D97% > 95% 95.6% 0.4% 0.004 96.7% 0.7% 0.16 96.3% 0.5% Rectum: V74Gy < 1cm3 0.5 cm3 0.2 cm3 0.16 0.8 cm3 0.3 cm3 0.38 0.7 cm3 0.3 cm3 Rectum: V70Gy < 20% 8.2% 2.7% 0.002 9.5% 3.2% 0.002 7.2% 2.6% Rectum: V65Gy < 30% 15.6% 4.9% 0.002 13.7% 4.2% 1.00 13.7% 4.5% Rectum: V60Gy < 50% 20.7% 6.3% 0.01 16.9% 4.9% 0.07 18.5% 5.8% Rectum: V50Gy < 60% 29.5% 8.5% 0.03 22.6% 5.8% 0.01 26.8% 7.9% Femur Cap (Left): Dmax < 52 Gy 37.4 Gy 5.5 Gy 0.01 39.1 Gy 4.7 Gy 0.002 31.9 Gy 6.1 Gy Femur Cap (Right): Dmax < 52 Gy 39.2 Gy 5.0 Gy 0.004 39.2 Gy 5.4 Gy 0.01 32.8 Gy 7.1 Gy Femur Cap (Left): V50Gy < 50% 0.0% 0.0% 1.00 0.0% 0.0% 1.00 0.0% 0.0% Femur Cap (Rigth): V50Gy < 50% 0.0% 0.0% 1.00 0.0% 0.0% 1.00 0.0% 0.0% Penile Bulb: V40Gy < 50% 22.9% 31.1% 0.02 13.3% 16.9% 0.03 8.6% 13.0% Bladder: V70Gy < 30% 17.0% 11.1% 0.63 17.3% 9.7% 1.00 17.5% 11.7% Bladder: V60Gy < 40% 27.2% 16.8% 0.43 24.7% 13.1% 0.38 26.7% 16.7% Bladder: V50Gy < 60% 36.8% 20.8% 0.16 31.9% 16.0% 0.16 34.9% 20.0%

Chapter

6.Adapting

automated

treatmentplanning

configurations210

Bowel Cavity: V50Gy < 50% 0.7% 2.2% 0.50 0.7% 2.0% 0.25 0.5% 1.5% PTV1 Conformity Index 0.99 0.02 0.002 1.11 0.03 0.70 1.11 0.03 PTV1 Homogeneity Index 0.11 0.01 0.002 0.12 0.01 0.70 1.12 0.01

Chapter

6.Adapting

automated

treatmentplanning

configurations211

Table A.6: Total recorded violations by each centre’s AP technique

AP Technique A

AP Technique B

AP Technique C

Constraint Protocol A Protocol B Rectum: V70Gy < 10% 1 7 RECT_78_03_NS: V80Gy < 1 cm3 1 Protocol C

High Protocol A Protocol B CTV1: D99% > 100% 1 1 Rectum: V75Gy < 5% 8 RECT_IN_PTV: V79Gy < 1 cm3 1 Protocol C

Medium Protocol A Femur Joint Area (Left): Dmax < 50 Gy 2

Femur Joint Area (Right): Dmax < 50 Gy 3 Protocol B PTV1: D95% > 100% 7 4 4 Protocol C PTV1imrt: D99% > 95% 10 4 5 PTV_prostate_imrt: D99% > 95% 9 4 5 Penile Bulb: V40Gy < 50% 1 Bladder: V70Gy < 30% 1 1 1 Bladder: V60Gy < 40% 1 1 1 Bladder: V50Gy < 60% 1 1 1

Low

Protocol A Protocol B Femoral Heads (Left): V40Gy < 10% 1 Protocol C Rectum: V70Gy < 10% 1 3 1 Bladder: V70Gy < 20% 3 4 4 Bladder: V60Gy < 25% 5 5 5 Bladder: V50Gy < 50% 2 1 1


Table A.7: Modified AP Technique B to meet distributed protocol B criteria

Target Optimisation Goals Dose PTV78_OPT 78.4 Gy CTV_ANTPROST 80 Gy Organs at Risk Objective Dose Volume Priority Compromise Rectum Max Dose 78.3 Gy High Yes Rectum Mean Dose 33.5 Gy Medium Yes Rectum Max DVH 73 Gy 5% High Yes Rectum Max DVH 60 Gy 40% Medium Yes Rectum Max DVH 38 Gy 50% High Yes RECT_78_03_NS Max Dose 80 Gy High Yes Bladder Max DVH 50 Gy 50% Medium Yes Bladder Max DVH 60 Gy 40% Medium Yes Bladder Mean Dose 15 Gy Low Yes POSTRECT_NS Max DVH 38 Gy 2% Medium Yes Tuning Balance 11% Dose Fall-Off 2.6 cm Hot-Spot Maximum Goal 103% Use Cold-Spot ROIs Yes


Table A.8: Original and modified AP technique B for protocol B, with significant differences between metrics bolded.

AP Technique B AP Technique B (Modified

Mean S.D. Mean S.D. p-value

PTV1: D95% > 100% 99.6% 0.9% 99.1% 0.7% 0.006 PTV1: D2% < 105% 104.0% 0.3% 104.4% 0.2% 0.01 CTV1: D99% > 100% 101.0% 0.5% 100.1% 0.5% 0.004 PTV1: Dmax < 107% 105.5% 0.7% 106.0% 0.7% 0.16 Rectum: V40Gy < 50% 25.7% 6.1% 25.0% 4.7% 0.32 Rectum: V50Gy < 40% 20.3% 5.1% 19.3% 4.4% 0.07 Rectum: V60Gy < 30% 15.8% 4.4% 14.7% 4.1% 0.05 Rectum: V65Gy < 20% 13.6% 4.0% 12.4% 3.9% 0.04 Rectum: V70Gy < 10% 11.0% 3.6% 9.7% 3.5% 0.01 Rectum: V75Gy < 5% 7.6% 2.7% 5.8% 2.1% 0.002 RECT_IN_PTV: V79Gy < 1 cm3 0.5 cm3 0.4 cm3 0.4 cm3 0.3 cm3 0.32 RECT_78_03_NS: V80Gy < 1 cm3 0.3 cm3 0.3 cm3 0.2 cm3 0.2 cm3 0.70 Femoral Heads (Left): V40Gy < 10% 1.7% 2.1% 0.9% 2.7% 0.81 Femoral Heads (Right): V40Gy < 10% 1.8% 2.9% 0.3% 0.5% 0.31 Bladder: V50Gy < 40% 18.8% 7.3% 18.3% 6.3% 0.85 Bladder: V60Gy < 30% 14.3% 5.3% 13.7% 4.5% 0.38 Peripheral Tissue: V50Gy < 1 cm3 0.1 cm3 0.1 cm3 < 0.1 cm3 < 0.1 cm3 0.03 PTV1 Conformity Index 1.33 0.08 1.24 0.05 0.002 PTV1 Homogeneity Index 0.06 0.01 0.07 0.01 0.002


Figure A.1: Mean left femoral head DVHs for all patients for AP Configurations A (blue), B (red), and C (green). Note that scale begins at 60 Gy for clarity. Solid lines correspond to host-centre AP configuration, modified configurations are dashed. Solid and dotted grey p-value curves, indicating significant differences between population mean DVHs, are also illustrated. The dashed black line shows p = 0.05.


Figure A.2: Mean right femoral head DVHs for all patients for AP Configurations A (blue), B (red), and C (green). Note that scale begins at 60 Gy for clarity. Solid lines correspond to host-centre AP configuration, modified configurations are dashed. Solid and dotted grey p-value curves, indicating significant differences between population mean DVHs, are also illustrated. The dashed black line shows p = 0.05.


Figure A.3: Mean bladder DVHs for all patients for AP Configurations A (blue), B (red), and C (green). Note that scale begins at 60 Gy for clarity. Solid lines correspond to host-centre AP configuration, modified configurations are dashed. Solid and dotted grey p-value curves, indicating significant differences between population mean DVHs, are also illustrated. The dashed black line shows p = 0.05.


CHAPTER 7

Discussion and Conclusion

218

Chapter 7. Discussion and Conclusion 219

7.1 General Discussion

This thesis investigated the utilisation of advanced analysis techniques to quantify vari-

ability observed during contouring and treatment planning. These developed techniques

allow for a determination of the clinical impact that results from variation in quality dur-

ing these steps. It is well established that these manual processes are prone to significant

inter-observer variations [25, 36], with uncertainties propagating throughout all aspects of

treatment due to these steps being at the beginning of the radiotherapy workflow (Figure

1.1). Incorporation of automated methods to reduce variation, along with utilisation of

the aforementioned analysis techniques presented in this thesis, will allow for significant

improvements in quality assurance during a clinical trial

Inter-observer contouring variability is widely investigated within the literature, however

there exists a lack of consensus regarding how these variations are to be quantified [26, 27].

Not only do commonly utilised contouring metrics fail to correlate with one another [28],

but often they provide no dosimetric context regarding the clinical impact of the varia-

tion. This was addressed in Chapter 3, whereby contouring metrics that correlated with

simulated treatment outcome for prostate radiotherapy were identified. These metrics

were subsequently utilised during investigations in Chapter 4, whereby inter-observer con-

touring variability of multiple male pelvic structures were assessed on both CT and MRI

datasets.

The utilisation of automation to reduce inter-observer variation in radiotherapy is of sig-

nificant interest within the literature [28]. Automatic contouring is a rapidly developing

field, with many studies investigating the use of atlas-based contouring methods [15]. Not

only does atlas-based contouring provide significant time-savings within the clinical work-

flow, but it can also be applied retrospectively to investigate large clinical trial datasets.

Variations between atlas and manually defined CTV contours across the RADAR, RT01,

and CHHiP prostate radiotherapy clinical trial datasets were investigated in Chapter 5.

This was assessed using vector mappings, whereby specific spatial regions with significant

variability surrounding the CTV were identified for each trial. Cox Regression analysis

was utilised to identify regions where contouring variability correlated with a significant


increase in risk of treatment failure. These results confirm that not only does contouring

variability impact the efficacy of treatment, but the clinical impact is highly dependent

on the spatial location of these discrepancies.

Automation has also been incorporated into treatment planning, with most vendors now

providing some form of automatic treatment planning module within their treatment plan-

ning system. Common methods of automation are knowledge-based models, favoured by

Varian3 within their RapidPlan™ module [402], and template-driven modules ,such as the

AutoPlanning™ module in Pinnacle3 [42]. AutoPlanning™ (AP) configurations are devel-

oped locally, with clinics developing site specific processes to generate treatment plans that

are clinically acceptable based on local protocol. Chapter 6 investigated the robustness

and quality of three different clinic’s AP configurations, and showed that each configu-

ration was able to be adapted to meet each other clinic’s treatment planning protocol.

This highlighted the ability for AP configurations to be shared between clinics, whereby

smaller centres that may lack the resources required to develop local AP configurations

could benefit from the work performed by larger centres.

This thesis investigated advanced analysis techniques aimed at addressing some of the

largest sources of uncertainty within the radiotherapy workflow, particularly with respect

to clinical trial quality assurance. Insight is provided into how these techniques could be

utilised to quantify, characterise, and reduce the variability that is observed. The main

themes of research carried out during this candidature were outlined in Chapter 1 and

include:

1. An assessment of inter-observer contouring uncertainties for target volumes and

organs-at-risk for intact prostate radiotherapy,

2. The effect of target volume contouring variability on recorded patient outcome within

prostate radiotherapy clinical trial datasets, and

3. An evaluation of quality and robustness of automated treatment planning configu-

rations shared between international centres.

These three themes each relate to improving quality assurance of radiotherapy clinical

trials (Figure 1.1). During a clinical trial it is imperative that there is an understanding


of possible uncertainties associated with each step within the radiotherapy workflow. This

is particularly true for investigations looking at emerging techniques and treatments, as

uncertainties and variations that arise early in the radiotherapy workflow will propagate

throughout all stages of treatment. Reductions in quality of treatment will not only

compromise potential findings from the clinical trial, but could also potentially place the

patient at risk if dangerous treatment is administered.

It is important when assessing contouring variability to understand the clinical impact

these variations would have on treatment outcome. Contouring metrics that quantify

variation, but provide no information regarding how a patient will respond to treatment

that utilises these contours, will provide no benefit to clinical trials aiming to improve

patient outcome. By identifying metrics that do correlate with treatment outcome for

prostate radiotherapy, future clinical trials could therefore incorporate these metrics during

quality assurance procedures. This would ensure that assessment of contouring quality

during benchmarking would be based on treatment outcome measures, as opposed to

simple topological descriptions.

An understanding of the extent of contouring variability is required when determining

potential links between delivered dose distributions to anatomical structures and recorded

patient outcomes. Any uncertainty in identifying the structure during contouring will

result in an associated uncertainty in the actual dose delivered to that structure. This

uncertainty and variation could not only weaken any potential correlations, but may addi-

tionally result in completely erroneous findings altogether. This is particularly true when

investigating anatomical regions where multiple small structures overlap. As was shown in

Chapter 4, significant differences can arise when contouring small structures of emerging

interest. Consequently, investigations studying dose delivered to these structures will have

uncertainty in any findings unless appropriate measures were undertaken to either reduce

or quantify this variability.

Conversely, if it is found that variations exist when defining the target volume contours

during a clinical trial, then a significant number of enrolled patients will receive inade-

quate radiotherapy treatment. This would drastically impact the patient’s care, as well

as limiting the clinical trial’s efficacy. This is also true for treatment planning, whereby


inadequate dose coverage of the target volume could fail to meet the trial’s protocols. For

these reasons, quantifying and reducing variability in contouring and treatment planning

are two of the most important considerations when creating a quality assurance protocol

for a clinical trial. This is the basis for investigations within this thesis. Through analysis

of clinical trial datasets and recorded patient outcomes, results were able to link these

variations and poorer treatment outcomes for these patient. Future clinical trials could

significantly improve the quality of patient treatment through use of the results from these

retrospective analyses, allowing us to move forward in standards of clinical care.

7.2 Inter-observer contouring uncertainties for intact prostate

radiotherapy

Chapter 3 identified multiple contouring variation metrics commonly utilised in contour-

ing studies, and assessed whether these metrics significantly correlated with simulated

treatment outcome for prostate radiotherapy. Two datasets were investigated; the first

comprised of 35 patients independently contoured by 3 observers, the second a cohort

of 5 patients contoured by 10 observers. Clinical target volume (CTV), planning target

volume (PTV), bladder and rectum were contoured on all datasets, with each structure

having gold standard volumes separately defined using majority voting and the STAPLE

algorithm. Treatment plans were generated for each patient based on local protocol at the

time (VMAT treatment with 78 Gy prescription dose). The impact contouring variations

between observer and gold standard contours had on treatment efficacy was investigated

using various radiobiological metrics such as tissue control probability (TCP), normal

tissue complication probability (NTCP), and equivalent uniform dose (EUD).

No significant correlations between any contouring variation metrics and all simulated

treatment outcome measures were observed for CTV contours. This was attributed to

the uniform dose distribution generated by the treatment plans for the PTV, resulting

in no substantial dose discrepancies across the CTV for all observer’s contours. Margins

utilised to generate PTV for prostate radiotherapy are therefore sufficient to account for

inter-observer contouring variation of the CTV. Volume similarity significantly correlated


with simulated treatment outcome for PTV across both patient datasets, with TCPPoisson

(ρ = 0.57, 0.65), TCPLogit (ρ = 0..39, 0.62), and EUD (ρ = 0.43, 0.61) all statistically

significant. PTV volume similarity was also shown to significantly correlate with rectum

NTCP (ρ = 0.0.33, 0.48), highlighting that contouring variation of the target volumes

can negatively impact toxicity in neighbouring organs-at-risk (OARs). Most notably, the

Dice Similarity Coefficient (DSC) and Hausdorff Distance both displayed no significant

correlation with any radiobiological metric. These contouring metrics are highly utilised

within the literature [25], however each fails to differentiate whether an observer’s contour

is over- or under-contoured with respect to the gold standard. As these two situations will

result in substantially different dose distributions delivered to the gold standard volume,

these metrics alone cannot be used to evaluate the dosimetric impact of contouring vari-

ability. Consequently, it is recommended by this work that volume similarity is utilised

along with these metrics during future contouring variation studies.

This ethos was adopted within Chapter 4, where multiple metrics were utilised to assess

contouring variations between 13 observers on 5 male pelvic datasets. Structures were

contoured on both CT and T2 weighted MRI scans, and included both standard clinical

structures as well as neighbouring OARs of growing interest in radiotherapy studies [34,

403, 404]. While studies had begun investigating inter-observer contouring variability for

some of these structures, no study had looked at the breadth of structures investigated

here. Additionally, the impact of contouring variability when all structures were contoured

simultaneously is often neglected in these studies, despite this being what would occur

clinically. Inter-observer contouring variation of structures contoured on CT were found

to be comparable to values within the literature [28], while CTV contoured on T2-weighted

MRI showed excellent inter-observer agreement (median DSC = 0.88). However, many of

the emerging structures contoured on MRI showed far more variation between observers,

with structures such as the left and right neurovascular bundle, membranous urethra and

trigone showing significant inter-observer variation (median DSC = 0.16, 0.15, 0.41, and

0.44 respectively).

It was found that the relative experience of observers did not impact results, with the

variation in contouring between the more experienced observers as substantial as those

with less experience contouring these structures. However, experience in contouring the


emerging structures was consistently low for all observers, and was attributed as the most

probable cause of variability. This is a salient point, with studies investigating dose de-

livered to these structures often failing to adequately address these uncertainties. Clinical

trials require the true volume to be contoured for findings to be statistically significant,

however as this is rarely known all clinical trials will have an associated uncertainty at

this early stage within the radiotherapy workflow. Consequently, it was a recommendation

from this work that all radiotherapy clinical trials must provide some details regarding the

quality assurance of each structure’s contours.

Additionally, automated contouring methods such as atlases utilise a priori clinical infor-

mation to automatically contour patient datasets. It is a well-known maxim in computer

science that “rubbish in equals rubbish out”, with this statement particularly true for

atlas-based contouring. Multiple observer’s contours are utilised when developing an at-

las, which due to resource demands will usually come from observer’s with a wide range

of experience. The variability in contours that comprise the atlas will therefore place a

cap on the quality of the automatically generated contours, no matter the accuracy of the

registration. As this study shows, this variation is quite significant for many structures.

Consequently, a thorough investigation of inter-observer contouring variability must be

incorporated as a first step during atlas development.

7.3 Contouring variability within clinical trial datasets

Chapter 5 utilised an atlas to retrospectively contour CTVAtlas on the RADAR, RT01, and

CHHiP prostate radiotherapy clinical trial datasets. Contours within the atlas originated

from the inter-observer study of Chapter 4, while atlas development and validation was

performed by Kennedy et al [397]. Following the automatic contouring of CTVAtlas on all

clinical trial patients, variations between these and the original contours from the clinical

trials (CTVManual) were assessed using vector mappings. Vectors propagating from the

centroid of CTVManual were sampled every 6 degrees in both polar coordinates (azimuthal

and elevation), allowing for the Distance-to-Agreement (DtA) between contours to be

measured across the entire CTV surface. This method allowed for the spatial regions

where significant contouring variations occurred to be identified, providing information


that is lost when a global metric that attributes a single value to describe contouring

variation is used.

For RADAR patients CTVManual were found to be significantly under-contoured in the

vicinity of the bladder with respect to CTVAtlas, while simultaneously being over-contoured

posteriorly towards the rectum. Additionally, both RADAR and RT01 CTVManual con-

tours overextended inferiorly in the direction of the penile bulb. RT01 and CHHiP patient’s

had CTVManual contours that were under-contoured near the rectum, while CHHiP pa-

tient’s CTVManual contours were over-contoured in the vicinity of the bladder. It was

noted that significant differences existed between the three clinical trials with respect to

dose prescriptions, contouring protocols, and treatment techniques. Each of these differ-

ences could be responsible for the differences in contouring variations observed between the

trials. This highlights the requirement for assessing inter-observer contouring variability

during every clinical trial, as variations will not be consistent even if the same treatment

site is investigated.

By spatially analysing contouring variations across these clinical trials, it was possible to

assess the clinical impact contouring variations in specific regions had on treatment efficacy.

For every vector generated, a DtA cut-point was identified such that the largest difference

in a particular treatment outcome was observed between patients dichotomised by this

DtA. This cut-point was unique not only for each vector, but also for each trial as well as

each treatment outcome measure investigated. Within this study, the treatment outcomes

considered were patient death, local progression, and PSA progression. Hazard Ratios were

calculated that indicated the associated risk of a patient observing each of these outcomes,

dependent on whether the patient’s DtA between CTVManual and CTVAtlas along this

vector either exceeded or fell short of the calculated cut-point. This was analogous to the

original CTVManual being over- or under-contoured respectively in this region.

It was found that under-contouring of CTVManual in the vicinity of the bladder resulted

in a significant increase in risk of patient death (median HR = 0.57, 0.36, and 0.70 for

RADAR, RT01, CHHiP respectively) and PSA progression (median HR = 0.37, 0.56

for RT01, CHHiP). Additionally, insufficient contouring in the vicinity of the rectum for

RADAR patients resulted in an increased risk of PSA progression (median HR = 0.55).


Finally, insufficient contouring in the anterior-inferior orientation (towards the penile bulb)

resulted in significantly increased risk of local progression across all three clinical trials

(median HR = 0.37, 0.16, 0.29 for RADAR, RT01, CHHiP).

These results verify that the specific spatial regions where contouring variations occur

is an important factor when assessing risk of treatment failure following radiotherapy.

The most significant regions of contouring variations occur near where the CTV borders

neighbouring OARs, such as the bladder and rectum. The soft tissue boundaries can often

be difficult to distinguish, particularly on CT, with the resultant manual CTV contours

often being conservatively defined to reduce dose delivered to the neighbouring OARs. If

the tumour volume was inadequately irradiated, it can be easily reasoned that this would

result in the poorer patient outcomes observed in this study. Consequently, this work not

only identified the clinical impact of contouring variability within prostate radiotherapy

clinical trials, but also provided the framework for a complete spatial analysis during future

contouring studies.

7.4 Quality of automated treatment planning techniques be-

tween centres

Automated treatment planning is increasingly utilised to improve efficiency within the

radiotherapy workflow. The Pinnacle3 treatment planning system had recently introduced

the AutoPlanning™ (AP) module, a template-based method of automatically generating

treatment plans through iterative re-optimisation of specified planning goals. It had been

shown that AP treatment plans were non-inferior in quality compared to manual plans

across multiple treatment sites, while being generated in a significantly shorter time [39,

41, 42]. However, significant resources are required by a clinic to develop AP configurations

that are specific to local treatment protocols, which can be a burden for centre’s lacking

the resources to complete this process. Chapter 6 investigated the ability to adapt AP

configurations developed by one clinic to meet the treatment protocols of other clinics,

with the robustness of these configurations and quality of the generated plans assessed.


Three centres (one Australian, two European) shared treatment protocols and training

datasets, with the intention of each centre modifying their local intact prostate AP con-

figurations to meet the treatment protocols of the other two centres. Once modified AP

configurations were developed for all centre’s protocols, validation datasets of 10 patients

were distributed by each centre. Treatment plans generated using the modified AP con-

figurations were compared against the original clinic’s treatment plan via DVH analysis

and protocol compliance. When assessing plans generated using the modified AP configu-

rations, only a single constraint and high priority deviation was recorded. These occurred

on a difficult prostate plan, whereby significant overlap between the PTV and rectum oc-

curred, and where the original clinically accepted plan also displayed these same violations.

Consequently, all modified AP configurations were able to generate clinically acceptable

treatment plans.

This study verified that AP configurations could be shared between clinics, being robust

enough to adapted to meet local treatment planning protocols. Smaller centre’s lacking the

resources to develop and validate individual AP configurations could therefore utilise AP

configurations previously developed by other centres, reducing the workload on therapists

while improving the quality of treatment. Additionally, clinical trials require specific

treatment protocols to be met by participating centres. Centres that utilise the Pinnacle3

AutoPlanning™ module could therefore adapt AP configurations to meet these clinical

trial protocols, reducing the impact of planning variation that is found within clinical

trials.

7.5 Future Work

The methods developed within this thesis provide opportunities for many additional future

investigations. Examples of potential future work following on from these studies include:

1. Further investigations correlating contouring variations metric with treatment out-

come for prostate radiotherapy when treated with advanced treatment techniques,

such as Stereotactic Body Radiotherapy (SBRT),


2. The utilisation of vector mappings to assess the impact of CTV contouring variability

on recorded urinary and toxicity outcomes within the RADAR, RT01, and CHHiP

clinical trial datasets,

3. Automatic contouring of additional male pelvic structures that are within the atlas

upon these clinical trial datasets, allowing for subsequent investigations of the clinical

impact of dose to these structures on treatment and toxicity, and

4. An investigation into the utilisation of AP configurations to improve treatment plan-

ning quality during a benchmarking exercise for a multicentre clinical trial.

Chapter 3 identified contouring variation metrics that significantly correlated with simu-

lated treatment outcome for prostate patients treated utilising Volumetric Modulated Arc

Therapy (VMAT). The primary findings were that the volumetric metrics, such as vol-

ume similarity, correlated significantly with outcome while boundary and overlap metrics

did not. This was due to the metric’s ability to distinguish target volume contours that

produced inadequate dose distributions to the gold standard volume. However, modern

treatment techniques such as Stereotactic Body Radiotherapy (SBRT) are increasingly

being investigated as alternatives to VMAT [202, 405]. These treatments deliver highly

ablative doses to the target volume, producing significantly different dose distributions

compared to VMAT [406]. This difference in treatment delivery could result in the ob-

served correlations no longer being statistically significant. Additionally, the margins used

to generate PTV many no longer be sufficient to ensure CTV contouring variability did

not impact the resultant CTV dose distribution. Consequently, further assessment in-

vestigating correlations between contouring metrics and treatment outcomes should be

undertaken for prostate patient treated with SBRT, as well as any other new treatment

technique.

The work presented in Chapter 5 displayed a novel method for investigating correlations

between contouring variations and treatment failure. Treatment failure was defined as pa-

tient death, local progression, or PSA progression sometime during 5 to 10 year follow-up

for the RADAR, RT01, and CHHiP clinical trials. It was found that contouring variabil-

ity in specific spatial regions resulted in significant increases in risk of observing these


outcomes. Further work could expand on this by utilising both the prostate and seminal

vesicle contours within the atlas to define the Clinical Target Volume. This would allow

for the inclusion of a greater portion of patients within the three datasets, improving the

statistical power and findings from this analysis. Increased statistical power could also be

achieved through the introduction of permutation testing methods within the methodol-

ogy to account for multiple testing issues. Additionally, the contouring similarity metrics

investigated in chapter 3 could also be brought into the analysis as well to complement

the analysis.

Additional follow-up data is also available from these clinical trial datasets. Urinary tox-

icities such as haematuria and dysuria were recorded for each patient, along with rectal

toxicities such as rectal bleeding, stool frequency, and tenesmus. As a treatment plan’s

dose distribution is inherently dependent on the contours, variability when defining the

target volume contour may result in additional dose being delivered to neighbouring OARs

such as the bladder and rectum. This could result in an increased risk in developing the

aforementioned toxicities. Investigating the correlations between CTV contouring vari-

ability and each urinary and rectal toxicity would improve our understanding concerning

how contouring uncertainties in prostate radiotherapy affects all aspects of treatment ef-

ficacy. The information gleaned from this work, combined with the results from Chapter

5, could then allow us to define target volume boundaries for prostate radiotherapy based

on pathological boundaries (i.e., defining the contour such that there is the smallest risk

of treatment failure or developing toxicities), as well as anatomical information.

The atlas utilised in Chapter 5 was derived from the multiple observer’s contours upon

the five patient datasets within Chapter 4. Significant results were obtained following

investigations utilising the CT-defined CTVAtlas contour, however this was only a single

structure contained within the atlas. As well as the MR-defined CTV, an additional

14 structures and OARs have each been propagated upon all patient datasets within the

RADAR, RT01, and CHHiP clinical trials. Future studies could investigate the known dose

distributions that were planned for each patient, and correlate these doses delivered to each

of the automatically contoured structures with the recorded patient outcomes. Utilisation

of this rich dataset could therefore provide further information not only on potential dose


constraints for these emerging structures, but also provide evidence regarding the clinical

benefit of utilising MRI to define the CTV boundary for prostate radiotherapy.

Finally, the assessment of AP configuration robustness in Chapter 6 showed that these

configurations were able to be adapted to meet other centre’s treatment planning proto-

cols in a retrospective assessment. This feasibility study was illuminating, however the

greatest potential benefit of this work would be during a large multicentre clinical trial,

where multiple participating centres are required to generate treatment plans that meet

the clinical trial protocols. Most clinical trials will incorporate a benchmarking exercise

at the beginning of the study to assess the capability of centre’s to generate treatment

plans of sufficient quality [283, 286]. Even so, treatment planning variability is prominent

in clinical trials, and has been shown to reduce both the effectiveness of treatment, as well

as the statistical power of the clinical trial [11]. By introducing an AP configuration that

is specific to the clinical trial’s protocol, it would be easily manageable to distribute this

configuration to all participating centres that utilise the Pinnacle3 AutoPlanning™ mod-

ule, removing the potential variability in treatment planning between these centres. An

assessment on the effectiveness of sharing a single AP configuration in a manner analogous

to Chapter 6, but in the context of a clinical trial benchmarking exercise, would signifi-

cantly reduce treatment planning variability while improving the quality and effectiveness

of the clinical trial.

7.6 Conclusion

The work compiled within this thesis investigated advanced analysis techniques which,

when combined with automated methods of contouring and treatment planning, can be

utilised to improve quality assurance in radiotherapy clinical trials. It was shown in this

thesis that the contouring metric volume similarity significantly correlated with simu-

lated treatment outcome for prostate radiotherapy, while commonly utilised overlap and

boundary contouring metrics did not. These metrics were utilised during an assessment of

inter-observer contouring variation for multiple male pelvic structures. Emerging struc-

tures were found to have significant variation between observer contours. Vector mappings

were incorporated to provide a spatial analysis of contouring variation between automatic


atlas CTV and original trial CTV. This method identified of specific regions surrounding

the CTV where contouring variations resulted in significantly increased risk of treatment

failure. Finally, when investigating automated treatment methods, it was found that mod-

ified AP configurations were successfully able to generate clinically acceptable treatment

plans for all treatment protocols.

All of these investigations have focussed on what are often regarded as the greatest sources

of uncertainty within radiotherapy. As future clinical trials in radiotherapy strive to find

key breakthroughs that will significantly improve quality of care for cancer patients, it is

essential to understand the uncertainties associated with all steps within the radiotherapy

workflow. It is hoped that the work presented herein will assist in this manner, with the

methods and analysis techniques utilised by future clinical trials in conjunction with the

automated methods of contouring and treatment planning explored in this work.

Bibliography

[1] Australian Institute of Health and Welfare (AIHW). Australian cancer incidence

and mortality (acim) books: Prostate cancer, 2017. URL http://www.aihw.gov.

au/acim-books.

[2] OpenStax College. Anatomy & Physiology. OpenStax College, 2013. URL http:

//cnx.org/content/col11496/latest/.

[3] Patrick W McLaughlin, Sara Troyer, Sally Berri, Vrinda Narayana, Amichay

Meirowitz, Peter L Roberson, and James Montie. Functional anatomy of the

prostate: implications for treatment planning. International Journal of Radiation

Oncology* Biology* Physics, 63(2):479–491, 2005. ISSN 0360-3016.

[4] Jonathan I Epstein. An update of the gleason grading system. The Journal of

urology, 183(2):433–440, 2010. ISSN 0022-5347.

[5] Matthew R Cooperberg and Peter R Carroll. Trends in management for patients

with localized prostate cancer, 1990-2013. JAMA, 314(1):80–82, 2015. ISSN 0098-

7484.

[6] Ian F Tannock, Richard P Hill, Robert G Bristow, and Lea Harrington. The Basic

Science of Oncology, 5th Edition. Boston: MacGraw Hill, 2013. ISBN 0071387749.

[7] Nagalingam Suntharalingam, Ervin B Podgorsak, and Jolyon H Hendry. Basic ra-

diobiology. Radiation oncology physics: A handbook for teachers and students, 2005.

[8] H Rodney Withers. The four r’s of radiotherapy. Adv Radiat Biol, 5(3):241–271,

1975.

232

http://www.aihw.gov.au/acim-books


http://cnx.org/content/col11496/latest/

http://cnx.org/content/col11496/latest/

References 233

[9] Michael J Zelefsky, Zvi Fuks, and Steven A Leibel. Intensity-modulated radiation

therapy for prostate cancer. In Seminars in radiation oncology, volume 12, pages

229–237. Elsevier, 2002. ISBN 1053-4296.

[10] Eckhart Dühmke, Jeremy Franklin, Michael Pfreundschuh, Susanne Sehlen, Norman

Willich, Ursula Rühl, Rolf-Peter Müller, Peter Lukas, Anton Atzinger, and Ursula

Paulus. Low-dose radiation is sufficient for the noninvolved extended-field treatment

in favorable early-stage hodgkin’s disease: long-term results of a randomized trial

of radiotherapy alone. Journal of clinical oncology, 19(11):2905–2914, 2001. ISSN

0732-183X. doi: https://doi.org/10.1200/JCO.2001.19.11.2905.

[11] Lester J Peters, Brian O’Sullivan, Jordi Giralt, Thomas J Fitzgerald, Andy Trotti,

Jacques Bernier, Jean Bourhis, Kally Yuen, Richard Fisher, and Danny Rischin.

Critical impact of radiotherapy protocol compliance and quality in the treatment

of advanced head and neck cancer: results from trog 02.02. Journal of Clinical

Oncology, 28(18):2996–3001, 2010. ISSN 0732-183X. doi: https://doi.org/10.1200/

JCO.2009.27.4498.

[12] Jacqueline E Livsey, James P Wylie, Ric Swindell, Vincent S Khoo, Richard A

Cowan, and John P Logue. Do differences in target volume definition in prostate

cancer lead to clinically relevant differences in normal tissue toxicity? International

Journal of Radiation Oncology* Biology* Physics, 60(4):1076–1081, 2004. ISSN

0360-3016.

[13] James W Denham, David Joseph, David S Lamb, Nigel A Spry, Gillian Duchesne,

John Matthews, Chris Atkinson, Keen-Hun Tai, David Christie, and Lizbeth Kenny.

Short-term androgen suppression and radiotherapy versus intermediate-term andro-

gen suppression and radiotherapy, with or without zoledronic acid, in men with

locally advanced prostate cancer (trog 03.04 radar): 10-year results from a ran-

domised, phase 3, factorial trial. The Lancet Oncology, 20(2):267–281, 2019. ISSN

1470-2045.

[14] Peressutti Devis and Schipaanboord Bas. Overview of atlas-based segmenta-

tion approaches. figshare, 2018. doi: 10.6084/m9.figshare.6809597.v1. URL

References 234

https://figshare.com/articles/Overview_of_atlas-based_segmentation_

approaches/6809597.

[15] Juan Eugenio Iglesias and Mert R Sabuncu. Multi-atlas segmentation of biomedical

images: a survey. Medical image analysis, 24(1):205–219, 2015. ISSN 1361-8415.

[16] Oscar Acosta, Jason Dowling, Gael Drean, Antoine Simon, Renaud De Crevoisier,

and Pascal Haigron. Multi-atlas-based segmentation of pelvic structures from CT

scans for planning in prostate cancer radiotherapy, pages 623–656. Springer, 2014.

ISBN 1461484979.

[17] Peter H Westfall and S Stanley Young. Resampling-based multiple testing: Examples

and methods for p-value adjustment, volume 279. John Wiley & Sons, 1993. ISBN

0471557617.

[18] Jonathan I Epstein, William C Allsbrook Jr, Mahul B Amin, Lars L Egevad, and

ISUP Grading Committee. The 2005 international society of urological pathology

(isup) consensus conference on gleason grading of prostatic carcinoma. The American

journal of surgical pathology, 29(9):1228–1242, 2005. ISSN 0147-5185.

[19] Damien C Weber, Philip MP Poortmans, Coen W Hurkmans, Edwin Aird, Akos

Gulyban, and Alysa Fairchild. Quality assurance for prospective eortc radiation

oncology trials: the challenges of advanced technology in a multicenter international

setting. Radiotherapy and Oncology, 100(1):150–156, 2011. ISSN 0167-8140.

[20] Australian Institute of Health and Welfare. Cancer in australia 2019.

Report, AIHW, 2019. URL https://www.aihw.gov.au/reports/cancer/

cancer-in-australia-2019.

[21] Mary Weinstein Dunn and Meredith Wallace Kazer. Prostate cancer overview. In

Seminars in oncology nursing, volume 27, pages 241–250. Elsevier, 2011. ISBN

0749-2081.

[22] Stephen R Thompson, Geoff P Delaney, Susannah Jacob, Jesmin Shafiq, Karen

Wong, Timothy P Hanna, Gabriel S Gabriel, and Michael B Barton. Estimation of

the optimal utilisation rates of radical prostatectomy, external beam radiotherapy

https://figshare.com/articles/Overview_of_atlas-based_segmentation_approaches/6809597

https://figshare.com/articles/Overview_of_atlas-based_segmentation_approaches/6809597

https://www.aihw.gov.au/reports/cancer/cancer-in-australia-2019

https://www.aihw.gov.au/reports/cancer/cancer-in-australia-2019

References 235

and brachytherapy in the treatment of prostate cancer by a review of clinical practice

guidelines. Radiotherapy and Oncology, 2016. ISSN 0167-8140.

[23] Christopher H Crane, Kathryn Winter, William F Regine, Howard Safran, Tyvin A

Rich, Walter Curran, Robert A Wolff, and Christopher G Willett. Phase ii study of

bevacizumab with concurrent capecitabine and radiation followed by maintenance

gemcitabine and bevacizumab for locally advanced pancreatic cancer: Radiation

therapy oncology group rtog 0411. Journal of Clinical Oncology, 27(25):4096–4102,

2009. ISSN 0732-183X.

[24] CF Njeh. Tumor delineation: The weakest link in the search for accuracy in radio-

therapy. Journal of medical physics/Association of Medical Physicists of India, 33

(4):136, 2008.

[25] Shalini K Vinod, Michael G Jameson, Myo Min, and Lois C Holloway. Uncertainties

in volume delineation in radiation oncology: A systematic review and recommenda-

tions for future studies. Radiotherapy and Oncology, 121(2):169–179, 2016. ISSN

0167-8140.

[26] Michael G Jameson, Lois C Holloway, Philip J Vial, Shalini K Vinod, and Peter E

Metcalfe. A review of methods of analysis in contouring studies for radiation oncol-

ogy. Journal of medical imaging and radiation oncology, 54(5):401–410, 2010. ISSN

1754-9485.

[27] I Fotina, C Lütgendorf-Caucig, M Stock, R Pötter, and D Georg. Critical discus-

sion of evaluation parameters for inter-observer variability in target definition for

radiation therapy. Strahlentherapie und Onkologie, 188(2):160–167, 2012. ISSN

0179-7158.

[28] Gregory Sharp, Karl D Fritscher, Vladimir Pekar, Marta Peroni, Nadya Shusharina,

Harini Veeraraghavan, and Jinzhong Yang. Vision 20/20: perspectives on automated

image segmentation for radiotherapy. Medical physics, 41(5):050902, 2014. ISSN

0094-2405.

[29] Michael G Jameson, Shivani Kumar, Shalini K Vinod, Peter E Metcalfe, and Lois C

Holloway. Correlation of contouring variation with modeled outcome for conformal

References 236

non-small cell lung cancer radiotherapy. Radiotherapy and Oncology, 112(3):332–336,

2014. ISSN 0167-8140.

[30] William J Beasley, Alan McWilliam, Adam Aitkenhead, Ranald I Mackay, and

Carl G Rowbottom. The suitability of common metrics for assessing parotid and

larynx autosegmentation accuracy. Journal of Applied Clinical Medical Physics, 17

(2), 2016. ISSN 1526-9914.

[31] Tobias Heimann and Hans-Peter Meinzer. Statistical shape models for 3d medical

image segmentation: a review. Medical image analysis, 13(4):543–563, 2009. ISSN

1361-8415.

[32] Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. nature, 521

(7553):436, 2015. ISSN 1476-4687.

[33] Mack Roach, Jiho Nam, Giovanna Gagliardi, Issam El Naqa, Joseph O Deasy, and

Lawrence B Marks. Radiation dose–volume effects and the penile bulb. International

Journal of Radiation Oncology* Biology* Physics, 76(3):S130–S134, 2010. ISSN

0360-3016.

[34] SL Gulliford, S Ghose, MA Ebert, A Kennedy, J Dowling, J Mitra, DJ Joseph,

and JW Denham. Radiotherapy dose-distribution to the perirectal fat space (prs) is

related to gastrointestinal control-related complications. Clinical and Translational

Radiation Oncology, 7:62–70, 2017. ISSN 2405-6308.

[35] Robert Jan Smeenk, Aswin L Hoffmann, Wim PM Hopman, Emile NJ Th van Lin,

and Johannes HAM Kaanders. Dose-effect relationships for individual pelvic floor

muscles and anorectal complaints after prostate radiotherapy. International Journal

of Radiation Oncology* Biology* Physics, 83(2):636–644, 2012. ISSN 0360-3016.

[36] Benjamin E Nelms, Greg Robinson, Jay Markham, Kyle Velasco, Steve Boyd,

Sharath Narayan, James Wheeler, and Mark L Sobczak. Variation in external

beam treatment plan quality: an inter-institutional study of planners and plan-

ning systems. Practical radiation oncology, 2(4):296–305, 2012. ISSN 1879-8500.

doi: https://doi.org/10.1016/j.prro.2011.11.012.

References 237

[37] Sean L Berry, Amanda Boczkowski, Rongtao Ma, James Mechalakos, and Margie

Hunt. Interobserver variability in radiation therapy plan output: Results of a single-

institution study. Practical radiation oncology, 6(6):442–449, 2016. ISSN 1879-8500.

doi: https://doi.org/10.1016/j.prro.2016.04.005.

[38] Ilma Xhaferllari, Eugene Wong, Karl Bzdusek, Michael Lock, and Jeff Z Chen.

Automated imrt planning with regional optimization using planning scripts. Journal

of applied clinical medical physics, 14(1):176–191, 2013. ISSN 1526-9914. doi: https:

//doi.org/10.1120/jacmp.v14i1.4052.

[39] Christian Rønn Hansen, Anders Bertelsen, Irene Hazell, Ruta Zukauskaite, Niels

Gyldenkerne, Jørgen Johansen, Jesper G Eriksen, and Carsten Brink. Automatic

treatment planning improves the clinical quality of head and neck cancer treatment

plans. Clinical and Translational Radiation Oncology, 1:2–8, 2016. ISSN 2405-6308.

doi: https://doi.org/10.1016/j.ctro.2016.08.001.

[40] Christian Rønn Hansen, Morten Nielsen, Anders Smedegaard Bertelsen, Irene Hazell,

Eva Holtved, Ruta Zukauskaite, Jon Kroll Bjerregaard, Carsten Brink, and Uffe

Bernchou. Automatic treatment planning facilitates fast generation of high-quality

treatment plans for esophageal cancer. Acta Oncologica, 56(11):1495–1500, 2017.

ISSN 0284-186X. doi: https://doi.org/10.1080/0284186X.2017.1349928.

[41] Irene Hazell, Karl Bzdusek, Prashant Kumar, Christian R Hansen, Anders Bertelsen,

Jesper G Eriksen, Jørgen Johansen, and Carsten Brink. Automatic planning of head

and neck treatment plans. Journal of Applied Clinical Medical Physics, 17(1), 2016.

ISSN 1526-9914. doi: https://doi.org/10.1120/jacmp.v17i1.5901.

[42] Kanabu Nawa, Akihiro Haga, Akihiro Nomoto, Raniel A Sarmiento, Kenshiro Shi-

raishi, Hideomi Yamashita, and Keiichi Nakagawa. Evaluation of a commercial au-

tomatic treatment planning system for prostate cancers. Medical Dosimetry, 2017.

ISSN 0958-3947. doi: https://doi.org/10.1016/j.meddos.2017.03.004.

[43] Freddie Bray, Jacques Ferlay, Isabelle Soerjomataram, Rebecca L Siegel, Lindsey A

Torre, and Ahmedin Jemal. Global cancer statistics 2018: Globocan estimates of

References 238

incidence and mortality worldwide for 36 cancers in 185 countries. CA: a cancer

journal for clinicians, 68(6):394–424, 2018. ISSN 0007-9235.

[44] Lindsey A Torre, Freddie Bray, Rebecca L Siegel, Jacques Ferlay, Joannie

Lortet‐Tieulent, and Ahmedin Jemal. Global cancer statistics, 2012. CA: a can-

cer journal for clinicians, 65(2):87–108, 2015. ISSN 1542-4863.

[45] Melissa M Center, Ahmedin Jemal, Joannie Lortet-Tieulent, Elizabeth Ward,

Jacques Ferlay, Otis Brawley, and Freddie Bray. International variation in prostate

cancer incidence and mortality rates. European urology, 61(6):1079–1092, 2012. ISSN

0302-2838.

[46] Muhammad Naeem Bashir. Epidemiology of prostate cancer. Asian Pac J Cancer

Prev, 16:5137–5141, 2015.

[47] Australian Institute of Health and Welfare (AIHW). Australian cancer incidence

and mortality (acim) books: Prostate cancer, 2015. URL http://www.aihw.gov.

au/acim-books.

[48] Peter D Baade, Danny R Youlden, Susanna M Cramb, Jeff Dunn, and Robert A

Gardiner. Epidemiology of prostate cancer in the asia-pacific region. Prostate inter-

national, 1(2):47–58, 2013. ISSN 2287-8882.

[49] Jacques Ferlay, Isabelle Soerjomataram, Rajesh Dikshit, Sultan Eser, Colin Mathers,

Marise Rebelo, Donald Maxwell Parkin, David Forman, and Freddie Bray. Cancer

incidence and mortality worldwide: sources, methods and major patterns in globocan

2012. International journal of cancer, 136(5):E359–E386, 2015. ISSN 1097-0215.

[50] Alan Pollack and Gunar K Zagars. External beam radiotherapy dose response of

prostate cancer. International Journal of Radiation Oncology* Biology* Physics, 39

(5):1011–1018, 1997. ISSN 0360-3016.

[51] John E McNeal. The zonal anatomy of the prostate. The prostate, 2(1):35–49, 1981.

ISSN 1097-0045.

[52] WJ Catalona, JP Richie, FR Ahmann, MA Hudson, PT Scardino, RC Flanigan,

JB Dekernion, TL Ratliff, LR Kavoussi, and BL Dalkin. Comparison of digital rectal



References 239

examination and serum prostate specific antigen in the early detection of prostate

cancer: results of a multicenter clinical trial of 6,630 men. The Journal of urology,

151(5):1283–1290, 1994. ISSN 0022-5347.

[53] National Comprehensive Cancer Network. Nccn clinical practice guidelines in on-

cology (nccn guidelines): Prostate cancer. National Comprehensive Cancer Network

(NCCN), 2015.

[54] Alan W Partin, Michael W Kattan, Eric NP Subong, Patrick CWalsh, Kirk J Wojno,

Joseph E Oesterling, Peter T Scardino, and JD Pearson. Combination of prostate-

specific antigen, clinical stage, and gleason score to predict pathological stage of

localized prostate cancer: a multi-institutional update. Jama, 277(18):1445–1451,

1997. ISSN 0098-7484.

[55] Donald F Gleason. Classification of prostatic carcinomas. Cancer chemotherapy

reports. Part 1, 50(3):125–128, 1966. ISSN 0069-0112.

[56] FH Schröder, P Hermanek, L Denis, WR Fair, MK Gospodarowicz, and

M Pavone‐Macaluso. The tnm classification of prostate cancer. The Prostate, 21

(S4):129–138, 1992. ISSN 1097-0045.

[57] Liang Cheng, Rodolfo Montironi, David G Bostwick, Antonio Lopez‐Beltran, and

Daniel M Berney. Staging of prostate cancer. Histopathology, 60(1):87–117, 2012.

ISSN 1365-2559.

[58] Ian M Thompson, Richard K Valicenti, Peter Albertsen, Brian J Davis, S Larry

Goldenberg, Carol Hahn, Eric Klein, Jeff Michalski, Mack Roach, and Oliver Sartor.

Adjuvant and salvage radiotherapy after prostatectomy: Aua/astro guideline. The

Journal of urology, 190(2):441–449, 2013. ISSN 0022-5347.

[59] Axel Heidenreich, Patrick J Bastian, Joaquim Bellmunt, Michel Bolla, Steven Joniau,

Theodor van der Kwast, Malcolm Mason, Vsevolod Matveev, Thomas Wiegel, and

F Zattoni. Eau guidelines on prostate cancer. part 1: screening, diagnosis, and

local treatment with curative intent—update 2013. European urology, 65(1):124–

137, 2014. ISSN 0302-2838.

References 240

[60] Australian Cancer Network Management of Metastatic Prostate Cancer Work-

ing Party. Clinical Practice Guidelines for the Management of Locally Advanced

and Metastatic Prostate Cancer. Cancer Council Australia and Australian Cancer

Network, 2010.

[61] George Miller MacKee. X-rays and Radium in the Treatment of Diseases of the Skin.

Lea& Febiger, 1921.

[62] Manuel Lederman. The early history of radiotherapy: 1895–1939. International

Journal of Radiation Oncology* Biology* Physics, 7(5):639–648, 1981. ISSN 0360-

3016.

[63] Michael B Barton, Susannah Jacob, Jesmin Shafiq, Karen Wong, Stephen R Thomp-

son, Timothy P Hanna, and Geoff P Delaney. Estimating the demand for radiother-

apy from the evidence: a review of changes from 2003 to 2012. Radiotherapy and

Oncology, 112(1):140–144, 2014. ISSN 0167-8140.

[64] Ervin B Podgorsak. Radiation Oncology Physics: A handbook for teachers and

students, volume 657. International Atomic Energy Agency, 2005.

[65] Piotr Wojcieszek and Brygida Białas. Prostate cancer brachytherapy: guidelines

overview. Journal of contemporary brachytherapy, 4(2):116, 2012.

[66] Daniel Ash, Anthony Flynn, Jan Battermann, Theodorous de Reijke, Paulo

Lavagnini, and Leo Blank. Estro/eau/eortc recommendations on permanent seed

implantation for localized prostate cancer. Radiotherapy and Oncology, 57(3):315–

321, 2000. ISSN 0167-8140.

[67] A Challapalli, E Jones, C Harvey, GO Hellawell, and SA Mangar. High dose rate

prostate brachytherapy: an overview of the rationale, experience and emerging ap-

plications in the treatment of prostate cancer. The British journal of radiology,

2012.

[68] Juanita Crook, Himu Lukka, Laurence Klotz, Nancy Bestic, and Mary Johnston.

Systematic overview of the evidence for brachytherapy in clinically localized prostate

cancer. Canadian Medical Association Journal, 164(7):975–981, 2001. ISSN 0820-

3946.

References 241

[69] Tom Pickles, Mira Keyes, W James Morris, Prostate Outcomes, and Provincial

Prostate Brachytherapy Program. Brachytherapy or conformal external radiotherapy

for prostate cancer: a single-institution matched-pair analysis. International Journal


[70] Al V Taira, Gregory S Merrick, Wayne M Butler, Robert W Galbreath, Jonathan

Lief, Edward Adamovich, and Kent E Wallner. Long-term outcome for clinically

localized prostate cancer treated with permanent interstitial brachytherapy. Inter-

national Journal of Radiation Oncology* Biology* Physics, 79(5):1336–1342, 2011.

ISSN 0360-3016.

[71] Daniel E Spratt, Zachary S Zumsteg, Pirus Ghadjar, Marisa A Kollmeier, Xin Pei,

Gilad Cohen, William Polkinghorn, Yoshiya Yamada, and Michael J Zelefsky. Com-

parison of high‐dose (86.4 gy) imrt vs combined brachytherapy plus imrt for in-

termediate‐risk prostate cancer. BJU international, 114(3):360–367, 2014. ISSN

1464-410X.

[72] Ruth Etzioni, David F Penson, Julie M Legler, Dante di Tommaso, Rob Boer, Pe-

ter H Gann, and Eric J Feuer. Overdiagnosis due to prostate-specific antigen screen-

ing: lessons from us prostate cancer incidence trends. Journal of the National Cancer

Institute, 94(13):981–990, 2002. ISSN 0027-8874.

[73] Gerrit Draisma, Rob Boer, Suzie J Otto, Ingrid W van der Cruijsen, Ronald AM

Damhuis, Fritz H Schröder, and Harry J de Koning. Lead times and overdetection

due to prostate-specific antigen screening: estimates from the european randomized

study of screening for prostate cancer. Journal of the National Cancer Institute, 95

(12):868–878, 2003. ISSN 0027-8874.

[74] H Gilbert Welch and William C Black. Overdiagnosis in cancer. Journal of the

National Cancer Institute, 102(9):605–613, 2010. ISSN 0027-8874.

[75] Marc A Dall’Era, Peter C Albertsen, Christopher Bangma, Peter R Carroll, H Bal-

lentine Carter, Matthew R Cooperberg, Stephen J Freedland, Laurence H Klotz,

Christopher Parker, and Mark S Soloway. Active surveillance for prostate cancer:

References 242

a systematic review of the literature. European urology, 62(6):976–983, 2012. ISSN

0302-2838.

[76] Viacheslav Iremashvili, Lisét Pelaez, Murugesan Manoharan, Mercé Jorda, Daniel L

Rosenberg, and Mark S Soloway. Pathologic prostate cancer characteristics in pa-

tients eligible for active surveillance: a head-to-head comparison of contemporary

protocols. European urology, 62(3):462–468, 2012. ISSN 0302-2838.

[77] Laurence Klotz. Active surveillance for prostate cancer: a review. Current urology

reports, 11(3):165–171, 2010. ISSN 1527-2737.

[78] Chris H Bangma, Meelan Bul, Theo H van der Kwast, Tom Pickles, Ida J Korfage,

Caroline M Hoeks, Ewout W Steyerberg, Guido Jenster, Michael W Kattan, and

Lara Bellardita. Active surveillance for low-risk prostate cancer. Critical reviews in

oncology/hematology, 85(3):295–302, 2013. ISSN 1040-8428.

[79] Laurence Klotz, Danny Vesprini, Perakaa Sethukavalan, Vibhuti Jethava, Liying

Zhang, Suneil Jain, Toshihiro Yamamoto, Alexandre Mamedov, and Andrew Loblaw.

Long-term follow-up of a large active surveillance cohort of patients with prostate

cancer. Journal of Clinical Oncology, page JCO. 2014.55. 1192, 2014. ISSN 0732-

183X.

[80] Matthew R Cooperberg. Long-term active surveillance for prostate cancer: Answers

and questions. Journal of Clinical Oncology, 33(3):238–240, 2015. ISSN 0732-183X.

[81] Vincent Fradet, John Kurhanewicz, Janet E Cowan, Alexander Karl, Fergus V Coak-

ley, Katsuto Shinohara, and Peter R Carroll. Prostate cancer managed with active

surveillance: Role of anatomic mr imaging and mr spectroscopic imaging1. Radiol-

ogy, 2010.

[82] Anna Bill-Axelson, Lars Holmberg, Hans Garmo, Jennifer R Rider, Kimmo Taari,

Christer Busch, Stig Nordling, Michael Häggman, Swen-Olof Andersson, and Anders

Spångberg. Radical prostatectomy or watchful waiting in early prostate cancer. New

England Journal of Medicine, 370(10):932–942, 2014. ISSN 0028-4793.

References 243

[83] David P Byar and FK Mostofi. Carcinoma of the prostate: prognostic evaluation of

certain pathologic features in 208 radical prostatectomies. Cancer, 30(1):5–13, 1972.

ISSN 1097-0142.

[84] Patrick C Walsh and Herbert Lepor. The role of radical prostatectomy in the man-

agement of prostatic cancer. Cancer, 60(S3):526–537, 1987. ISSN 1097-0142.

[85] Jochen Walz, Arthur L Burnett, Anthony J Costello, James A Eastham, Markus

Graefen, Bertrand Guillonneau, Mani Menon, Francesco Montorsi, Robert P Myers,

and Bernardo Rocco. A critical analysis of the current knowledge of surgical anatomy

related to optimization of cancer control and preservation of continence and erection

in candidates for radical prostatectomy. European urology, 57(2):179–192, 2010. ISSN

0302-2838.

[86] Ahmed Magheli and Arthur L Burnett. Erectile dysfunction following prostatectomy:

prevention and treatment. Nature Reviews Urology, 6(8):415–427, 2009. ISSN 1759-

4812.

[87] Ashutosh Tewari, James O Peabody, Melissa Fischer, Richard Sarle, Guy Vallancien,

V Delmas, Mazen Hassan, Aditya Bansal, Ashok K Hemal, and Bertrand Guillon-

neau. An operative and anatomic study to help in nerve sparing during laparoscopic

and robotic radical prostatectomy. European urology, 43(5):444–454, 2003. ISSN

0302-2838.

[88] Ingolf Türk, Serdar Deger, Björn Winkelmann, Bernd Schönberger, and Stefan A

Loening. Laparoscopic radical prostatectomy. European urology, 40(1):46–53, 2001.

ISSN 1421-993X.

[89] Ashutosh Tewari, Prasanna Sooriakumaran, Daniel A Bloch, Usha Seshadri-

Kreaden, April E Hebert, and Peter Wiklund. Positive surgical margin and pe-

rioperative complication rates of primary surgical treatments for prostate cancer: a

systematic review and meta-analysis comparing retropubic, laparoscopic, and robotic

prostatectomy. European urology, 62(1):1–15, 2012. ISSN 0302-2838.

[90] Nima Sharifi, James L Gulley, and William L Dahut. Androgen deprivation therapy

for prostate cancer. Jama, 294(2):238–244, 2005. ISSN 0098-7484.

References 244

[91] Anthony V D’Amico, Judith Manola, Marian Loffredo, Andrew A Renshaw, Alyssa

DellaCroce, and Philip W Kantoff. 6-month androgen suppression plus radiation

therapy vs radiation therapy alone for patients with clinically localized prostate

cancer: a randomized controlled trial. Jama, 292(7):821–827, 2004. ISSN 0098-7484.

[92] Colleen A Lawton, Kathryn Winter, Roger Byhardt, William T Sause, Gerald E

Hanks, Anthony H Russell, Marvin Rotman, Arthur Porter, David G McGowan,

and John D DelRowe. Androgen suppression plus radiation versus radiation alone

for patients with d1 (pn+) adenocarcinoma of the prostate (results based on a na-

tional prospective randomized trial, rtog 85-31). International Journal of Radiation


[93] Colleen A Lawton, Kathryn Winter, Kevin Murray, Mitchell Machtay, John B Mesic,

Gerald E Hanks, Christopher T Coughlin, and Miljenko V Pilepich. Updated results

of the phase iii radiation therapy oncology group (rtog) trial 85-31 evaluating the

potential benefit of androgen suppression following standard radiation therapy for

unfavorable prognosis carcinoma of the prostate. International Journal of Radiation


[94] Miljenko V Pilepich, Kathryn Winter, Colleen A Lawton, Robert E Krisch, Harvey B

Wolkov, Benjamin Movsas, Eugen B Hug, Sucha O Asbell, and David Grignon. An-

drogen suppression adjuvant to definitive radiotherapy in prostate carcinoma—long-

term results of phase iii rtog 85–31. International Journal of Radiation Oncology*

Biology* Physics, 61(5):1285–1290, 2005. ISSN 0360-3016.

[95] Miljenko V Pilepich, Kathryn Winter, Madhu J John, John B Mesic, William Sause,

Phillip Rubin, Colleen Lawton, Mitchell Machtay, and David Grignon. Phase iii ra-

diation therapy oncology group (rtog) trial 86-10 of androgen deprivation adjuvant

to definitive radiotherapy in locally advanced carcinoma of the prostate. Inter-


ISSN 0360-3016.

[96] Eric M Horwitz, Kathryn Winter, Gerald E Hanks, Colleen A Lawton, Anthony H

Russell, and Mitchell Machtay. Subset analysis of rtog 85-31 and 86-10 indicates an

advantage for long-term vs. short-term adjuvant hormones for patients with locally

References 245

advanced nonmetastatic prostate cancer treated with radiation therapy. Interna-

tional Journal of Radiation Oncology* Biology* Physics, 49(4):947–956, 2001. ISSN

0360-3016.

[97] Michael J Zelefsky, Yoshiya Yamada, Marisa A Kollmeier, Alison M Shippy,

and Michelle A Nedelka. Long-term outcome following three-dimensional

conformal/intensity-modulated external-beam radiotherapy for clinical stage t3

prostate cancer. european urology, 53(6):1172–1179, 2008. ISSN 0302-2838.

[98] Grace L Lu-Yao, Peter C Albertsen, Dirk F Moore, Weichung Shih, Yong Lin,

Robert S DiPaola, and Siu-Long Yao. Survival following primary androgen de-

privation therapy among men with localized prostate cancer. Jama, 300(2):173–181,

2008. ISSN 0098-7484.

[99] Arnold L Potosky, Kevin Knopf, Limin X Clegg, Peter C Albertsen, Janet L Stan-

ford, Ann S Hamilton, Frank D Gilliland, J William Eley, Robert A Stephenson,

and Richard M Hoffman. Quality-of-life outcomes after primary androgen depriva-

tion therapy: results from the prostate cancer outcomes study. Journal of Clinical

Oncology, 19(17):3750–3757, 2001. ISSN 0732-183X.

[100] Matthew R Cooperberg, Andrew J Vickers, Jeanette M Broering, and Peter R Car-

roll. Comparative risk‐adjusted mortality outcomes after primary surgery, radio-

therapy, or androgen‐deprivation therapy for localized prostate cancer. Cancer, 116

(22):5226–5234, 2010. ISSN 1097-0142.

[101] Rowan G Casey, Niall M Corcoran, and S Larry Goldenberg. Quality of life issues

in men undergoing androgen deprivation therapy: a review. Asian J Androl, 14(2):

226–31, 2012.

[102] Mariët Hagedoorn, Robbert Sanderman, Hilde N Bolks, Jolanda Tuinstra, and

James C Coyne. Distress in couples coping with cancer: a meta-analysis and crit-

ical review of role and gender effects. Psychological bulletin, 134(1):1, 2008. ISSN

1939-1455.

References 246

[103] Dominik R Berthold, Gregory R Pond, Freidele Soban, Ronald de Wit, Mario Eisen-

berger, and Ian F Tannock. Docetaxel plus prednisone or mitoxantrone plus pred-

nisone for advanced prostate cancer: updated survival in the tax 327 study. Journal

of Clinical Oncology, 26(2):242–245, 2008. ISSN 0732-183X.

[104] Ian F Tannock, Ronald de Wit, William R Berry, Jozsef Horti, Anna Pluzanska,

Kim N Chi, Stephane Oudard, Christine Théodore, Nicholas D James, and Ingela

Turesson. Docetaxel plus prednisone or mitoxantrone plus prednisone for advanced

prostate cancer. New England Journal of Medicine, 351(15):1502–1512, 2004. ISSN

0028-4793.

[105] eviQ. Prostate metastatic docetaxel three weekly and prednisolone, 2015.

URL https://www.eviq.org.au/medical-oncology/urogenital/prostate/

249-prostate-metastatic-docetaxel-three-weekly-and#literature-search.

[106] 2010.

[107] Channing J Paller and Emmanuel S Antonarakis. Cabazitaxel: a novel second-line

treatment for metastatic castration-resistant prostate cancer. Drug design, develop-

ment and therapy, 5:117, 2011.

[108] Stephane Oudard. TROPIC: Phase iii trial of cabazitaxel for the treatment of

metastatic castration-resistant prostate cancer. Future Oncology, 7(4):497–506, 2011.

ISSN 1479-6694.

[109] JS De Bono, S Oudard, M Ozguroglu, S Hansen, JH Machiels, L Shen, P Matthews,

AO Sartor, and TROPIC Investigators. Cabazitaxel or mitoxantrone with prednisone

in patients with metastatic castration-resistant prostate cancer (mCRPC) previously

treated with docetaxel: Final results of a multinational phase iii trial (TROPIC).

Journal of Clinical Oncology, 28(15):4508–4508, 2010. ISSN 0732-183X.

[110] 2013.

[111] Charles G Drake. Prostate cancer as a model for tumour immunotherapy. Nature

Reviews Immunology, 10(8):580–593, 2010. ISSN 1474-1741.

https://www.eviq.org.au/medical-oncology/urogenital/prostate/249-prostate-metastatic-docetaxel-three-weekly-and#literature-search

https://www.eviq.org.au/medical-oncology/urogenital/prostate/249-prostate-metastatic-docetaxel-three-weekly-and#literature-search

References 247

[112] Eric J Small, Paige Fratesi, David M Reese, George Strang, Reiner Laus, Mad-

husudan V Peshwa, and Frank H Valone. Immunotherapy of hormone-refractory

prostate cancer with antigen-loaded dendritic cells. Journal of Clinical Oncology, 18

(23):3894–3903, 2000. ISSN 0732-183X.

[113] Philip W Kantoff, Celestia S Higano, Neal D Shore, E Roy Berger, Eric J Small,

David F Penson, Charles H Redfern, Anna C Ferrari, Robert Dreicer, and Robert B

Sims. Sipuleucel-t immunotherapy for castration-resistant prostate cancer. New

England Journal of Medicine, 363(5):411–422, 2010. ISSN 0028-4793.

[114] Massimo Valerio, Hashim U Ahmed, Mark Emberton, Nathan Lawrentschuk, Mas-

simo Lazzeri, Rodolfo Montironi, Paul L Nguyen, John Trachtenberg, and Thomas J

Polascik. The role of focal therapy in the management of localised prostate cancer:

a systematic review. European urology, 66(4):732–751, 2014. ISSN 0302-2838.

[115] John F Ward and J Stephen Jones. Focal cryotherapy for localized prostate cancer:

a report from the national cryo on‐line database (cold) registry. BJU international,

109(11):1648–1654, 2012. ISSN 1464-410X.

[116] Sébastien Crouzet, Olivier Rouviere, Xavier Martin, and Albert Gelet. High-

intensity focused ultrasound as focal therapy of prostate cancer. Current opinion

in urology, 24(3):225–230, 2014. ISSN 0963-0643.

[117] S Khaled and KD Held. Radiation Biology: A Handbook for Teachers and Students.

Training Coarse Series. International Atomic Energy Agency, 2010.

[118] Penny Jeggo and Martin F Lavin. Cellular radiosensitivity: how much better do we

understand it? International journal of radiation biology, 85(12):1061–1081, 2009.

ISSN 0955-3002.

[119] Judith Campisi and Fabrizio d’Adda di Fagagna. Cellular senescence: when bad

things happen to good cells. Nature reviews Molecular cell biology, 8(9):729–740,

2007. ISSN 1471-0072.

[120] J Martin Brown and Laura D Attardi. The role of apoptosis in cancer development

and treatment response. Nature Reviews Cancer, 5(3):231–237, 2005. ISSN 1474-

175X.

References 248

[121] Maria Castedo, Jean-Luc Perfettini, Thomas Roumier, Karine Andreau, Rene

Medema, and Guido Kroemer. Cell death by mitotic catastrophe: a molecular

definition. Oncogene, 23(16):2825–2837, 2004. ISSN 0950-9232.

[122] Wei-Xing Zong, Dara Ditsworth, Daniel E Bauer, Zhao-Qi Wang, and Craig B

Thompson. Alkylating dna damage stimulates a regulated form of necrotic cell

death. Genes& development, 18(11):1272–1282, 2004. ISSN 0890-9369.

[123] Dana E Christofferson and Junying Yuan. Necroptosis as an alternative form of

programmed cell death. Current opinion in cell biology, 22(2):263–268, 2010. ISSN

0955-0674.

[124] Larry Bodgi, Aurélien Canet, Laurent Pujo-Menjouet, Annick Lesne, Jean-Marc

Victor, and Nicolas Foray. Mathematical models of radiation action on living cells:

from the target theory to the modern approaches. a historical and critical review.

Journal of Theoretical Biology, 2016. ISSN 0022-5193.

[125] Albrecht M Kellerer and Harald H Rossi. A generalized formulation of dual radiation

action. Radiation Research, 75(3):471–488, 1978. ISSN 0033-7587.

[126] KH Chadwick and HP Leenhouts. A molecular theory of cell survival. Physics in

medicine and biology, 18(1):78, 1973. ISSN 0031-9155.

[127] Jian Z Wang, M Guerrero, and X Allen Li. How low is the α/β ratio for prostate

cancer? International Journal of Radiation Oncology* Biology* Physics, 55(1):

194–203, 2003. ISSN 0360-3016.

[128] David J Brenner, Alvaro A Martinez, Gregory K Edmundson, Christina Mitchell,

Howard D Thames, and Elwood P Armour. Direct evidence that prostate tumors

show high sensitivity to fractionation (low α/β ratio), similar to late-responding

normal tissue. International Journal of Radiation Oncology* Biology* Physics, 52

(1):6–13, 2002. ISSN 0360-3016.

[129] Ivan R Vogelius and Søren M Bentzen. Meta-analysis of the alpha/beta ratio for

prostate cancer in the presence of an overall time factor: Bad news, good news, or

no news? International Journal of Radiation Oncology* Biology* Physics, 85(1):

89–94, 2013. ISSN 0360-3016.

References 249

[130] Alexandru Dasu and Iuliana Toma-Dasu. Prostate alpha/beta revisited–an analysis

of clinical results from 14 168 patients. Acta Oncologica, 51(8):963–974, 2012. ISSN

0284-186X.

[131] Felix Leborgne, Jack Fowler, José H Leborgne, and Julieta Mezzera. Later outcomes

and alpha/beta estimate from hypofractionated conformal three-dimensional radio-

therapy versus standard fractionation for localized prostate cancer. International


0360-3016.

[132] Cécile Proust-Lima, Jeremy MG Taylor, Solène Sécher, Howard Sandler, Larry

Kestin, Tom Pickles, Kyoungwha Bae, Roger Allison, and Scott Williams. Con-

firmation of a low α/β ratio for prostate cancer treated by external beam radiation

therapy alone using a post-treatment repeated-measures model for psa dynamics. In-

ternational Journal of Radiation Oncology* Biology* Physics, 79(1):195–201, 2011.

ISSN 0360-3016.

[133] G Gordon Steel, Trevor J McMillan, and JH Peacock. The 5rs of radiobiology.

International journal of radiation biology, 56(6):1045–1048, 1989. ISSN 0955-3002.

[134] HR Withers, JMG Taylor, and B Maciejewski. The hazard of accelerated tumor

clonogen repopulation during radiotherapy. Acta Oncologica, 27(2):131–146, 1988.

ISSN 0284-186X.

[135] George D Wilson. Radiation and the cell cycle, revisited. Cancer and Metastasis

Reviews, 23(3-4):209–225, 2004. ISSN 0167-7659.

[136] Michael B Kastan and Jiri Bartek. Cell-cycle checkpoints and cancer. Nature, 432

(7015):316–323, 2004. ISSN 0028-0836.

[137] ICRU. Icru report 83: Prescribing, recording, and reporting photon-beam intensity-

modulated radiation therapy (imrt). Journal of the ICRU, 10(1), 2010.

[138] Douglas Jones. Icru report 50—prescribing, recording and reporting photon beam

therapy. Medical Physics, 21(6):833–834, 1994. ISSN 0094-2405.

References 250

[139] A Wambersie and T Landgerg. Icru report 62: Prescribing, recording and reporting

photon beam therapy. ICRU Publ Bethesda MD, 1999.

[140] P Metcalfe, GP Liney, L Holloway, A Walker, M Barton, GP Delaney, S Vinod, and

W Tome. The potential for an enhanced role for mri in radiation-therapy treatment

planning. Technology in cancer research& treatment, 12(5):429–446, 2013. ISSN

1533-0346.

[141] Peter B Greer, Jason A Dowling, Jonathon A Lambert, Jurgen Fripp, Joel Parker,

James W Denham, Chris Wratten, Anne Capp, and Olivier Salvado. A magnetic

resonance imaging-based workflow for planning radiation therapy for prostate cancer.

Medical Journal of Australia, 194(4):S24, 2011. ISSN 0025-729X.

[142] Marc Debois, Raymond Oyen, Frederik Maes, Geert Verswijvel, Giovanna Gatti,

Hilde Bosmans, Michel Feron, Erwin Bellon, Gerald Kutcher, and Hein Van Poppel.

The contribution of magnetic resonance imaging to the three-dimensional treatment

planning of localized prostate cancer. International Journal of Radiation Oncology*


[143] Louise Dickinson, Hashim U Ahmed, Clare Allen, Jelle O Barentsz, Brendan Carey,

Jurgen J Futterer, Stijn W Heijmink, Peter J Hoskin, Alex Kirkham, and Anwar R

Padhani. Magnetic resonance imaging for the detection, localisation, and charac-

terisation of prostate cancer: recommendations from a european consensus meeting.

European urology, 59(4):477–494, 2011. ISSN 0302-2838.

[144] CC Parker, A Damyanovich, T Haycocks, M Haider, A Bayley, and CN Catton. Mag-

netic resonance imaging in the radiation treatment planning of localized prostate can-

cer using intra-prostatic fiducial markers for computed tomography co-registration.

Radiotherapy and oncology, 66(2):217–224, 2003. ISSN 0167-8140.

[145] Carl Salembier, Geert Villeirs, Berardino De Bari, Peter Hoskin, Bradley R. Pieters,

Marco Van Vulpen, Vincent Khoo, Ann Henry, Alberto Bossi, Gert De Meerleer,

and Valérie Fonteyne. ESTRO ACROP consensus guideline on ct- and mri-based

target volume delineation for primary radiation therapy of localized prostate cancer.

Radiotherapy and Oncology, 127(1):49–61, 2018. ISSN 0167-8140.

References 251

[146] Claudio Fiorino, Riccardo Valdagni, Tiziana Rancati, and Giuseppe Sanguineti.

Dose–volume effects for normal tissues in external radiotherapy: pelvis. Radio-

therapy and Oncology, 93(2):153–167, 2009. ISSN 0167-8140.

[147] Hiram A Gay, H Joseph Barthold, Elizabeth O’Meara, Walter R Bosch, Issam

El Naqa, Rawan Al-Lozi, Seth A Rosenthal, Colleen Lawton, W Robert Lee, and

Howard Sandler. Pelvic normal tissue contouring guidelines for radiation therapy:

a radiation therapy oncology group consensus panel atlas. International Journal of

Radiation Oncology* Biology* Physics, 83(3):e353–e362, 2012. ISSN 0360-3016.

[148] Anders Brahme, J-E Roos, and Ingemar Lax. Solution of an integral equation en-

countered in rotation therapy. Physics in medicine and biology, 27(10):1221, 1982.

ISSN 0031-9155.

[149] Anders Brahme. Optimization of stationary and moving beam radiation therapy

techniques. Radiotherapy and Oncology, 12(2):129–140, 1988. ISSN 0167-8140.

[150] Karl Bzdusek, Henrik Friberger, Kjell Eriksson, Björn Hårdemark, David Robinson,

and Michael Kaus. Development and evaluation of an efficient approach to volumetric

arc therapy planning. Medical physics, 36(6):2328–2339, 2009. ISSN 0094-2405.

[151] Vladimir Feygelman, Geoffrey Zhang, and Craig Stevens. Initial dosimetric eval-

uation of smartarc–a novel vmat treatment planning module implemented in a

multi‐vendor delivery chain. Journal of applied clinical medical physics, 11(1):99–

116, 2010. ISSN 1526-9914.

[152] MA Earl, DM Shepard, S Naqvi, XA Li, and CX Yu. Inverse planning for intensity-

modulated arc therapy using direct aperture optimization. Physics in Medicine&

Biology, 48(8):1075, 2003. ISSN 0031-9155.

[153] Steven A Leibel, Zvi Fuks, Michael J Zelefsky, Suzanne L Wolden, Kenneth E

Rosenzweig, Kaled M Alektiar, Margie A Hunt, Ellen D Yorke, Linda X Hong,

and Howard I Amols. Intensity‐modulated radiotherapy. The Cancer Journal, 8(2):

164–176, 2002. ISSN 1528-9117.

[154] Z Fuks and A Horwich. Clinical and technical aspects of conformal therapy. Radio-

therapy and Oncology, 29(2):219, 1993. ISSN 0167-8140.

References 252

[155] Alan Pollack, Gunar K Zagars, George Starkschall, Constance H Childress, Susan

Kopplin, Arthur L Boyer, and Isaac I Rosen. Conventional vs. conformal radiother-

apy for prostate cancer: preliminary results of dosimetry and acute toxicity. In-


ISSN 0360-3016.

[156] David P Dearnaley, Vincent S Khoo, Andrew R Norman, Lesley Meyer, Alan Nahum,

Diana Tait, John Yarnold, and Alan Horwich. Comparison of radiation side-effects

of conformal and conventional radiotherapy in prostate cancer: a randomised trial.

The Lancet, 353(9149):267–272, 1999. ISSN 0140-6736.

[157] David E Morris, Bahman Emami, Peter M Mauch, Andre A Konski, May L Tao,

Andrea K Ng, Eric A Klein, Najeeb Mohideen, Mark D Hurwitz, and Bendick A

Fraas. Evidence-based review of three-dimensional conformal radiotherapy for local-

ized prostate cancer: an astro outcomes initiative. International Journal of Radiation


[158] Michael Brundage, Himu Lukka, Juanita Crook, Padraig Warde, Glenn Bauman,

Charles Catton, Barbara R Markman, Manya Charette, and Cancer care Ontario

Practice Guidelines Initiative Genitourinary Cancer Disease Site Group. The use

of conformal radiotherapy and the selection of radiation dose in t1 or t2 low or

intermediate risk prostate cancer–a systematic review. Radiotherapy and oncology,

64(3):239–250, 2002. ISSN 0167-8140.

[159] Isabel Syndikus, Rachel C Morgan, Matthew R Sydes, John D Graham, and David P

Dearnaley. Late gastrointestinal toxicity after dose-escalated conformal radiotherapy

for early prostate cancer: results from the uk medical research council rt01 trial

(isrctn47772397). International Journal of Radiation Oncology* Biology* Physics,

77(3):773–783, 2010. ISSN 0360-3016.

[160] Lynn J Verhey. Comparison of three-dimensional conformal radiation therapy and

intensity-modulated radiation therapy systems. In Seminars in radiation oncology,

volume 9, pages 78–98. Elsevier, 1999. ISBN 1053-4296.

References 253

[161] Yasumasa Nishimura and Ritsuko Komaki. Intensity-Modulated Radiation Therapy:

Clinical Evidence and Techniques. Tokyo: Springer Japan, 2015. ISBN 4431554858.

[162] Oren Cahlon, Michael J Zelefsky, Alison Shippy, Heather Chan, Zvi Fuks, Yoshiya

Yamada, Margie Hunt, Steven Greenstein, and Howard Amols. Ultra-high dose

(86.4 gy) imrt for localized prostate cancer: toxicity and biochemical outcomes.

International Journal of Radiation Oncology* Biology* Physics, 71(2):330–337, 2008.

ISSN 0360-3016.

[163] Christopher R King, Thomas A DiPetrillo, and David E Wazer. Optimal radio-

therapy for prostate cancer: predictions for conventional external beam, imrt, and

brachytherapy from radiobiologic models. International Journal of Radiation On-

cology* Biology* Physics, 46(1):165–172, 2000. ISSN 0360-3016.

[164] Michael J Zelefsky, Zvi Fuks, Margie Hunt, Henry J Le, Danna Lombardi, Clifton C

Ling, Victor E Reuter, Ennapadam S Venkatraman, and Steven A Leibel. High

dose radiation delivered by intensity modulated conformal radiotherapy improves

the outcome of localized prostate cancer. The Journal of urology, 166(3):876–881,

2001. ISSN 0022-5347.

[165] Michael J Zelefsky, Heather Chan, Margie Hunt, Yoshiya Yamada, Alison M Shippy,

and Howard Amols. Long-term outcome of high dose intensity modulated radiation

therapy for patients with clinically localized prostate cancer. The Journal of urology,

176(4):1415–1419, 2006. ISSN 0022-5347.

[166] Gary Luxton, Steven L Hancock, and Arthur L Boyer. Dosimetry and radiobiologic

model comparison of imrt and 3d conformal radiotherapy in treatment of carcinoma

of the prostate. International Journal of Radiation Oncology* Biology* Physics, 59

(1):267–284, 2004. ISSN 0360-3016.

[167] Pascal Fenoglietto, Benoit Laliberte, Ali Allaw, Norbert Ailleres, Katia Idri,

Meng Huor Hay, Carmen Llacer Moscardo, Sophie Gourgou, Jean-Bernard Dubois,

and David Azria. Persistently better treatment planning results of intensity-

modulated (imrt) over conformal radiotherapy (3d-crt) in prostate cancer patients

References 254

with significant variation of clinical target volume and/or organs-at-risk. Radiother-

apy and Oncology, 88(1):77–87, 2008. ISSN 0167-8140.

[168] Maria T Vlachaki, Terrance N Teslow, Chad Amosson, Nathan W Uy, and Salahud-

din Ahmad. Imrt versus conventional 3dcrt on prostate and normal tissue dosimetry

using an endorectal balloon for prostate immobilization. Medical Dosimetry, 30(2):

69–75, 2005. ISSN 0958-3947.

[169] Silvia Hummel, EL Simpson, P Hemingway, MD Stevenson, and A Rees. Intensity-

modulated radiotherapy for the treatment of prostate cancer: a systematic review and

economic evaluation, volume 14. Health Technology Assessment, 2010.

[170] Oren Cahlon, Margie Hunt, and Michael J Zelefsky. Intensity-modulated radiation

therapy: supportive data for prostate cancer. In Seminars in radiation oncology,


[171] Justin E Bekelman, Nandita Mitra, Jason Efstathiou, Kaijun Liao, Robert Sun-

derland, Deborah N Yeboa, and Katrina Armstrong. Outcomes after intensity-

modulated versus conformal radiotherapy in older men with nonmetastatic prostate

cancer. International Journal of Radiation Oncology* Biology* Physics, 81(4):e325–

e334, 2011. ISSN 0360-3016.

[172] Michael J Zelefsky, Zvi Fuks, Margie Hunt, Yoshiya Yamada, Christine Marion,

C Clifton Ling, Howard Amols, Ennapadam S Venkatraman, and Steven A Leibel.

High-dose intensity modulated radiation therapy for prostate cancer: early toxic-

ity and biochemical outcome in 772 patients. International Journal of Radiation


[173] Michael J Zelefsky, Emily J Levin, Margie Hunt, Yoshiya Yamada, Alison M Shippy,

Andrew Jackson, and Howard I Amols. Incidence of late rectal and urinary toxicities

after three-dimensional conformal radiotherapy and intensity-modulated radiother-

apy for localized prostate cancer. International Journal of Radiation Oncology*


[174] Michael J Zelefsky, Zvi Fuks, Laura Happersett, Henry J Lee, C Clifton Ling,

Chandra M Burman, Margie Hunt, Theresa Wolfe, ES Venkatraman, and Andrew

References 255

Jackson. Clinical experience with intensity modulated radiation therapy (imrt) in

prostate cancer. Radiotherapy and Oncology, 55(3):241–249, 2000. ISSN 0167-8140.

[175] Nathan C Sheets, Gregg H Goldin, Anne-Marie Meyer, Yang Wu, YunKyung Chang,

Til Stürmer, Jordan A Holmes, Bryce B Reeve, Paul A Godley, and William R

Carpenter. Intensity-modulated radiation therapy, proton therapy, or conformal

radiation therapy and morbidity and disease control in localized prostate cancer.

Jama, 307(15):1611–1620, 2012. ISSN 0098-7484.

[176] Eric J Hall and Cheng-Shie Wuu. Radiation-induced second cancers: the impact of

3d-crt and imrt. International Journal of Radiation Oncology* Biology* Physics, 56

(1):83–88, 2003. ISSN 0360-3016.

[177] Hannah E Carter, Andrew Martin, Deborah Schofield, Gillian Duchesne, Annette

Haworth, Colin Hornby, Mark Sidhom, and Michael Jackson. A decision model to

estimate the cost-effectiveness of intensity modulated radiation therapy (imrt) com-

pared to three dimensional conformal radiation therapy (3dcrt) in patients receiving

radiotherapy to the prostate bed. Radiotherapy and Oncology, 112(2):187–193, 2014.

ISSN 0167-8140.

[178] Robert W Kopp, Michael Duff, Frank Catalfamo, Dhiren Shah, Michael Rajecki,

and Kehkashan Ahmad. Vmat vs. 7-field-imrt: assessing the dosimetric parameters

of prostate cancer treatment with a 292-patient sample. Medical Dosimetry, 36(4):

365–372, 2012. ISSN 0958-3947.

[179] Enzhuo M Quan, Xiaoqiang Li, Yupeng Li, Xiaochun Wang, Rajat J Kudchadker,

Jennifer L Johnson, Deborah A Kuban, Andrew K Lee, and Xiaodong Zhang. A com-

prehensive comparison of imrt and vmat plan quality for prostate cancer treatment.

International Journal of Radiation Oncology* Biology* Physics, 83(4):1169–1178,

2012. ISSN 0360-3016.

[180] Karl Otto. Volumetric modulated arc therapy: Imrt in a single gantry arc. Medical

physics, 35(1):310–317, 2008. ISSN 0094-2405.

[181] R Shaffer, WJ Morris, V Moiseenko, M Welsh, C Crumley, S Nakano, M Schmu-

land, T Pickles, and K Otto. Volumetric modulated arc therapy and conventional

References 256

intensity-modulated radiotherapy for simultaneous maximal intraprostatic boost: a

planning comparison study. Clinical oncology, 21(5):401–407, 2009. ISSN 0936-6555.

[182] Pengpeng Zhang, Laura Happersett, Margie Hunt, Andrew Jackson, Michael Zelef-

sky, and Gig Mageras. Volumetric modulated arc therapy: planning and evaluation

for prostate cancer cases. International Journal of Radiation Oncology* Biology*

Physics, 76(5):1456–1462, 2010. ISSN 0360-3016.

[183] David Palma, Emily Vollans, Kerry James, Sandy Nakano, Vitali Moiseenko,

Richard Shaffer, Michael McKenzie, James Morris, and Karl Otto. Volumetric mod-

ulated arc therapy for delivery of prostate radiotherapy: comparison with intensity-

modulated radiotherapy and three-dimensional conformal radiotherapy. Interna-


0360-3016.

[184] Chiao-Ling Tsai, Jian-Kuen Wu, Hsiao-Ling Chao, Yi-Chun Tsai, and Jason Chia-

Hsien Cheng. Treatment and dosimetric advantages between vmat, imrt, and helical

tomotherapy in prostate cancer. Medical Dosimetry, 36(3):264–271, 2011. ISSN

0958-3947.

[185] Min Rao, Wensha Yang, Fan Chen, Ke Sheng, Jinsong Ye, Vivek Mehta, David

Shepard, and Daliang Cao. Comparison of elekta vmat with helical tomotherapy

and fixed field imrt: plan quality, delivery efficiency and accuracy. Medical physics,

37(3):1350–1359, 2010. ISSN 0094-2405.

[186] Melanie TM Davidson, Samuel J Blake, Deidre L Batchelar, Patrick Cheung, and

Katherine Mah. Assessing the role of volumetric modulated arc therapy (vmat)

relative to imrt and helical tomotherapy in the management of localized, locally

advanced, and post-operative prostate cancer. International Journal of Radiation


[187] Dirk Wolff, Florian Stieler, Grit Welzel, Friedlieb Lorenz, Yasser Abo-Madyan,

Sabine Mai, Carsten Herskind, Martin Polednik, Volker Steil, and Frederik Wenz.

Volumetric modulated arc therapy (vmat) vs. serial tomotherapy, step-and-shoot

References 257

imrt and 3d-conformal rt for treatment of prostate cancer. Radiotherapy and oncol-

ogy, 93(2):226–233, 2009. ISSN 0167-8140.

[188] David Pryor, Mark Sidhom, Sankar Arumugam, Joseph Bucci, Sarah Gallagher,

Joanne Smart, Melissa Grand, Lee Wilton, and Jarad Martin. Phase 2 multicentre

study of gantry-based stereotactic radiotherapy boost for intermediate and high risk

prostate cancer (prometheus). Frontiers in oncology, 9:217, 2019. ISSN 2234-943X.

[189] Matthew Richardson, Mark Sidhom, Sarah Gallagher, Mel Grand, David Pryor,

Joseph Bucci, Lee Wilton, Sankar Arumugam, Sarah Keats, and Jarad M Martin.

Prostate multicentre external beam radiotherapy using a stereotactic boost: the

prometheus study protocol. BMC cancer, 18(1):588, 2018. ISSN 1471-2407.

[190] Stanley H Benedict, Kamil M Yenice, David Followill, James M Galvin, William

Hinson, Brian Kavanagh, Paul Keall, Michael Lovelock, Sanford Meeks, and Lech

Papiez. Stereotactic body radiation therapy: the report of aapm task group 101.

Medical physics, 37(8):4078–4101, 2010. ISSN 0094-2405.

[191] Brian K Chang and Robert D Timmerman. Stereotactic body radiation therapy: a

comprehensive review. American journal of clinical oncology, 30(6):637–644, 2007.

ISSN 0277-3732.

[192] Jack Fowler, Rick Chappell, and Mark Ritter. Is α/β for prostate tumors really

low? International Journal of Radiation Oncology* Biology* Physics, 50(4):1021–

1031, 2001. ISSN 0360-3016.

[193] Jack F Fowler, Mark A Ritter, Rick J Chappell, and David J Brenner. What hy-

pofractionated protocols should be tested for prostate cancer? International Journal


[194] Debra E Freeman and Christopher R King. Stereotactic body radiotherapy for low-

risk prostate cancer: five-year outcomes. Radiation Oncology, 6(1), 2011. ISSN

1748-717X.

[195] Berit L Madsen, R Alex Hsi, Huong T Pham, Jack F Fowler, Laura Esagui, and

John Corman. Stereotactic hypofractionated accurate radiotherapy of the prostate

References 258

(sharp), 33.5 gy in five fractions for localized disease: first clinical trial results.


2007. ISSN 0360-3016.

[196] B Detti, P Bonomo, L Masi, R Doro, S Cipressi, C Iermano, I Bonucci, D Frances-

chini, V Di Cataldo, and L Di Brina. Cyberknife treatment for low and intermediate

risk prostate cancer. Cancer investigation, 2015. ISSN 0735-7907.

[197] Alan J Katz, Michael Santoro, Fred Diblasio, and Richard Ashley. 6 year cyberknife

prostate cancer study released. Radiation Oncology, 8:118, 2013.

[198] Christopher R King, James D Brooks, Harcharan Gill, Todd Pawlicki, Cristian

Cotrutz, and Joseph C Presti. Stereotactic body radiotherapy for localized prostate

cancer: interim results of a prospective phase ii clinical trial. International Journal


[199] Christopher R King, James D Brooks, Harcharan Gill, and Joseph C Presti. Long-

term outcomes from a prospective trial of stereotactic body radiotherapy for low-risk

prostate cancer. International Journal of Radiation Oncology* Biology* Physics, 82

(2):877–882, 2012. ISSN 0360-3016.

[200] Ellen A Wiegner and Christopher R King. Sexual function after stereotactic body

radiotherapy for prostate cancer: results of a prospective clinical trial. International


3016.

[201] Thomas P Boike, Yair Lotan, L Chinsoo Cho, Jeffrey Brindle, Paul DeRose, Xian-Jin

Xie, Jingsheng Yan, Ryan Foster, David Pistenmaa, and Alida Perkins. Phase i dose-

escalation study of stereotactic body radiation therapy for low-and intermediate-risk

prostate cancer. Journal of Clinical Oncology, 29(15):2020–2026, 2011. ISSN 0732-

183X.

[202] Leonard N Chen, Simeng Suy, Sunghae Uhm, Eric K Oermann, Andrew W Ju, Viola

Chen, Heather N Hanscom, Sarah Laing, Joy S Kim, and Siyuan Lei. Stereotactic

body radiation therapy (sbrt) for clinically localized prostate cancer: the georgetown

university experience. Radiat Oncol, 8(1):58, 2013.

References 259

[203] Jay L Friedland, Debra E Freeman, Mary Ellen Masterson-McGary, and David M

Spellberg. Stereotactic body radiotherapy: an emerging treatment approach for

localized prostate cancer. Technology in cancer research& treatment, 8(5):387–392,

2009. ISSN 1533-0346.

[204] William C Jackson, Jessica Silva, Holly E Hartman, Robert T Dess, Amar U Kishan,

Whitney H Beeler, Laila A Gharzai, Elizabeth M Jaworski, Rohit Mehra, Jason WD

Hearn, Todd M Morgan, Simpa S Salami, Matthew R Cooperberg, Payal D Mahal,

Brandon Aand Soni, Samuel Kaffenberger, Paul L Nguyen, Neil Desai, Felix Y Feng,

Zachary S Zumsteg, and Daniel E Spratt. Stereotactic body radiation therapy for

localized prostate cancer: a systematic review and meta-analysis of over 6,000 pa-

tients treated on prospective studies. International Journal of Radiation Oncology*


[205] Anders Widmark, Adalsteinn Gunnlaugsson, Lars Beckman, Camilla Thellenberg-

Karlsson, Morten Hoyer, Magnus Lagerlund, Jon Kindblom, Claes Ginman, Bengt

Johansson, Kirsten Björnlinger, Mihajl Seke, Mans Agrup, Per Fransson, Björn

Tavelin, David Norman, Björn Zackrisson, Harald Anderson, Elisabeth Kjellen, Lars

Franzen, and Per Nilsson. Ultra-hypofractionated versus conventionally fractionated

radiotherapy for prostate cancer: 5-year outcomes of the HYPO-RT-PC randomised,

non-inferiority, phase 3 trial. The Lancet, 394(10196):385–395, 2019. ISSN 0140-6736.

[206] Douglas H Brand, Alison C Tree, Peter Ostler, Hans van der Voet, Andrew Loblaw,

William Chu, Daniel Ford, Shaun Tolan, Suneil Jain, Alexander Martin, John

Staffurth, Philip Camilleri, Kiran Kancherla, John Frew, Andrew Chan, Ian S.

Dayes, Daniel Henderson, Stephanie Brown, Clare Cruickshank, Stephanie Burnett,

Aileen Duffton, Clare Griffin, Victoria Hinder, Kirsty Morrison, Olivia Naismith,

Emma Hall, and Nicholas van As. Intensity-modulated fractionated radiotherapy

versus stereotactic body radiotherapy for prostate cancer (pace-b): acute toxicity

findings from an international, randomised, open-label, phase 3, non-inferiority trial.

The Lancet Oncology, 20(11):1531–1543, 2019. ISSN 1470-2045.

[207] James M Galvin, Gary Ezzell, Avraham Eisbrauch, Cedric Yu, Brian Butler, Ying

Xiao, Isaac Rosen, Julian Rosenman, Michael Sharpe, and Lei Xing. Implementing

References 260

imrt in clinical practice: a joint document of the american society for therapeutic

radiology and oncology and the american association of physicists in medicine. Inter-


ISSN 0360-3016.

[208] James M Balter, Howard M Sandler, Kwok Lam, Robert L Bree, Allen S Lichter,

and Randall K Ten Haken. Measurement of prostate movement over the course of

routine radiotherapy using implanted markers. International Journal of Radiation


[209] JM Crook, Y Raymond, D Salhani, H Yang, and B Esche. Prostate motion during

standard radiotherapy as assessed by fiducial markers. Radiotherapy and Oncology,

37(1):35–42, 1995. ISSN 0167-8140.

[210] Marciana Nona Duma and Patrick Kupelian. IGRT: How and When, pages 103–

107. Springer, 2015. ISBN 3642370985. URL http://www.springer.com/fr/book/

9783642370984.

[211] Michael J Zelefsky, Marisa Kollmeier, Brett Cox, Anthony Fidaleo, Dahlia Sperling,

Xin Pei, Brett Carver, Jonathan Coleman, Michael Lovelock, and Margie Hunt.

Improved clinical outcomes with high-dose image guided radiotherapy compared

with non-igrt for the treatment of clinically localized prostate cancer. International


3016.

[212] Ken Takeda, Yoshihiro Takai, Kakutaro Narazaki, Masatoshi Mitsuya, Rei

Umezawa, Noriyuki Kadoya, Yukio Fujita, Toshiyuki Sugawara, Masaki Kubozono,

and Eiji Shimizu. Treatment outcome of high-dose image-guided intensity-modulated

radiotherapy using intra-prostate fiducial markers for localized prostate cancer at a

single institute in japan. Radiat Oncol, 7(1):105, 2012.

[213] J Van Dyk, JJ Batista, and Glenn S Bauman. Accuracy and uncertainty considera-

tions in modern radiation oncology. The Modern Technology of Radiation Oncology,

3:361–412, 2013.

http://www.springer.com/fr/book/9783642370984

http://www.springer.com/fr/book/9783642370984

References 261

[214] Damien C Weber, Coen W Hurkmans, Christos Melidis, Wilfried Budach, Jo-

hannes H Langendijk, Lester J Peters, Vincent Grégoire, Philippe Maingon, and

Christophe Combescure. Outcome impact and cost-effectiveness of quality assur-

ance for radiotherapy planned for the eortc 22071–24071 prospective study for head

and neck cancer. Radiotherapy and Oncology, 111(3):393–399, 2014. ISSN 0167-8140.

[215] David Thwaites, Pierre Scalliet, Jan Willem Leer, and Jens Overgaard. Quality

assurance in radiotherapy: European society for therapeutic radiology and oncology

advisory report to the commission of the european union for the ‘europe against

cancer programme’. Radiotherapy and Oncology, 35(1):61–73, 1995. ISSN 0167-8140.

[216] Eric E Klein, Joseph Hanley, John Bayouth, Fang‐Fang Yin, William Simon, Sean

Dresser, Christopher Serago, Francisco Aguirre, Lijun Ma, and Bijan Arjomandy.

Task group 142 report: Quality assurance of medical accelerators a. Medical physics,

36(9Part1):4197–4212, 2009. ISSN 0094-2405.

[217] Damien C Weber, Milan Tomsej, Christos Melidis, and Coen W Hurkmans. Qa

makes a clinical trial stronger: evidence-based medicine in radiation therapy. Ra-

diotherapy and Oncology, 105(1):4–8, 2012. ISSN 0167-8140.

[218] Morten N Pettersen, Edwin Aird, and Dag R Olsen. Quality assurance of dosime-

try and the impact on sample size in randomized clinical trials. Radiotherapy and

Oncology, 86(2):195–199, 2008. ISSN 0167-8140.

[219] Christos Melidis, Walther R Bosch, Joanna Izewska, Elena Fidarova, Eduardo Zu-

bizarreta, Kenneth Ulin, Satoshi Ishikura, David Followill, James Galvin, and An-

nette Haworth. Global harmonization of quality assurance naming conventions in

radiation therapy clinical trials. International Journal of Radiation Oncology* Biol-

ogy* Physics, 90(5):1242–1249, 2014. ISSN 0360-3016.

[220] Lawrence B Marks, Robert D Adams, Todd Pawlicki, Albert L Blumberg, David

Hoopes, Michael D Brundage, and Benedick A Fraass. Enhancing the role of case-

oriented peer review to improve quality and safety in radiation oncology: executive

summary. Practical radiation oncology, 3(3):149–156, 2013. ISSN 1879-8500.

References 262

[221] Nitin Ohri, Xinglei Shen, Adam P Dicker, Laura A Doyle, Amy S Harrison, and

Timothy N Showalter. Radiotherapy protocol deviations and clinical outcomes: a

meta-analysis of cooperative group clinical trials. Journal of the National Cancer

Institute, page djt001, 2013. ISSN 0027-8874.

[222] Ross A Abrams, Kathryn A Winter, William F Regine, Howard Safran, John P

Hoffman, Robert Lustig, Andre A Konski, Al B Benson, John S Macdonald, and

Tyvin A Rich. Failure to adhere to protocol specified radiation therapy guidelines

was associated with decreased survival in rtog 9704—a phase iii trial of adjuvant

chemotherapy and chemoradiotherapy for patients with resected adenocarcinoma of

the pancreas. International Journal of Radiation Oncology* Biology* Physics, 82(2):

809–816, 2012. ISSN 0360-3016. doi: https://doi.org/10.1016/j.ijrobp.2010.11.039.

[223] Avraham Eisbruch, Jonathan Harris, Adam S Garden, Clifford KS Chao, William

Straube, Paul M Harari, Giuseppe Sanguineti, Christopher U Jones, Walter R Bosch,

and K Kian Ang. Multi-institutional trial of accelerated hypofractionated intensity-

modulated radiation therapy for early-stage oropharyngeal cancer (rtog 00-22). In-


ISSN 0360-3016.

[224] Berthe MP Aleman, Théodore Girinsky, Richard WM van Der Maazen, Simon Strijk,

Paul Meijnders, Roberto Bortolus, Manouk JJ Olofsen-van Acht, Marnix LM Ly-

beert, Yolande Lievens, and Houchingue Eghbali. Quality control of involved-field

radiotherapy in patients with advanced hodgkin’s lymphoma (eortc 20884). Inter-


ISSN 0360-3016.

[225] William F Regine, Kathryn A Winter, Ross A Abrams, Howard Safran, John P

Hoffman, Andre Konski, Al B Benson, John S Macdonald, Mahesh R Kudrimoti,

and Mitchel L Fromm. Fluorouracil vs gemcitabine chemotherapy before and after

fluorouracil-based chemoradiation following resection of pancreatic adenocarcinoma:

a randomized controlled trial. Jama, 299(9):1019–1026, 2008. ISSN 0098-7484.

[226] Carol J Fabian, Carl M Mansfield, Steve Dahlberg, Stephen E Jones, Thomas P

Miller, Ellis Van Slyck, Petre N Grozea, Francis S Morrison, Charles A Coltman, and

References 263

Richard I Fisher. Low-dose involved field radiation after chemotherapy in advanced

hodgkin disease: A southwest oncology group randomized study. Annals of internal

medicine, 120(11):903–912, 1994. ISSN 0003-4819.

[227] Danny Rischin, Lester J Peters, Brian O’Sullivan, Jordi Giralt, Richard Fisher, Kally

Yuen, Andy Trotti, Jacques Bernier, Jean Bourhis, and Jolie Ringash. Tirapazamine,

cisplatin, and radiation versus cisplatin and radiation for advanced squamous cell

carcinoma of the head and neck (trog 02.02, headstart): a phase iii trial of the trans-

tasman radiation oncology group. Journal of Clinical Oncology, page JCO. 2009.27.

4449, 2010. ISSN 0732-183X.

[228] Elisabeth Weiss and Clemens F Hess. The impact of gross tumor volume (gtv)

and clinical target volume (ctv) definition on the total accuracy in radiotherapy.

Strahlentherapie und Onkologie, 179(1):21–30, 2003. ISSN 0179-7158.

[229] Roel JHM Steenbakkers, Joop C Duppen, Isabelle Fitton, Kirsten EI Deurloo, Lam-

bert J Zijp, Emile FI Comans, Apollonia LJ Uitterhoeve, Patrick TR Rodrigus,

Gijsbert WP Kramer, and Johan Bussink. Reduction of observer variation using

matched ct-pet for lung cancer delineation: a three-dimensional analysis. Interna-


0360-3016.

[230] Femke OB Spoelstra, Suresh Senan, Cecile Le Péchoux, Satoshi Ishikura, Francesc

Casas, David Ball, Allan Price, Dirk De Ruysscher, John R van Sörnsen de Koste,

and Lung Adjuvant Radiotherapy Trial Investigators Group. Variations in target vol-

ume definition for postoperative radiotherapy in stage iii non–small-cell lung cancer:

Analysis of an international contouring study. International Journal of Radiation


[231] Dee H Wu, Nina A Mayr, Yasemin Karatas, Rifat Karatas, Mustafa Adli, Susan M

Edwards, James D Wolff, Allen Movahed, Joseph Montebello, and William TC Yuh.

Inter-observer variation in cervical cancer tumor delineation for image-based radio-

therapy planning among and within different specialties. Journal of Applied Clinical

Medical Physics, 6(4), 2005. ISSN 1526-9914.

References 264

[232] Primož Petrič, Robert Hudej, Peter Rogelj, Mateja Blas, Kari Tanderup, Elena Fi-

darova, Christian Kirisits, Daniel Berger, Johannes Carl Athanasios Dimopoulos,

and Richard Pötter. Uncertainties of target volume delineation in mri guided adap-

tive brachytherapy of cervix cancer: A multi-institutional study. Radiotherapy and

Oncology, 107(1):6–12, 2013. ISSN 0167-8140.

[233] Taran Paulsen Hellebust, Kari Tanderup, Christoffer Lervåg, Elena Fidarova, Daniel

Berger, Eirik Malinen, Richard Pötter, and Primož Petrič. Dosimetric impact of

interobserver variability in mri-based delineation for cervical cancer brachytherapy.


[234] X Allen Li, An Tai, Douglas W Arthur, Thomas A Buchholz, Shannon Macdonald,

Lawrence B Marks, Jean M Moran, Lori J Pierce, Rachel Rabinovitch, and Alphonse

Taghian. Variability of target and normal structure delineation for breast cancer

radiotherapy: an rtog multi-institutional and multiobserver study. International


3016.

[235] Ross P Petersen, Pauline T Truong, Hosam A Kader, Eric Berthelet, Junella C Lee,

Michelle L Hilts, Adam S Kader, Wayne A Beckham, and Ivo A Olivotto. Target

volume delineation for partial breast radiotherapy planning: clinical characteristics

associated with low interobserver concordance. International Journal of Radiation


[236] John P Logue, Carole L Sharrock, Richard A Cowan, Graham Read, Julie Marrs,

and David Mott. Clinical variability of target volume description in conformal radio-

therapy planning. International Journal of Radiation Oncology* Biology* Physics,

41(4):929–932, 1998. ISSN 0360-3016.

[237] Anne E Saarnak, Menno Boersma, Bart NFM van Bunningen, René Wolterink, and

Marcel J Steggerda. Inter-observer variation in delineation of bladder and rectum

contours for brachytherapy of cervical cancer. Radiotherapy and Oncology, 56(1):

37–42, 2000. ISSN 0167-8140.

References 265

[238] Benjamin E Nelms, Wolfgang A Tomé, Greg Robinson, and James Wheeler. Varia-

tions in the contouring of organs at risk: test case from a patient with oropharyngeal

cancer. International Journal of Radiation Oncology* Biology* Physics, 82(1):368–

378, 2012. ISSN 0360-3016.

[239] Coen Rasch, Isabelle Barillot, Peter Remeijer, Adriaan Touw, Marcel van Herk, and

Joos V Lebesque. Definition of the prostate in ct and mri: a multi-observer study.


ISSN 0360-3016.

[240] Eric LH Khoo, Karlissa Schick, Ashley W Plank, Michael Poulsen, Winnie WG

Wong, Mark Middleton, and Jarad M Martin. Prostate contouring variation: can

it be fixed? International Journal of Radiation Oncology* Biology* Physics, 82(5):

1923–1929, 2012. ISSN 0360-3016.

[241] Stephen L Breen, Julia Publicover, Shiroma De Silva, Greg Pond, Kristy Brock,

Brian O’Sullivan, Bernard Cummings, Laura Dawson, Anne Keller, and John Kim.

Intraobserver and interobserver variability in gtv delineation on fdg-pet-ct images

of head and neck cancers. International Journal of Radiation Oncology* Biology*

Physics, 68(3):763–770, 2007. ISSN 0360-3016.

[242] Gabriela Altorjai, Irina Fotina, Carola Lütgendorf-Caucig, Markus Stock, Richard

Pötter, Dietmar Georg, and Karin Dieckmann. Cone-beam ct-based delineation

of stereotactic lung targets: the influence of image modality and target size on

interobserver variability. International Journal of Radiation Oncology* Biology*

Physics, 82(2):e265–e272, 2012. ISSN 0360-3016.

[243] Coen Rasch, Roel Steenbakkers, and Marcel van Herk. Target definition in prostate,

head, and neck. In Seminars in radiation oncology, volume 15, pages 136–145.

Elsevier, 2005. ISBN 1053-4296.

[244] Darren M Mitchell, Lesley Perry, Steve Smith, Tony Elliott, James P Wylie,

Richard A Cowan, Jacqueline E Livsey, and John P Logue. Assessing the effect

of a contouring protocol on postprostatectomy radiotherapy clinical target volumes

References 266

and interphysician variation. International Journal of Radiation Oncology* Biology*

Physics, 75(4):990–993, 2009. ISSN 0360-3016.

[245] Patrick Bowden, Richard Fisher, Michael Mac Manus, Andrew Wirth, Gillian Duch-

esne, Michael Millward, Allan McKenzie, Judy Andrews, and David Ball. Measure-

ment of lung tumor volumes using three-dimensional computer planning software.


ISSN 0360-3016.

[246] Edwin PM Jansen, Jasper Nijkamp, Michael Gubanski, Pehr ARM Lind, and Marcel

Verheij. Interobserver variation of clinical target volume delineation in gastric cancer.


2010. ISSN 0360-3016.

[247] Luigi Franco Cazzaniga, Maria Antonella Marinoni, Alberto Bossi, Ernestina

Bianchi, Emanuela Cagna, Dorian Cosentino, Luciano Scandolaro, Marica Valli, and

Milena Frigerio. Interphysician variability in defining the planning target volume in

the irradiation of prostate and seminal vesicles. Radiotherapy and oncology, 47(3):

293–296, 1998. ISSN 0167-8140.

[248] Benedikt Engels, Guy Soete, D Verellen, and Guy Storme. Conformal arc radio-

therapy for prostate cancer: increased biochemical failure in patients with distended

rectum on the planning computed tomogram despite image guidance by implanted

markers. International Journal of Radiation Oncology* Biology* Physics, 74(2):

388–391, 2009. ISSN 0360-3016.

[249] Zhanrong Gao, David Wilkins, Libni Eapen, Christopher Morash, Youssef Wassef,

and Lee Gerig. A study of prostate delineation referenced against a gold standard

created from the visible human data. Radiotherapy and oncology, 85(2):239–246,

2007. ISSN 0167-8140.

[250] Alistair M Pooler, Helen M Mayles, Olivia F Naismith, John P Sage, and David P

Dearnaley. Evaluation of margining algorithms in commercial treatment planning

systems. Radiotherapy and Oncology, 86(1):43–47, 2008. ISSN 0167-8140.

References 267

[251] Shalini K Vinod, Myo Min, Michael G Jameson, and Lois C Holloway. A review of

interventions to reduce inter‐observer variability in volume delineation in radiation

oncology. Journal of medical imaging and radiation oncology, 60(3):393–406, 2016.

ISSN 1754-9485.

[252] Claudio Fiorino, Michele Reni, Angelo Bolognesi, Giovanni Mauro Cattaneo, and

Riccardo Calandrino. Intra-and inter-observer variability in contouring prostate and

seminal vesicles: implications for conformal treatment planning. Radiotherapy and

oncology, 47(3):285–292, 1998. ISSN 0167-8140.

[253] Peter Remeijer, Coen Rasch, Joos V Lebesque, and Marcel van Herk. A general

methodology for three-dimensional analysis of variation in target volume delineation.

Medical Physics, 26(6):931–940, 1999. ISSN 0094-2405.

[254] Beatrice Seddon, Margaret Bidmead, Jane Wilson, Vincent Khoo, and David Dear-

naley. Target volume definition in conformal radiotherapy for prostate cancer: qual-

ity assurance in the mrc rt-01 trial. Radiotherapy and Oncology, 56(1):73–83, 2000.

ISSN 0167-8140.

[255] Piet Ost, Gert De Meerleer, Tom Vercauteren, Werner De Gersem, Liv Veldeman,

Katrien Vandecasteele, Valérie Fonteyne, and Geert Villeirs. Delineation of the post-

prostatectomy prostate bed using computed tomography: interobserver variability

following the eortc delineation guidelines. International Journal of Radiation On-

cology* Biology* Physics, 81(3):e143–e149, 2011. ISSN 0360-3016.

[256] Geert M Villeirs, Koen Van Vaerenbergh, Luc Vakaet, Samuel Bral, Filip Claus,

Wilfried J De Neve, Koenraad L Verstraete, and Gert O De Meerleer. Interobserver

delineation variation using ct versus combined ct+ mri in intensity–modulated radio-

therapy for prostate cancer. Strahlentherapie und Onkologie, 181(7):424–430, 2005.

ISSN 0179-7158.

[257] Mack Roach, Pamela Faillace-Akazawa, Christine Malfatti, John Holland, and Hed-

vig Hricak. Prostate volumes defined by magnetic resonance imaging and computer-

ized tomographic scans for three-dimensional conformal radiotherapy. International

References 268


0360-3016.

[258] Carola Lütgendorf-Caucig, Irina Fotina, Markus Stock, Richard Pötter, Gregor

Goldner, and Dietmar Georg. Feasibility of cbct-based target and normal struc-

ture delineation in prostate cancer radiotherapy: multi-observer and image multi-

modality study. Radiotherapy and Oncology, 98(2):154–161, 2011. ISSN 0167-8140.

[259] Donald F Dubois, Bradley R Prestidge, Laura A Hotchkiss, James J Prete, and

William S Bice Jr. Intraobserver and interobserver variability of mr imaging-

and ct-derived prostate volumes after transperineal interstitial permanent prostate

brachytherapy. Radiology, 207(3):785–789, 1998. ISSN 0033-8419.

[260] Michael Milosevic, Sachi Voruganti, Ralph Blend, Hamideh Alasti, Padraig Warde,

Michael McLean, Pamela Catton, Charles Catton, and Mary Gospodarowicz. Mag-

netic resonance imaging (mri) for localization of the prostatic apex: comparison to

computed tomography (ct) and urethrography. Radiotherapy and oncology, 47(3):

277–284, 1998. ISSN 0167-8140.

[261] Jana Sefrova, Karel Odrazka, Petr Paluska, Zdenek Belobradek, Milos Brodak, Mar-

tin Dolezel, Petr Prosvic, Zuzana Macingova, Milan Vosmik, and Petr Hoffmann.

Magnetic resonance imaging in postprostatectomy radiotherapy planning. Interna-


0360-3016.

[262] Richard K Valicenti, John W Sweet, Walter W Hauck, Richard S Hudes, Tony Lee,

Adam P Dicker, Frank M Waterman, Pramila R Anne, Benjamin W Corn, and

James M Galvin. Variation of clinical target volume definition in three-dimensional

conformal radiation therapy for prostate cancer. International Journal of Radiation


[263] Elise M Pogson, Jarrad Begg, Michael G Jameson, Claire Dempsey, Drew Latty,

Vikneswary Batumalai, Andrew Lim, Kankean Kandasamy, Peter E Metcalfe, and

Lois C Holloway. A phantom assessment of achievable contouring concordance across

References 269

multiple treatment planning systems. Radiotherapy and Oncology, 2015. ISSN 0167-

8140.

[264] Lili Chen, Robert A Price, Lu Wang, Jinsheng Li, Lihong Qin, Shawn McNeeley, C-

M Charlie Ma, Gary M Freedman, and Alan Pollack. Mri-based treatment planning

for radiotherapy: dosimetric verification for prostate imrt. International Journal of

Radiation Oncology* Biology* Physics, 60(2):636–647, 2004. ISSN 0360-3016.

[265] Patrick W McLaughlin, Cheryl Evans, Mary Feng, and Vrinda Narayana. Radio-

graphic and anatomic basis for prostate contouring errors and methods to improve

prostate contouring accuracy. International Journal of Radiation Oncology* Biol-

ogy* Physics, 76(2):369–378, 2010. ISSN 0360-3016.

[266] Claudio Fiorino, Vittorio Vavassori, Giuseppe Sanguineti, Carla Bianchi, Gio-

vanni Mauro Cattaneo, Anna Piazzolla, and Cesare Cozzarini. Rectum contouring

variability in patients treated for prostate cancer: impact on rectum dose–volume

histograms and normal tissue complication probability. Radiotherapy and oncology,

63(3):249–255, 2002. ISSN 0167-8140.

[267] Marcel van Herk, Peter Remeijer, and Joos V Lebesque. Inclusion of geometric uncer-

tainties in treatment plan evaluation. International Journal of Radiation Oncology*


[268] Kevin L Moore, R Scott Brame, Daniel A Low, and Sasa Mutic. Quantitative metrics

for assessing plan quality. In Seminars in radiation oncology, volume 22, pages 62–

69. Elsevier, 2012. ISBN 1053-4296. doi: https://doi.org/10.1016/j.semradonc.2011.

09.005.

[269] Jörg Bohsung, Sofie Gillis, Rafael Arrans, Annemarie Bakai, Carlos De Wagter,

Tommy Knöös, Ben J Mijnheer, Marta Paiusco, Bruce A Perrin, and Hans Welle-

weerd. Imrt treatment planning—a comparative inter-system and inter-centre plan-

ning exercise of the estro quasimodo group. Radiotherapy and oncology, 76(3):354–

361, 2005. ISSN 0167-8140.

References 270

[270] MJ Williams, MJ Bailey, D Forstner, and Peter E Metcalfe. Multicentre quality

assurance of intensity‐modulated radiation therapy plans: A precursor to clinical

trials. Australasian radiology, 51(5):472–479, 2007. ISSN 0004-8461.

[271] Indra J Das, Chee-Wai Cheng, Kashmiri L Chopra, Raj K Mitra, Shiv P Srivas-

tava, and Eli Glatstein. Intensity-modulated radiation therapy dose prescription,

recording, and delivery: patterns of variability among institutions and treatment

planning systems. Journal of the National Cancer Institute, 100(5):300–307, 2008.

ISSN 0027-8874.

[272] B Nelms, J Caudell, D Gintz, and V Feygelman. Head and neck plan quality: where

we are and does it matter? In Journal of Physics: Conference Series, volume 1154,

page 012018. IOP Publishing, 2019. ISBN 1742-6596.

[273] Vikneswary Batumalai, Michael G Jameson, Dion F Forstner, Philip Vial, and Lois C

Holloway. How important is dosimetrist experience for intensity modulated radia-

tion therapy? a comparative analysis of a head and neck case. Practical radiation

oncology, 3(3):e99–e106, 2013. ISSN 1879-8500. doi: https://doi.org/10.1016/j.prro.

2012.06.009.

[274] Roel GJ Kierkels, Ruurd Visser, Hendrik P Bijl, Johannes A Langendijk, Aart A

van‘t Veld, Roel JHM Steenbakkers, and Erik W Korevaar. Multicriteria optimiza-

tion enables less experienced planners to efficiently produce high quality treatment

plans in head and neck cancer radiotherapy. Radiation oncology, 10(1):87, 2015.

ISSN 1748-717X.

[275] David L Craft, Theodore S Hong, Helen A Shih, and Thomas R Bortfeld. Improved

planning time and plan quality through multicriteria optimization for intensity-

modulated radiotherapy. International Journal of Radiation Oncology* Biology*

Physics, 82(1):e83–e90, 2012. ISSN 0360-3016.

[276] N Alves, T Ventura, J Mateus, M Capela, and MC Lopes. Do a priori expectations

of plan quality offset planning variability in head and neck imrt? Radiation Physics

and Chemistry, page 108580, 2019. ISSN 0969-806X.

References 271

[277] Radar trial - randomised androgen deprivation and radiotherapy, 2005. URL https:

//clinicaltrials.gov/ct2/show/study/NCT00193856.

[278] James W Denham, Allison Steigler, David S Lamb, David Joseph, Hedy Mameghan,

Sandra Turner, John Matthews, Ian Franklin, Chris Atkinson, and John North.

Short-term androgen deprivation and radiotherapy for locally advanced prostate

cancer: results from the trans-tasman radiation oncology group 96.01 randomised

controlled trial. The lancet oncology, 6(11):841–850, 2005. ISSN 1470-2045.

[279] James W Denham, Allison Steigler, David S Lamb, David Joseph, Sandra Turner,

John Matthews, Chris Atkinson, John North, David Christie, and Nigel A Spry.

Short-term neoadjuvant androgen deprivation and radiotherapy for locally advanced

prostate cancer: 10-year data from the trog 96.01 randomised trial. The lancet

oncology, 12(5):451–459, 2011. ISSN 1470-2045.

[280] James W Denham, David Joseph, David S Lamb, Nigel A Spry, Gillian Duchesne,

John Matthews, Chris Atkinson, Keen-Hun Tai, David Christie, and Lizbeth Kenny.

Short-term androgen suppression and radiotherapy versus intermediate-term andro-

gen suppression and radiotherapy, with or without zoledronic acid, in men with lo-

cally advanced prostate cancer (trog 03.04 radar): an open-label, randomised, phase

3 factorial trial. The Lancet Oncology, 15(10):1076–1089, 2014. ISSN 1470-2045.

[281] James W Denham, Chantelle Wilcox, David Joseph, Nigel A Spry, David S Lamb,

Keen-Hun Tai, John Matthews, Chris Atkinson, Sandra Turner, and David Christie.

Quality of life in men with locally advanced prostate cancer treated with leuprorelin

and radiotherapy with or without zoledronic acid (trog 03.04 radar): secondary

endpoints from a randomised phase 3 factorial trial. The Lancet Oncology, 13(12):

1260–1270, 2012. ISSN 1470-2045.

[282] James W Denham, Chantelle Wilcox, David S Lamb, Nigel A Spry, Gillian Duchesne,

Chris Atkinson, John Matthews, Sandra Turner, Lizbeth Kenny, and Keen-Hun Tai.

Rectal and urinary dysfunction in the trog 03.04 radar trial for locally advanced

prostate cancer. Radiotherapy and Oncology, 105(2):184–192, 2012. ISSN 0167-8140.

https://clinicaltrials.gov/ct2/show/study/NCT00193856

https://clinicaltrials.gov/ct2/show/study/NCT00193856

References 272

[283] Annette Haworth, Rachel Kearvell, Peter B Greer, Ben Hooton, James W Denham,

David Lamb, Gillian Duchesne, Judy Murray, and David Joseph. Assuring high

quality treatment delivery in clinical trials–results from the trans-tasman radiation

oncology group (trog) study 03.04 “radar” set-up accuracy study. Radiotherapy and

Oncology, 90(3):299–306, 2009. ISSN 0167-8140.

[284] James W Denham, Allison Steigler, David Joseph, David S Lamb, Nigel A Spry,

Gillian Duchesne, Chris Atkinson, John Matthews, Sandra Turner, and Lizbeth

Kenny. Radiation dose escalation or longer androgen suppression for locally advanced

prostate cancer? data from the trog 03.04 radar trial. Radiotherapy and Oncology,

115(3):301–307, 2015. ISSN 0167-8140.

[285] Rachel Kearvell, Annette Haworth, Martin A Ebert, Judy Murray, Ben Hooton,

Sharon Richardson, David J Joseph, David Lamb, Nigel A Spry, and Gillian Duch-

esne. Quality improvements in prostate radiotherapy: outcomes and impact of com-

prehensive quality assurance during the trog 03.04 ‘radar’trial. Journal of medical

imaging and radiation oncology, 57(2):247–257, 2013. ISSN 1754-9485.

[286] SE Griffiths, S Stanley, and M Sydes. Recommendations on best practice for ra-

diographer set-up of conformal radiotherapy treatment for patients with prostate

cancer: experience developed during the mrc rt01 trial (isrtcn 47772397). Journal

of Radiotherapy in Practice, 4(2-3):107–117, 2004. ISSN 1467-1131.

[287] Martin A Ebert, Annette Haworth, Rachel Kearvell, Ben Hooton, Rhonda Coleman,

Nigel Spry, Sean Bydder, and David Joseph. Detailed review and analysis of complex

radiotherapy clinical trial planning data: evaluation and initial experience with the

swan software system. Radiotherapy and Oncology, 86(2):200–210, 2008. ISSN 0167-

8140.

[288] M Skala, M Berry, G Duchesne, K Gogna, K‐H Tai, S Turner, A Kneebone, A Rolfo,

and A Haworth. Australian and new zealand three‐dimensional conformal radiation

therapy consensus guidelines for prostate cancer. Australasian radiology, 48(4):493–

501, 2004. ISSN 1440-1673.

References 273

[289] Martin A Ebert, Max Bulsara, Annette Haworth, Rachel Kearvell, Sharon Richard-

son, Angel Kennedy, Nigel A Spry, Sean A Bydder, David J Joseph, and James W

Denham. Technical quality assurance during the trog 03.04 radar prostate radiother-

apy trial: Are the results reflected in observed toxicity rates? Journal of medical

imaging and radiation oncology, 59(1):99–108, 2015. ISSN 1754-9485.

[290] Marco Marcello, Martin Ebert, Annette Haworth, Allison Steigler, Angel Kennedy,

David Joseph, and James Denham. Association between treatment planning and

delivery factors and disease progression in prostate cancer radiotherapy: Results

from the trog 03.04 radar trial. Radiotherapy and Oncology, 126(2):249–256, 2018.

ISSN 0167-8140.

[291] Noorazrul Yahya, Martin A Ebert, Max Bulsara, Annette Haworth, Rachel Kearvell,

Kerwyn Foo, Angel Kennedy, Sharon Richardson, Michele Krawiec, and David J

Joseph. Impact of treatment planning and delivery factors on gastrointestinal tox-

icity: an analysis of data from the radar prostate radiotherapy trial. Radiation

Oncology, 9(1):282, 2014. ISSN 1748-717X.

[292] Noorazrul Yahya, Martin A Ebert, Max Bulsara, Annette Haworth, Angel Kennedy,

David J Joseph, and Jim W Denham. Dosimetry, clinical factors and medication

intake influencing urinary symptoms after prostate radiotherapy: An analysis of

data from the radar prostate radiotherapy trial. Radiotherapy and Oncology, 116(1):

112–118, 2015. ISSN 0167-8140.

[293] David P Dearnaley, Matthew R Sydes, John D Graham, Edwin G Aird, David Bot-

tomley, Richard A Cowan, Robert A Huddart, Chakiath C Jose, John HL Matthews,

and Jeremy Millar. Escalated-dose versus standard-dose conformal radiotherapy in

prostate cancer: first results from the mrc rt01 randomised controlled trial. The

lancet oncology, 8(6):475–487, 2007. ISSN 1470-2045.

[294] David P Dearnaley, Gordana Jovic, Isabel Syndikus, Vincent Khoo, Richard A

Cowan, John D Graham, Edwin G Aird, David Bottomley, Robert A Huddart,

and Chakiath C Jose. Escalated-dose versus control-dose conformal radiotherapy

for prostate cancer: long-term results from the mrc rt01 randomised controlled trial.

The lancet oncology, 15(4):464–473, 2014. ISSN 1470-2045.

References 274

[295] Deborah A Kuban, Susan L Tucker, Lei Dong, George Starkschall, Eugene H Huang,

M Rex Cheung, Andrew K Lee, and Alan Pollack. Long-term results of the md

anderson randomized dose-escalation trial for prostate cancer. International Journal


[296] Anthony L Zietman, Michelle L DeSilvio, Jerry D Slater, Carl J Rossi, Daniel W

Miller, Judith A Adams, and William U Shipley. Comparison of conventional-dose

vs high-dose conformal radiation therapy in clinically localized adenocarcinoma of

the prostate: a randomized controlled trial. Jama, 294(10):1233–1239, 2005. ISSN

0098-7484.

[297] Michael J Zelefsky, Steven A Leibel, Paul B Gaudin, Gerald J Kutcher, Neil E

Fleshner, E S Venkatramen, Victor E Reuter, William R Fair, C Clifton Ling, and Zvi

Fuks. Dose escalation with three-dimensional conformal radiation therapy affects the

outcome in prostate cancer. International Journal of Radiation Oncology* Biology*

Physics, 41(3):491–500, 1998. ISSN 0360-3016.

[298] Matthew R Sydes, Richard J Stephens, A Rollo Moore, Edwin G Aird, A Margaret

Bidmead, Lesley J Fallowfield, John Graham, Sue Griffiths, W Philip Mayles, and

Ali McGuire. Implementing the uk medical research council (mrc) rt01 trial (isrctn

47772397): methods and practicalities of a randomised controlled trial of conformal

radiotherapy in men with localised prostate cancer. Radiotherapy and oncology, 72

(2):199–211, 2004. ISSN 0167-8140.

[299] David P Dearnaley, Matthew R Sydes, Ruth E Langley, John D Graham, Robert A

Huddart, Isabel Syndikus, John HL Matthews, Christopher D Scrase, Chakiath C

Jose, and John Logue. The early toxicity of escalated versus standard dose conformal

radiotherapy with neo-adjuvant androgen suppression for patients with localised

prostate cancer: results from the mrc rt01 trial (isrctn47772397). Radiotherapy and

oncology, 83(1):31–41, 2007. ISSN 0167-8140.

[300] GC Barnett, Gert De Meerleer, SL Gulliford, MR Sydes, Rebecca M Elliott, and

DP Dearnaley. The impact of clinical factors on the development of late radia-

tion toxicity: results from the medical research council rt01 trial (isrctn47772397).

Clinical Oncology, 23(9):613–624, 2011. ISSN 0936-6555.

References 275

[301] Stephen A Mangar, Matthew R Sydes, Helen L Tucker, Jerome Coffey, Syed A

Sohaib, Stefano Gianolini, Steve Webb, Vincent S Khoo, MRC RT01 Trial Manage-

ment Group, and David P Dearnaley. Evaluating the relationship between erectile

dysfunction and dose received by the penile bulb: using data from a randomised con-

trolled trial of conformal radiotherapy in prostate cancer (mrc rt01, isrctn47772397).

Radiotherapy and oncology, 80(3):355–362, 2006. ISSN 0167-8140.

[302] Sarah L Gulliford, Kerwyn Foo, Rachel C Morgan, Edwin G Aird, A Margaret

Bidmead, Helen Critchley, Philip M Evans, Stefano Gianolini, W Philip Mayles, and

A Rollo Moore. Dose–volume constraints to reduce rectal side effects from prostate

radiotherapy: evidence from mrc rt01 trial isrctn 47772397. International Journal


[303] Florian Buettner, Sarah L Gulliford, Steve Webb, Matthew R Sydes, David P Dear-

naley, and Mike Partridge. Assessing correlations between the spatial distribution

of the dose to the rectal wall and late rectal toxicity after prostate radiotherapy: an

analysis of data from the mrc rt01 trial (isrctn 47772397). Physics in Medicine&

Biology, 54(21):6535, 2009. ISSN 0031-9155.

[304] Florian Buettner, Sarah L Gulliford, Steve Webb, Matthew R Sydes, David P Dear-

naley, and Mike Partridge. The dose–response of the anal sphincter region–an anal-

ysis of data from the mrc rt01 trial. Radiotherapy and Oncology, 103(3):347–352,

2012. ISSN 0167-8140.

[305] W Philip M Mayles, A Rollo Moore, Edwin GA Aird, A Margaret Bidmead, David P

Dearnaley, Sue E Griffiths, Richard J Stephens, and AP Jim Warrington. Ques-

tionnaire based quality assurance for the rt01 trial of dose escalation in conformal

radiotherapy for prostate cancer (isrctn 47772397). Radiotherapy and oncology, 73

(2):199–207, 2004. ISSN 0167-8140.

[306] A Rollo Moore, AP Jim Warrington, Edwin GA Aird, A Margaret Bidmead, and

David P Dearnaley. A versatile phantom for quality assurance in the uk medi-

cal research council (mrc) rt01 trial (isrctn47772397) in conformal radiotherapy for

prostate cancer. Radiotherapy and oncology, 80(1):82–85, 2006. ISSN 0167-8140.

References 276

[307] S Stanley, S Griffiths, MR Sydes, AR Moore, I Syndikus, and DP Dearnaley.

Accuracy and reproducibility of conformal radiotherapy using data from a ran-

domised controlled trial of conformal radiotherapy in prostate cancer (mrc rt01,

isrctn47772397). Clinical Oncology, 20(8):582–590, 2008. ISSN 0936-6555.

[308] David Dearnaley, Isabel Syndikus, Helen Mossop, Vincent Khoo, Alison Birtle,

David Bloomfield, John Graham, Peter Kirkbride, John Logue, and Zafar Malik.

Conventional versus hypofractionated high-dose intensity-modulated radiotherapy

for prostate cancer: 5-year outcomes of the randomised, non-inferiority, phase 3

chhip trial. The Lancet Oncology, 17(8):1047–1060, 2016. ISSN 1470-2045.

[309] Bridget F Koontz, Alberto Bossi, Cesare Cozzarini, Thomas Wiegel, and Anthony

D’Amico. A systematic review of hypofractionation for primary management of

prostate cancer. European urology, 68(4):683–691, 2015. ISSN 0302-2838.

[310] Nicholas G Zaorsky, Nitin Ohri, Timothy N Showalter, Adam P Dicker, and Robert B

Den. Systematic review of hypofractionated radiation therapy for prostate cancer.

Cancer treatment reviews, 39(7):728–736, 2013. ISSN 0305-7372.

[311] VS Khoo and DP Dearnaley. Question of dose, fractionation and technique: in-

gredients for testing hypofractionation in prostate cancer–the chhip trial. Clinical

oncology, 20(1):12–14, 2008. ISSN 0936-6555.

[312] David Dearnaley, Isabel Syndikus, Georges Sumo, Margaret Bidmead, David Bloom-

field, Catharine Clark, Annie Gao, Shama Hassan, Alan Horwich, and Robert Hud-

dart. Conventional versus hypofractionated high-dose intensity-modulated radio-

therapy for prostate cancer: preliminary safety results from the chhip randomised

controlled trial. The lancet oncology, 13(1):43–54, 2012. ISSN 1470-2045.

[313] David Dearnaley, Helen Mossop, Isabel Syndikus, Vincent Khoo, DJ Bloomfield,

Chris C Parker, John Logue, Christopher D Scrase, Alison J Birtle, and John

Staffurth. 5-year patient-reported outcomes of bowel and urinary bother in the

chhip trial (cruk/06/016). Journal of Clinical Oncology, 2017.

References 277

[314] James M Wilson, David P Dearnaley, Isabel Syndikus, Vincent Khoo, Alison Birtle,

David Bloomfield, Ananya Choudhury, John Graham, Catherine Ferguson, and Za-

far Malik. The efficacy and safety of conventional and hypofractionated high-dose

radiation therapy for prostate cancer in an elderly population: A subgroup analysis

of the chhip trial. International Journal of Radiation Oncology* Biology* Physics,

100(5):1179–1189, 2018. ISSN 0360-3016.

[315] OF Naismith, C Griffin, I Syndikus, C South, H Mayles, P Mayles, V Khoo, C Scrase,

J Graham, and S Hassan. Forward-and inverse-planned intensity-modulated radio-

therapy in the chhip trial: A comparison of dosimetry and normal tissue toxicity.

Clinical Oncology, 2019. ISSN 0936-6555.

[316] David Dearnaley and Emma Hall. How will the chhip trial affect the future of

prostate radiotherapy? Expert Review of Anticancer Therapy, 2018.

[317] O Naismith, H Mayles, M Bidmead, CH Clark, S Gulliford, S Hassan, V Khoo,

K Roberts, C South, and E Hall. Radiotherapy quality assurance for the chhip trial:

Conventional versus hypofractionated high-dose intensity-modulated radiotherapy

in prostate cancer. Clinical Oncology, 2019. ISSN 0936-6555.

[318] Christos Davatzikos, Dengfeng Liu, Dinggang Shen, and Edward H Herskovits. Spa-

tial normalization of spine mr images for statistical correlation of lesions with clinical

symptoms. Radiology, 224(3):919–926, 2002. ISSN 0033-8419.

[319] Dinggang Shen, Edward H Herskovits, and Christos Davatzikos. An adaptive-focus

statistical shape model for segmentation and shape modeling of 3-d brain structures.

IEEE transactions on medical imaging, 20(4):257–270, 2001. ISSN 0278-0062.

[320] Dominik Fritz, Daniel Rinck, Rüdiger Dillmann, and Michael Scheuering. Segmen-

tation of the left and right cardiac ventricle using a combined bi-temporal statistical

model. In Medical Imaging 2006: Visualization, Image-Guided Procedures, and Dis-

play, volume 6141, page 614121. International Society for Optics and Photonics,

2006.

References 278

[321] Yiqiang Zhan and Dinggang Shen. Deformable segmentation of 3-d ultrasound

prostate images using statistical texture matching method. IEEE Transactions on

Medical Imaging, 25(3):256–272, 2006. ISSN 0278-0062.

[322] Vladimir Pekar, Michael R Kaus, Cristian Lorenz, Steven Lobregt, Roel Truyen,

and Juergen Weese. Shape-model-based adaptation of 3d deformable meshes for

segmentation of medical images. InMedical Imaging 2001: Image Processing, volume

4322, pages 281–289. International Society for Optics and Photonics, 2001.

[323] Berkman Sahiner, Aria Pezeshk, Lubomir M Hadjiiski, Xiaosong Wang, Karen

Drukker, Kenny H Cha, Ronald M Summers, and Maryellen L Giger. Deep learning

in medical imaging and radiation therapy. Medical physics, 46(1):e1–e36, 2019. ISSN

0094-2405.

[324] Carlos E Cardenas, Jinzhong Yang, Brian M Anderson, Laurence E Court, and

Kristy B Brock. Advances in auto-segmentation. In Seminars in radiation oncology,


[325] Kunihiko Fukushima. Neocognitron: A self-organizing neural network model for a

mechanism of pattern recognition unaffected by shift in position. Biological cyber-

netics, 36(4):193–202, 1980. ISSN 0340-1200.

[326] Matthew Lai. Deep learning for medical image segmentation. arXiv preprint

arXiv:1505.02000, 2015.

[327] Erik Smistad, Thomas L Falch, Mohammadmehdi Bozorgi, Anne C Elster, and Frank

Lindseth. Medical image segmentation on gpus–a comprehensive review. Medical

image analysis, 20(1):1–18, 2015. ISSN 1361-8415.

[328] MR Avendi, Arash Kheradvar, and Hamid Jafarkhani. A combined deep-learning

and deformable-model approach to fully automatic segmentation of the left ventricle

in cardiac mri. Medical image analysis, 30:108–119, 2016. ISSN 1361-8415.

[329] Pim Moeskops, Max A Viergever, Adriënne M Mendrik, Linda S de Vries,

Manon JNL Benders, and Ivana Išgum. Automatic segmentation of mr brain images

with a convolutional neural network. IEEE transactions on medical imaging, 35(5):

1252–1261, 2016. ISSN 0278-0062.

References 279

[330] Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional net-

works for biomedical image segmentation. In International Conference on Medical

image computing and computer-assisted intervention, pages 234–241. Springer, 2015.

[331] Ruida Cheng, Holger R Roth, Le Lu, Shijun Wang, Baris Turkbey, William Gandler,

Evan S McCreedy, Harsh K Agarwal, Peter Choyke, and Ronald M Summers. Active

appearance model and deep learning for more accurate prostate segmentation on mri.

InMedical Imaging 2016: Image Processing, volume 9784, page 97842I. International

Society for Optics and Photonics, 2016.

[332] Mohamed S Elmahdy, Thyrza Jagt, R Th Zinkstok, Yuchuan Qiao, Rahil Shahzad,

Hessam Sokooti, Sahar Yousefi, Luca Incrocci, CAM Marijnen, and Mischa Hooge-

man. Robust contour propagation using deep learning and image registration for

online adaptive proton therapy of prostate cancer. Medical physics, 2019. ISSN

0094-2405.

[333] Ruida Cheng, Holger R Roth, Nathan S Lay, Le Lu, Baris Turkbey, William Gandler,

Evan S McCreedy, Thomas J Pohida, Peter A Pinto, and Peter L Choyke. Automatic

magnetic resonance prostate segmentation by deep learning with holistically nested

networks. Journal of Medical Imaging, 4(4):041302, 2017. ISSN 2329-4302.

[334] Zhiqiang Tian, Lizhi Liu, Zhenfeng Zhang, and Baowei Fei. Psnet: prostate segmen-

tation on mri based on a convolutional neural network. Journal of Medical Imaging,

5(2):021208, 2018. ISSN 2329-4302.

[335] Qikui Zhu, Bo Du, Baris Turkbey, Peter L Choyke, and Pingkun Yan. Deeply-

supervised cnn for prostate segmentation. In 2017 International Joint Conference

on Neural Networks (Ijcnn), pages 178–184. IEEE, 2017. ISBN 1509061827.

[336] Shu Liao, Yaozong Gao, Aytekin Oto, and Dinggang Shen. Representation learn-

ing: a unified deep learning framework for automatic prostate mr segmentation.

In International Conference on Medical Image Computing and Computer-Assisted

Intervention, pages 254–261. Springer, 2013.

References 280

[337] Yanrong Guo, Yaozong Gao, and Dinggang Shen. Deformable mr prostate segmen-

tation via deep feature learning and sparse patch matching. IEEE transactions on

medical imaging, 35(4):1077–1089, 2015. ISSN 0278-0062.

[338] Emran Mohammad Abu Anas, Parvin Mousavi, and Purang Abolmaesumi. A deep

learning approach for real time prostate segmentation in freehand ultrasound guided

biopsy. Medical image analysis, 48:107–116, 2018. ISSN 1361-8415.

[339] Ling Ma, Rongrong Guo, Guoyi Zhang, Funmilayo Tade, David M Schuster, Peter

Nieh, Viraj Master, and Baowei Fei. Automatic segmentation of the prostate on

ct images using deep learning and multi-atlas fusion. In Medical Imaging 2017:

Image Processing, volume 10133, page 101332O. International Society for Optics

and Photonics, 2017.

[340] Samaneh Kazemifar, Anjali Balagopal, Dan Nguyen, Sarah McGuire, Raquibul Han-

nan, Steve Jiang, and Amir Owrangi. Segmentation of the prostate and organs at

risk in male pelvic ct images using deep learning. Biomedical Physics & Engineering

Express, 4(5):055003, 2018. ISSN 2057-1976.

[341] Sharif Elguindi, Michael J Zelefsky, Jue Jiang, Harini Veeraraghavan, Joseph O

Deasy, Margie A Hunt, and Neelam Tyagi. Deep learning-based auto-segmentation of

targets and organs-at-risk for magnetic resonance imaging only planning of prostate

radiotherapy. Physics and Imaging in Radiation Oncology, 12:80–86, 2019. ISSN

2405-6316.

[342] Shekhar S Chandra, Jason A Dowling, Kai-Kai Shen, Parnesh Raniga, Josien PW

Pluim, Peter B Greer, Olivier Salvado, and Jurgen Fripp. Patient specific prostate

segmentation in 3-d magnetic resonance images. Medical Imaging, IEEE Transac-

tions on, 31(10):1955–1964, 2012. ISSN 0278-0062.

[343] JA Dowling, N Burdett, PB Greer, J Sun, J Parker, P Pichler, P Stanwell, S Chan-

dra, D Rivest-Hénault, and S Ghose. Automatic atlas based electron density and

structure contouring for mri-based prostate radiation therapy on the cloud. In Jour-

nal of Physics: Conference Series, volume 489, page 012048. IOP Publishing, 2014.

ISBN 1742-6596.

References 281

[344] Jason A Dowling, Jonathan Lambert, Joel Parker, Olivier Salvado, Jurgen Fripp,

Anne Capp, Chris Wratten, James W Denham, and Peter B Greer. An atlas-based

electron density mapping method for magnetic resonance imaging (mri)-alone treat-

ment planning and adaptive mri-based prostate radiation therapy. International

Journal of Radiation Oncology* Biology* Physics, 83(1):e5–e11, 2012. ISSN 0360-

3016.

[345] Jason A Dowling, Jidi Sun, Peter Pichler, David Rivest-Hénault, Soumya Ghose,

Haylea Richardson, Chris Wratten, Jarad Martin, Jameen Arm, and Leah Best.

Automatic substitute computed tomography generation and contouring for magnetic

resonance imaging (mri)-alone external beam radiation therapy from standard mri

sequences. International Journal of Radiation Oncology* Biology* Physics, 93(5):

1144–1153, 2015. ISSN 0360-3016.

[346] Paul Aljabar, Rolf A Heckemann, Alexander Hammers, Joseph V Hajnal, and Daniel

Rueckert. Multi-atlas based segmentation of brain images: atlas selection and its

effect on accuracy. Neuroimage, 46(3):726–738, 2009. ISSN 1053-8119.

[347] Rolf A Heckemann, Joseph V Hajnal, Paul Aljabar, Daniel Rueckert, and Alexander

Hammers. Automatic anatomical brain mri segmentation combining label propaga-

tion and decision fusion. NeuroImage, 33(1):115–126, 2006. ISSN 1053-8119.

[348] Minjie Wu, Caterina Rosano, Pilar Lopez-Garcia, Cameron S Carter, and Howard J

Aizenstein. Optimum template selection for atlas-based segmentation. Neuroimage,

34(4):1612–1618, 2007. ISSN 1053-8119.

[349] Chao Lu, Sudhakar Chelikani, Xenophon Papademetris, Jonathan P Knisely,

Michael F Milosevic, Zhe Chen, David A Jaffray, Lawrence H Staib, and James S

Duncan. An integrated approach to segmentation and nonrigid registration for appli-

cation in image-guided pelvic radiotherapy. Medical Image Analysis, 15(5):772–785,

2011. ISSN 1361-8415.

[350] Jeremiah Hwee, Alexander V Louie, Stewart Gaede, Glenn Bauman, David D’Souza,

Tracy Sexton, Michael Lock, Belal Ahmad, and George Rodrigues. Technology

References 282

assessment of automated atlas based segmentation in prostate bed contouring. Radiat

Oncol, 6:110, 2011.

[351] Rongrong Zhou, Zhongxing Liao, Tinsu Pan, Sarah A Milgrom, Chelsea C Pinnix,

Anhui Shi, Linglong Tang, Ju Yang, Ying Liu, and Daniel Gomez. Cardiac atlas

development and validation for automatic segmentation of cardiac substructures.


[352] Robert Finnegan, Jason Dowling, Eng-Siew Koh, Simon Tang, James Otton, Ge-

off Delaney, Vikneswary Batumalai, Carol Luo, Pramukh Atluri, and Athiththa

Satchithanandha. Feasibility of multi-atlas cardiac segmentation from thoracic plan-

ning ct in a probabilistic framework. Physics in Medicine& Biology, 64(8):085006,

2019. ISSN 0031-9155.

[353] Torsten Rohlfing, Robert Brandt, Randolf Menzel, Daniel B Russakoff, and Calvin R

Maurer Jr. Quo vadis, atlas-based segmentation?, pages 435–486. Springer, 2005.

ISBN 0306486075.

[354] Torsten Rohlfing, Robert Brandt, Randolf Menzel, and Calvin R Maurer. Evaluation

of atlas selection strategies for atlas-based image segmentation with application to

confocal microscopy images of bee brains. NeuroImage, 21(4):1428–1442, 2004. ISSN

1053-8119.

[355] Dzung L Pham, Chenyang Xu, and Jerry L Prince. Current methods in medical

image segmentation. Annual review of biomedical engineering, 2(1):315–337, 2000.

ISSN 1523-9829.

[356] Nhat Trung Doan, Jonathan Orban de Xivry, and Benoît Macq. Effect of inter-

subject variation on the accuracy of atlas-based segmentation applied to human

brain structures. In Medical Imaging 2010: Image Processing, volume 7623, page

76231S. International Society for Optics and Photonics, 2010.

[357] Torsten Rohlfing, Robert Brandt, Randolf Menzel, and Calvin R Maurer Jr. Seg-

mentation of three-dimensional images using non-rigid registration: Methods and

validation with application to confocal microscopy images of bee brains. In Medical

References 283

Imaging 2003: Image Processing, volume 5032, pages 363–374. International Society

for Optics and Photonics, 2003.

[358] John Ashburner and Karl J Friston. Unified segmentation. Neuroimage, 26(3):

839–851, 2005. ISSN 1053-8119.

[359] Hyunjin Park, Peyton H Bland, and Charles R Meyer. Construction of an abdominal

probabilistic atlas and its application in segmentation. IEEE Transactions on medical

imaging, 22(4):483–492, 2003. ISSN 0278-0062.

[360] BT Thomas Yeo, Mert R Sabuncu, Rahul Desikan, Bruce Fischl, and Polina Golland.

Effects of registration regularization and atlas sharpness on segmentation accuracy.

Medical image analysis, 12(5):603–615, 2008. ISSN 1361-8415.

[361] Arno Klein, Brett Mensh, Satrajit Ghosh, Jason Tourville, and Joy Hirsch. Mind-

boggle: automated brain labeling with multiple atlases. BMC medical imaging, 5

(1):7, 2005. ISSN 1471-2342.

[362] Jinzhong Yang, Yongbin Zhang, Lifei Zhang, and Lei Dong. Automatic segmentation

of parotids from ct scans using multiple atlases. Medical image analysis for the clinic:

a grand challenge, pages 323–330, 2010.

[363] Thomas R Langerak, Floris F Berendsen, Uulke A Van der Heide, Alexis NTJ Kotte,

and Josien PW Pluim. Multiatlas‐based segmentation with preregistration atlas

selection. Medical physics, 40(9):091701, 2013. ISSN 0094-2405.

[364] Saman Nouranian, S Sara Mahdavi, Ingrid Spadinger, William J Morris, Septimu E

Salcudean, and Purang Abolmaesumi. A multi-atlas-based segmentation framework

for prostate brachytherapy. IEEE transactions on medical imaging, 34(4):950–961,

2014. ISSN 0278-0062.

[365] Oscar Acosta, Antoine Simon, Frédéric Monge, Frédéric Commandeur, Cunka

Bassirou, Guillaume Cazoulat, Renaud De Crevoisier, and Pascal Haigron. Evalua-

tion of multi-atlas-based segmentation of ct scans in prostate cancer radiotherapy.

In Biomedical Imaging: From Nano to Macro, 2011 IEEE International Symposium

on, pages 1966–1969. IEEE, 2011. ISBN 1424441277.

References 284

[366] Bas Schipaanboord, Djamal Boukerroui, Devis Peressutti, Johan Van Soest, Tim

Lustberg, Timor Kadir, Andre Dekker, Wouter van Elmpt, and Mark Gooding. Can

atlas-based auto-segmentation ever be perfect? insights from extreme value theory.

IEEE transactions on medical imaging, 38(1):99–106, 2018. ISSN 0278-0062.

[367] Jimit Doshi, Guray Erus, Yangming Ou, Susan M Resnick, Ruben C Gur,

Raquel E Gur, Theodore D Satterthwaite, Susan Furth, Christos Davatzikos, and

Alzheimer’s Neuroimaging Initiative. Muse: Multi-atlas region segmentation utiliz-

ing ensembles of registration algorithms and parameters, and locally optimal atlas

selection. Neuroimage, 127:186–195, 2016. ISSN 1053-8119.

[368] Daniel J Blezek and James V Miller. Atlas stratification. Medical image analysis,

11(5):443–457, 2007. ISSN 1361-8415.

[369] John A Bogovic, Bruno Jedynak, Rachel Rigg, Annie Du, Bennett A Landman,

Jerry L Prince, and Sarah H Ying. Approaching expert results using a hierarchical

cerebellum parcellation protocol for multiple inexpert human raters. NeuroImage,

64:616–629, 2013. ISSN 1053-8119.

[370] Hongzhi Wang and Paul A Yushkevich. Guiding automatic segmentation with multi-

ple manual segmentations. In International Conference on Medical Image Computing

and Computer-Assisted Intervention, pages 429–436. Springer, 2012.

[371] William R Crum, Thomas Hartkens, and DLG Hill. Non-rigid image registration:

theory and practice. The British Journal of Radiology, 2004.

[372] Aristeidis Sotiras, Christos Davatzikos, and Nikos Paragios. Deformable medical

image registration: A survey. IEEE transactions on medical imaging, 32(7):1153,

2013.

[373] BT Thomas Yeo, Mert R Sabuncu, Tom Vercauteren, Daphne J Holt, Katrin

Amunts, Karl Zilles, Polina Golland, and Bruce Fischl. Learning task-optimal reg-

istration cost functions for localizing cytoarchitecture and function in the cerebral

cortex. IEEE transactions on medical imaging, 29(7):1424–1441, 2010. ISSN 0278-

0062.

References 285

[374] Daniel Rueckert, Luke I Sonoda, Carmel Hayes, Derek LG Hill, Martin O Leach, and

David J Hawkes. Nonrigid registration using free-form deformations: application to

breast mr images. Medical Imaging, IEEE Transactions on, 18(8):712–721, 1999.

ISSN 0278-0062.

[375] Gustavo K Rohde, Akram Aldroubi, and Benoit M Dawant. The adaptive bases

algorithm for intensity-based nonrigid image registration. IEEE transactions on

medical imaging, 22(11):1470–1479, 2003. ISSN 0278-0062.

[376] Tom Vercauteren, Xavier Pennec, Aymeric Perchant, and Nicholas Ayache. Non-

parametric diffeomorphic image registration with the demons algorithm. Medical

Image Computing and Computer-Assisted Intervention–MICCAI 2007, pages 319–

326, 2007.

[377] Josien PW Pluim, JB Antoine Maintz, and Max A Viergever. Interpolation arte-

facts in mutual information-based image registration. Computer vision and image

understanding, 77(2):211–232, 2000. ISSN 1077-3142.

[378] Xabier Artaechevarria, Arrate Munoz-Barrutia, and Carlos Ortiz-de Solórzano.

Combination strategies in multi-atlas image segmentation: application to brain mr

data. IEEE transactions on medical imaging, 28(8):1266–1277, 2009. ISSN 0278-

0062.

[379] Michaël Sdika. Combining atlas based segmentation and intensity classification with

nearest neighbor transform and accuracy weighted vote. Medical Image Analysis, 14

(2):219–226, 2010. ISSN 1361-8415.

[380] Xabier Artaechevarria, Arrate Muñoz-Barrutia, and Carlos Ortiz-de Solórzano. Ef-

ficient classifier generation and weighted voting for atlas-based segmentation: Two

small steps faster and closer to the combination oracle. In Medical Imaging 2008:

Image Processing, volume 6914, page 69141W. International Society for Optics and

Photonics, 2008.

References 286

[381] Ivana Isgum, Marius Staring, Annemarieke Rutten, Mathias Prokop, Max A

Viergever, and Bram Van Ginneken. Multi-atlas-based segmentation with local de-

cision fusion—application to cardiac and aortic segmentation in ct scans. IEEE

transactions on medical imaging, 28(7):1000–1010, 2009. ISSN 0278-0062.

[382] Alireza Akhondi-Asl and Simon K Warfield. Simultaneous truth and performance

level estimation through fusion of probabilistic segmentations. Medical Imaging,

IEEE Transactions on, 32(10):1840–1852, 2013. ISSN 0278-0062.

[383] Simon K Warfield, Kelly H Zou, and William M Wells. Simultaneous truth and

performance level estimation (staple): an algorithm for the validation of image seg-

mentation. Medical Imaging, IEEE Transactions on, 23(7):903–921, 2004. ISSN

0278-0062.

[384] Thomas Robin Langerak, Uulke A van der Heide, Alexis NTJ Kotte, Max A

Viergever, Marco Van Vulpen, and Josien PW Pluim. Label fusion in atlas-based

segmentation using a selective and iterative method for performance level estima-

tion (simple). IEEE transactions on medical imaging, 29(12):2000–2008, 2010. ISSN

0278-0062.

[385] I Simmat, P Georg, D Georg, W Birkfellner, G Goldner, and M Stock. Assessment

of accuracy and efficiency of atlas-based autosegmentation for prostate radiotherapy

in a variety of clinical conditions. Strahlentherapie und Onkologie, 188(9):807–815,

2012. ISSN 0179-7158.

[386] Geert Litjens, Robert Toth, Wendy van de Ven, Caroline Hoeks, Sjoerd Kerkstra,

Bram van Ginneken, Graham Vincent, Gwenael Guillard, Neil Birbeck, and Jin-

dang Zhang. Evaluation of prostate segmentation algorithms for mri: the promise12

challenge. Medical image analysis, 18(2):359–373, 2014. ISSN 1361-8415.

[387] Carl Sjöberg, Martin Lundmark, Christoffer Granberg, Silvia Johansson, Anders

Ahnesjö, and Anders Montelius. Clinical evaluation of multi-atlas based segmenta-

tion of lymph node regions in head and neck and prostate cancer patients. Radiation

Oncology, 8(1):229, 2013. ISSN 1748-717X.

References 287

[388] Jean-François Daisne and Andreas Blumhofer. Atlas-based automatic segmentation

of head and neck organs at risk and nodal target volumes: a clinical validation.

Radiation oncology, 8(1):154, 2013. ISSN 1748-717X.

[389] Kirrily Cloak, Michael G Jameson, Andrea Paneghel, Kirsty Wiltshire, Andrew

Kneebone, Maria Pearse, Mark Sidhom, Colin Tang, Carol Fraser‐Browne, and

Lois C Holloway. Contour variation is a primary source of error when delivering

post prostatectomy radiotherapy: Results of the trans‐tasman radiation oncology

group 08.03 radiotherapy adjuvant versus early salvage (raves) benchmarking exer-

cise. Journal of Medical Imaging and Radiation Oncology, 2019. ISSN 1754-9477.

[390] Amy Tien Yee Chang, Li Tee Tan, Simon Duke, and Wai-Tong Ng. Challenges

for quality assurance of target volume delineation in clinical trials. Frontiers in

Oncology, 7:221, 2017. ISSN 2234-943X.

[391] Samantha Cox, Anne Cleves, Enrico Clementel, Elizabeth Miles, John Staffurth,

and Sarah Gwynne. Impact of deviations in target volume delineation–time for a

new rtqa approach? Radiotherapy and Oncology, 137:1–8, 2019. ISSN 0167-8140.

[392] Marie Lambrecht, Enrico Clementel, Jan-Jacob Sonke, Ursula Nestle, Sonja Ade-

bahr, Mathias Guckenberger, Nicolaus Andratschke, Damien C Weber, Marcel Ver-

heij, and Coen W Hurkmans. Radiotherapy quality assurance of sbrt for patients

with centrally located lung tumours within the multicentre phase ii eortc lungtech

trial: Benchmark case results. Radiotherapy and Oncology, 132:63–69, 2019. ISSN

0167-8140.

[393] Barbara Segedin and Primoz Petric. Uncertainties in target volume delineation in

radiotherapy–are they relevant and what can we do about them? Radiology and

oncology, 50(3):254–262, 2016. ISSN 1581-3207.

[394] Dale Roach, Michael G Jameson, Jason A Dowling, Martin Andrew Ebert, Peter B

Greer, Angel M Kennedy, Sandie Watt, and Lois C Holloway. Correlations be-

tween contouring similarity metrics and simulated treatment outcome for prostate

radiotherapy. Physics in medicine and biology, 2018. ISSN 1361-6560.

References 288

[395] MB Altman, JA Kavanaugh, HO Wooten, OL Green, TA DeWees, H Gay,

WL Thorstad, H Li, and S Mutic. A framework for automated contour quality

assurance in radiation therapy including adaptive techniques. Physics in medicine

and biology, 60(13):5199, 2015. ISSN 0031-9155.

[396] M Jameson, J Dowling, J Faustino, K Cloak, M Sidhom, J Martin, J De Leon,

M Berry, D Pryor, and L Holloway. Po-0823: Trac: Automated atlas based ma-

chine learning qa of contouring accuracy for the prometheus trial. Radiotherapy and

Oncology, 127:S429–S430, 2018. ISSN 0167-8140.

[397] Angel Kennedy, Jason Dowling, Peter B Greer, Lois Holloway, Michael G Jameson,

Dale Roach, Soumya Ghose, David Rivest‐Hénault, Marco Marcello, and Martin A

Ebert. Similarity clustering based atlas selection for pelvic ct image segmentation.

Medical physics, 2019. ISSN 0094-2405.

[398] Dale Roach, Lois C Holloway, Michael G Jameson, Jason A Dowling, Angel Kennedy,

Peter B Greer, Michele Krawiec, Robba Rai, Jim Denham, and Jeremiah De Leon.

Multi‐observer contouring of male pelvic anatomy: Highly variable agreement across

conventional and emerging structures of interest. Journal of medical imaging and

radiation oncology, 2019. ISSN 1754-9477.

[399] T Hothorn. maxstat: Maximally selected rank statistics. r package version 0.7-22,

2015.

[400] Ralf Bender and Stefan Lange. Adjusting for multiple testing—when and how?

Journal of clinical epidemiology, 54(4):343–349, 2001. ISSN 0895-4356.

[401] CP South, VS Khoo, O Naismith, A Norman, and DP Dearnaley. A comparison of

treatment planning techniques used in two randomised uk external beam radiother-

apy trials for localised prostate cancer. Clinical oncology, 20(1):15–21, 2008. ISSN

0936-6555.

[402] Carolin Schubert, Oliver Waletzko, Christian Weiss, Dirk Voelzke, Sevda Toperim,

Arnd Roeser, Silvia Puccini, Marc Piroth, Christian Mehrens, and Jan-Dirk Kueter.

Intercenter validation of a knowledge based model for automated planning of vol-

umetric modulated arc therapy for prostate cancer. the experience of the german

References 289

rapidplan consortium. PloS one, 12(5):e0178034, 2017. ISSN 1932-6203. doi:

https://doi.org/10.1371/journal.pone.0178034.

[403] Gerard J van der Wielen, John P Mulhall, and Luca Incrocci. Erectile dysfunction

after radiotherapy for prostate cancer and radiation dose to the penile structures: a

critical review. Radiotherapy and Oncology, 84(2):107–113, 2007. ISSN 0167-8140.

[404] Pirus Ghadjar, Michael J Zelefsky, Daniel E Spratt, Per Munck af Rosenschöld,

Jung Hun Oh, Margie Hunt, Marisa Kollmeier, Laura Happersett, Ellen Yorke, and

Joseph O Deasy. Impact of dose to the bladder trigone on long-term urinary function

after high-dose intensity modulated radiation therapy for localized prostate cancer.


ISSN 0360-3016.

[405] Alan Jay Katz and Josephine Kang. Quality of life and toxicity after sbrt for organ-

confined prostate cancer, a 7-year study. Frontiers in oncology, 4:301, 2014. ISSN

2234-943X.

[406] Sergiu Scobioala, Christopher Kittel, Khaled Elsayad, Kai Kroeger, Michael Oertel,

Laith Samhouri, Uwe Haverkamp, and Hans Theodor Eich. A treatment planning

study comparing imrt techniques and cyber knife for stereotactic body radiotherapy

of low-risk prostate carcinoma. Radiation Oncology, 14(1):143, 2019. ISSN 1748-

717X.

Looking back to move forward: - UNSWorks

Documents

Transcript of Looking back to move forward: - UNSWorks