PHYSICS CONTRIBUTION

3D DOSE VERIFICATION USING TOMOTHERAPY CT DETECTOR ARRAY

KE SHENG, PH.D.,* RYAN JONES, B.S.,* WENSHAYANG, PH.D.,* SIDDHARTH SARAIYA, B.S.,*BERNARD SCHNEIDER, M.D.,* QUAN CHEN, PH.D.,y GEOFF SOBERING, PH.D.,y GUSTAVO OLIVERA, PH.D.,y

AND PAUL READ, M.D., PH.D.*

*Department of Radiation Oncology, University of Virginia, Charlottesville, VA, and yTomoTherapy, Inc., Madison, WI

Purpose: To evaluate a three-dimensional dose verification method based on the exit dose using the onboard de-tector of tomotherapy.Methods andMaterials: The study included 347 treatment fractions from 24 patients, including 10 prostate, 5 headand neck (HN), and 9 spinal stereotactic body radiation therapy (SBRT) cases. Detector sonograms were retrievedand back-projected to calculate entrance fluence, which was then forward-projected on the CT images to calculatethe verification dose, which was compared with ion chamber and film measurement in the QA plans and with theplanning dose in patient plans.Results: Root mean square (RMS) errors of 2.0%, 2.2%, and 2.0% were observed comparing the dose verification(DV) and the ion chamber measured point dose in the phantom plans for HN, prostate, and spinal SBRT patients,respectively. When cumulative dose in the entire treatment is considered, for HN patients, the error of the meandose to the planning target volume (PTV) varied from 1.47% to 5.62% with a RMS error of 3.55%. For prostatepatients, the error of the mean dose to the prostate target volume varied from –5.11% to 3.29%, with a RMS errorof 2.49%. The RMS error of maximum doses to the bladder and the rectum were 2.34% (–4.17% to 2.61%) and2.64% (–4.54% to 3.94%), respectively. For the nine spinal SBRT patients, the RMS error of the minimum dose tothe PTVwas 2.43% (–5.39% to 2.48%). The RMS error of maximum dose to the spinal cord was 1.05% (–2.86% to0.89%).Conclusions: An excellent agreement was observed between the measurement and the verification dose. In thepatient treatments, the agreement in doses to the majority of PTVs and organs at risk is within 5% for the cumu-lative treatment course doses. The dosimetric error strongly depends on the error in multileaf collimator leafopening time with a sensitivity correlating to the gantry rotation period. ! 2011 Elsevier Inc.

In vivo dosimetry, Exit dosimetry QA, Tomotherapy, Dose verification.

INTRODUCTION

Standard quality assurance (QA)measurements to ensure thatthe delivered dose on a treatment unit is within specified tol-erance when compared with the planned dose for intensity-modulated radiation therapy (IMRT) include pretreatmention chamber and film measurements using a phantom. Forthe vast majority of patients, the delivered dose is not directlymonitored at the time of patient treatment. A methodology tomonitor the dose in real time would increase patient safetythrough verification of daily treatment accuracy and eventu-ally may allow online adaptive radiotherapy. The clinical im-plementation of IMRT with extremely complex multileafblocking sequences and stereotactic body radiation therapy(SBRT) with high dose per fraction delivery with both tech-niques employing sharp dose gradients between tumor and

organs at risk (OARs) require extreme treatment accuracyfor optimal patient outcomes. Complicating treatmentfurther, complex multileaf collimator (MLC) motion ofIMRT may also require synchronized gantry rotation suchas with helical tomotherapy (1) or volumetric modulatedarc therapy (2), potentially increasing the risk of misadminis-tration. A catastrophic dose misadministration may occur ifthere is mechanical malfunction or human error in any aspectof IMRT or SBRT treatment planning or delivery.

Current dose verification methodologies measuring doseat the time of patient treatment are limited to point measure-ments on the patient surface or using expensive implanteddosimeters that are not practical for all patients. Point dosemeasurements are routinely taken at the patient surface toestimate the dose using thermoluminescent dosimeters andvarious types of semiconductor dosimeters, such as diodes

Reprint requests to: Ke Sheng, Ph.D., Department of RadiationOncology, University of Virginia, P.O. Box 800375, Charlottesville,VA 22908. Tel: (434) 243-0030; Fax: (434) 982-3520; E-mail:ks2mc@virginia.eduConflict of interest: Quan Chen, Geoff Sobering, and Gustavo

Olivera are employees of TomoTherapy, Inc. Ke Sheng and Paul

Read received research grants and honorarium from TomoTherapy,Inc.Received Sept 8, 2010, and in revised form Dec 22, 2010.

Accepted for publication Dec 29, 2010.

1

Int. J. Radiation Oncology Biol. Phys., Vol. -, No. -, pp. 1–8, 2011Copyright ! 2011 Elsevier Inc.

Printed in the USA. All rights reserved0360-3016/$ - see front matter

doi:10.1016/j.ijrobp.2010.12.043

and metal-oxide-semiconductor field-effect transistor; how-ever, these measure dose in the superficial buildup regionwhere dose calculation is less accurate and where theremay be steep dose falloff, reducing their ability to verifydose delivery. Surface dose alone does not reflect dose todeep-seated tumors, and therefore these approaches do notgenerally measure the target dose accuracy. Moreover, pointdose measurements can be difficult to interpret and insuffi-cient to assess the entire three-dimensional dose distribution,particularly in the high dose gradient regions. The limitationin location of measurement was overcome to a certain degreeby implantable dosimeters (3, 4) that may be implanted intothe target volume and transmit dose readings wirelessly toa receiver. Implantation of the dosimeter is an invasiveprocedure that cannot be used in many patients, such asthose with intracranial target volumes, and adds severalthousand dollars to the cost of treatment. The method doesnot provide a three-dimensional dose distribution of thetreated site. Because there is no method to directly measurethe three-dimensional dose in the patient, alternative ap-proaches to reconstruct the delivered three-dimensionaldose distribution based on measurement of either entranceor exit dose and back-projecting the measurements onto sim-ulation or image guidance CT image sets are desired.

The opportunity to reconstruct dose from informationcollected during treatment became available with the intro-duction of radiation imaging detectors being incorporatedinto treatment delivery systems, such as electronic portal im-aging device (EPID) and CT detector arrays. EPID, whendeployed during treatment, collected exit fluence from thepatient, which was back-projected to X-ray fluence beforeentering the patient; the dose on the patient or phantomwas then recomputed using the entrance fluence and planningCT images. A challenge to implement EPID-based dose re-construction is from the scatter photons from the patient,with cross-plane scatter photons contributing substantiallyto the fluence measured by EPID with cone-beam geometry.Reported solutions to overcome this scatter problem, includeiterative reconstruction (5) and deconvolution of MonteCarlo simulated kernels (6, 7) that were used to removethe noise from scatter photons. Alternatively, with theassumption that the dose delivery is perfectly reproducible,a pretreatment portal imaging sequence is acquired,without the patient, to obtain the entrance fluence (8–11).To take the possible changes in patient geometry intoconsideration, volumetric images collected before eachfraction of the treatment are used for back-projection anddose recalculation (12). Additional challenges exist in thatthe CT numbers derived from online cone-beam CTs are in-accurate up to 15% uncorrected and 5% corrected, alsobecause of the presence of scatter photons (13, 14). Acorrection has to be made before these CBCT images canbe used for dose calculation. Despite these challenges, invivo dosimetry based on the MV EPID has refinedsubstantially in accuracy and reliability for clinical use (8).

The reconstruction of three-dimensional doses on HelicalTomotherapy (HT; TomoTherapy, Madison, WI) using exit

dose measurement via the in-line CT detector array isfeasible. Dose verification on HTwas first studied by Kapa-toes et al. (15–17) and, in principle, employs the similarmethods as EPID-based approaches adapted for HT. Theentrance fluence was derived from the exit dose usinga transfer matrix, which was calculated using the radiologi-cal path length from the source to the detector. In the originalstudy, 3% and 3-mm distance to agreement were observedfor phantom. However, the actual clinical application ofthe method was not demonstrated. Here, we report the initialresults of the dose verification (DV) tool used retrospectivelyto study the characteristics of in vivo dose delivered toa diverse group of patients treated on a HI-ART HT unit.

METHODS AND MATERIALS

Treatment data from 24 patients, including 10 prostate, 5 headand neck (HN), and 9 spinal SBRT cases, treated over a 10-monthperiod were included in this study, resulting in 402 deliveredtreatment fractions, out of the which 55 procedures with thetreatment interruption or data corruption were excluded, leaving347 fractions for analysis. Prostate patients were treated with an ac-celerated regimen of 65 Gy delivered in 25 fractions. OARs ofprostate patients include bladder, rectum, and femoral head. HNpatients were treated by 50 Gy in 25 fractions followed by addi-tional boost of 20 Gy that was not included in the study. OARs ofHN patients include eyes, lenses, optical nerves, optical chiasm,parotids, brainstem, and spinal cord. Spinal SBRT patients weretreated to a total of 24 Gy in 3 fractions. Spinal cord was consideredas the OAR in this study. Depending on the gantry speed, which isallowed between 1 to 6 rpm, each fraction in the SBRT treatmentwas further divided in 2–4 passes. All patients were treated witha jaw size of 2.5 cm. Twenty-one patients were treated with a pitchof 0.287 to minimize the thread effect (18), and the remaining threepatients were treated with pitches of 0.2, 0.25, and 0.25, respec-tively. Gantry rotation period and modulation factor for thesepatients are summarized in Table 1. For patients with interruptedprocedures, the cumulated verification dose was scaled up propor-tionally to match the total number of actually delivered dose.Before treatment, a patient-specific QAwas performed for each

patient. In the QA plans, the patient CTwas replaced by a cylindri-cal phantom with a diameter of 30 cm, and the dose was recalcu-lated. Point dose measurement was performed at the center of thephantom using an electrometer (PTW unidos E, Freiburg,Germany) and an ion chamber (Standard Imaging A1SL, Middle-ton, WI). Kodak EDR-2 film was also placed between the twohalves of the cylindrical phantom to measure the coronal planardose and compared with the calculated two-dimensional dose.The ratio between the ion chamber measurement and the calculatedpoint dose is shown in Table 2. The film measurement was

Table 1. Gantry rotation period and modulation factor ofpatients in the study

Disease site

Average (range)

Gantry rotation period (s) Modulation factor

Prostate 26.8 (19.0–34.0) 2.0 (1.8–2.4)Head and neck 19.7 (15.0–24.3) 2.1 (1.9–2.3)Spinal SBRT 43.6 (32.0–56.0) 2.2 (1.8–2.5)

Abbreviation: SBRT = stereotactic body radiation therapy.

2 I. J. Radiation Oncology d Biology d Physics Volume -, Number -, 2011

renormalized to match the maximum dose before the gamma anal-ysis using 3%/3-mm criteria. For the gamma analysis, the region ofinterest was selected to be a 24 ! 17 cm rectangular region of filmthat was within the cylindrical phantom.The details of the dose verification calculation were described by

Kapatoes et al. (17). Briefly, for every treatment fraction delivered,including the QA procedures, the detector signal in the form of a si-nogram was retrieved from the HT archives and imported into theDV tool. For the QA plans, the kVCT image of the cylindrical phan-tom was used for back-projection. For the patient plans, the MVCT(Mega-Voltage Computed Tomography) acquired before each treat-ment was used for back-projection with a CT number-density cali-bration curve established on the HT unit. The flow chart of the DVcalculation is shown in Fig. 1.Mathematically, the relationship between entrance and exit

fluence is described by (17):

s ¼ Dj ;

j ¼ D#1s ;

where s is the detector sinogram, j is the entrance fluence, and D isthe transfer matrix (15, 16). To calculate j from s, the transfermatrix D needs to be determined. D was calculated on the basis

of the radiological pathlength and the detector-to-patient distanceusing the kVCTorMVCT images as previously described by Kapa-toes et al. (15, 16, 19). Leakage and transmission and tongue andgroove were measured with closed MLC and individually openedleaves. Both were corrected from the detector sonogram beforeback-projection.The entrance fluence was converted to the unit of equivalent leaf

opening time (LOT), which was compared with the planned LOT tocalculate LOT error. With the entrance fluence, dose can be recal-culated according to either the kVCT or the MVCT. We chose thekVCT images because the emphasis of this study is in the sensitivityof DV to machine variations, instead of patient anatomical changes.For the same reason, dose statistics were calculated based on theoriginal patient contours from the planning CT with the recon-structed dose.For the reconstructed phantom doses, the point dose at the center,

where the point dose measurement is taken, was compared with thecalculated value. Gamma at the film plane was also calculatedbetween the DV dose and the dose predicted by the planning stationafter normalization to the maximum value. For patients, recon-structed doses to the planning target volume (PTV) and OARswere compared with the planning dose.

Fig. 1. Flow chart of the dose verification calculation process.kVCT stands for kilo-voltage computed tomography and MVCTis the abbreviation of the mega-voltage computed tomography.

Fig. 2. Dose volume histogram of a head and neck patient. Thesolid lines are from the planned dose, the dotted lines are fromthe dose verification reconstructed doses from 5 fractions with 1-week intervals between fractions. PTV = planning target volume.ON stands for optical nerve.

Table 2. Comparison of the ion chamber/film measurementand the dose reconstruction results on the phantom using the

DQA plans

Prostate

Ion chamberpoint dose

Measuredgamma (%)

DV pointdose

DVgamma (%)

Ave = 1.004 Ave = 98.6 Ave = 1.010 Ave = 97.6

Pt. 1 0.998 98.7 1.017 97.4Pt. 2 0.996 99.5 0.974 98.2Pt. 3 1.019 97.8 0.980 97.5Pt. 4 1.005 98.3 1.012 96.8Pt. 5 1.000 98.6 1.017 97.8Pt. 6 1.000 98.9 0.995 98.0Pt. 7 0.991 99.2 1.013 98.3Pt. 8 0.992 97.9 1.000 96.9Pt. 9 1.009 98.5 1.044 97.8Pt. 10 1.000 99.0 1.017 97.4

Head-and-neck Ave = 1.018 Ave = 97.1 Ave = 1.006 Ave = 95.3

Pt. 1 0.996 96.8 1.017 95.4Pt. 2 1.012 97.2 1.011 94.8Pt. 3 0.989 97.1 1.010 95.6Pt. 4 1.010 98.0 0.990 94.6Pt. 5 1.002 96.5 1.004 96.0

Spinal SBRT Ave = 0.999 Ave = 98.0 Ave = 1.002 Ave = 97.0

Pt. 1 0.993 97.8 0.991 96.3Pt. 2 1.001 98.2 0.993 97.4Pt. 3 0.991 98.7 0.954 96.9Pt. 4 0.952 97.8 0.971 97.4Pt. 5 1.044 96.9 1.031 95.2Pt. 6 0.989 97.6 1.016 96.8Pt. 7 1.023 98.5 1.021 98.8Pt. 8 1.001 97.0 1.019 95.7Pt. 9 1.005 99.2 1.023 98.4

Abbreviations: Ave = average; DV = dose verification; Pt =patient; SBRT = stereotactic body radiation therapy.

3D Dose verification on tomotherapy d K. SHENG et al. 3

To study the linear correlation of dosimetric variation and theaverage LOT error, the error in the average PTV dose for each an-alyzed fraction is plotted against the average error of MLC openingtime. Least square fitting was performed to determine the slopebetween them. Because the dosimetric error can possibly correlateto the modulation factor and gantry rotation period, linear correla-tion between them was also studied. The coefficient of the linearcorrelation was calculated using the following equation:

Pðx# xÞðy# yÞffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPðx# xÞ2

Pðy# yÞ2

q ;

where x and y are two vectors in the comparison, xand y are themean values. It was previously reported that HT plans with shorteraverage LOT tend to result in higher delivery errors (20). Therefore,the average LOT, in addition to the average LOT error, was alsostudied as an indicator of potential dosimetric errors.

RESULTS

Table 2 shows the ratio between both the ion chambermeasurement and the DV reconstructed point dose to thecalculated dose for the phantom plans. Root mean square(RMS) errors of 2.0%, 2.2 % and 2.0% were observed com-paring the DV and ion chamber measured point dose in thephantom plans for HN, prostate, and spinal SBRT patients,respectively. Gamma analysis shows good agreement witha higher pass rate from the prostate patients comparedwith SBRT and HN patients in both the film and DV-basedanalyses. Larger gamma errors typically present at the pen-umbra regions in both cases. Quantitatively, the linear corre-

lation between the gamma calculated from the film and DVdose is 0.78. When cumulative dose in the entire treatment isconsidered, for the five HN patients, the error of the meandose to the PTV varied from 1.47% to 5.62% with anRMS error of 3.55%; dose volume histogram (DVH) ofa typical HN patient is shown in Fig. 2. For the 10 prostatepatients, the error of the mean dose to the prostate targetvolume varied from –5.11% to 3.29%, with a RMS errorof 2.49%. The RMS error of maximum dose to the bladderwas 2.34% with a range of –4.17% to 2.61%. The RMS errorof maximum dose to the rectum was 2.64%, with a range of–4.54% to 3.94%. For the nine spinal SBRT patients, the er-ror of the minimum dose to the PTV varied from –5.39% to2.48%, with a RMS error of 2.43%. The RMS error of max-imum dose to the spinal cord was 1.05% with a range of–2.86% to 0.89%. The dosimetric deviation in the minimal,maximal, and average doses is summarized in Tables 3–5.

The error in average PTV dose vs. the relative MLCfluence error is plotted in Fig. 3a–3e. A strong linear rela-tionship was observed. For the prostate patients, two groupsof patients are clearly shown in the plot (Fig. 3b) withdistinctly different slopes. An improved linear fitting wasachieved with R2 = 0.97 once the 10 prostate patientswere separated in two groups, one group with gantry rotationperiod less than 28 seconds and the other greater than 28seconds (Fig. 3c and 3d, respectively).

In Fig. 4, the scatter points of a specific patient tend to ag-gregate, indicating the correlation between the dosimetric er-ror and certain intrinsic planning parameters of that patient.

Table 3. Root mean square errors of the dose verification doses for planning target volume and organs at risk of head-and-neck cancerpatients compared with the planning dose

Structure

Average RMS % error (range)

Mean dose Minimum dose Maximum dose

PTV 3.55 (1.47 to 5.62) 2.70 (0.16 to 3.57) 5.92 (0.71 to 9.70)Brainstem 0.71 (#0.18 to 1.11) 0.29 (#0.39 to 0.49) 2.04 (#0.13 to 3.30)Spinal cord 0.58 (#0.67 to 0.83) 0.04 (#0.02 to 0.07) 2.27 (#1.76 to 3.91)Optic chiasm 3.16 (0.89 to 5.43) 2.17 (0.68 to 4.18) 4.34 (1.69 to 5.88)Right optic nerve 3.63 (1.48 to 5.58) 3.41 (#0.06 to 6.34) 3.94 (2.42 to 5.56)Left optic nerve 3.66 (2.22 to 5.12) 3.30 (2.16 to 4.58) 3.96 (2.96 to 4.64)Right parotid 1.13 (#0.49 to 2.14) 1.06 (#1.70 to 0.84) 3.89 (2.13 to 5.18)Left parotid 0.96 (0.59 to 1.50) 0.84 (#1.62 to 0.08) 2.82 (0.73 to 4.75)Right lens 2.13 (#0.09 to 3.62) 0.88 (#0.40 to 1.63) 3.61 (0.14–6.18)Left lens 0.99 (0.54–1.45) 0.49 (#0.50 to 0.85) 2.63 (1.21 to 4.43)Right eye 1.96 (0.11 to 3.05) 0.37 (#0.40 to 0.64) 2.71 (#0.52 to 4.21)Left eye 1.82 (0.75 to 2.45) 0.50 (#0.51 to 0.90) 5.03 (3.22 to 7.04)

Abbreviations: PTV = planning target volume; RMS = root mean square.

Table 4. Root mean square errors of the dose verification doses for PTV and organs at risk of prostate patients compared with theplanning dose

Structure

Average RMS % error (range)

Mean dose Minimum dose Maximum dose

PTV 2.49 (#5.11 to 3.29) 2.73 (#5.08 to 4.42) 2.65 (#4.19 to 5.33)Bladder 0.96 (#1.53 to 1.55) 0.35 (#0.22-1.05) 2.34 (#4.17 to 2.61)Rectum 1.34 (#2.32 to 1.89) 0.15 (#0.31 to 0.02) 2.64 (#4.54 to 3.94)

Abbreviations: PTV = planning target volume; RMS = root mean square.

4 I. J. Radiation Oncology d Biology d Physics Volume -, Number -, 2011

For all patients, the linear correlation coefficient between do-simetric error, the modulation factor and the gantry rotationperiod is shown in Table 6. Dosimetric error is correlated tothe modulation factor with correlation coefficient of –0.62to –0.73 for non-SBRT treatments and 0.04 for SBRT treat-ments. Most interestingly, a strong linear correlation (R2 =0.97) between the slope of the linear fitting in Fig. 3 andthe gantry rotation period was observed (Fig. 3). The correla-tion shows that the sensitivity of dosimetric error to the LOTis inversely proportional to the gantry rotation period.

Figure 5 shows the correlation between the mean doseerror to the PTV and the average LOT. A weak correlationwas observed for plans with longer average LOT.

DISCUSSION

ADV tool based on the linear transfer matrix of the radio-logical pathlength was evaluated on 24 patients treated onHT. The patient cohort covers a wide range of dosimetricand treatment characteristics. Compared with the prostatepatients, a higher modulation factor is usually necessaryon HN patients because of the complexity of the PTV shapevolume and proximity to multiple OARs. Spinal SBRT treat-ments are hypofractionated to deliver higher dose in a singlefraction, which usually results in a longer gantry rotationperiod. HN and spinal SBRT patients are better immobilizedeither by mask and alpha cradle or by a double vacuum im-mobilization cushion. Loss of body weight was commonlyobserved in HN patients but not with the SBRT patientswho were treated in a week or prostate cancer patients.From the study, a slightly larger gamma error was observedin QA plans of HN patients, but for the patient plans, theagreement between reconstructed dose and the planneddose is not affected by the type of cancer patient studiedwith varying treatment-dependent parameters indicatingthe robustness of the DV method.

The analysis emphasized the correlation between MLCleaf opening time error and the dosimetric error, but severalother factors could also contribute to the dosimetric error.The machine output varies from day to day and within thesame day. A daily output checker (DailyQA Checker, Stan-dard Imaging) was used to measure the output variationduring the time these patients were treated. The variationwas between –1.4% and 1.5% during the period of thepatient studied. Intraday variation was not independentlychecked. The variation in output results in error of estimat-ing the LOT. Additional errors may arise from the MVCT

density calibration curves, which were assumed constantbut can change over the timeframe of 10 months, duringwhich the analyzed patients were treated. Frequent calibra-tion of the MVCT was not carried out for this retrospectivestudy. The detector response to each individual leaf openingwas measured one time to set the baseline of the DVsoftware, but its constancy was not monitored. The Xenondetector is energy dependent, further contributing to the un-certainties in dose calculation. Despite these uncertainties,good agreement between the ion chamber measurementand the DV reconstructed dose was observed on the phantomplans with a RMS error around 2%. For 22 of the 24 patients,ion chamber measurement in the DQA (Dosimetric QualityAssurance) plans returned an error less than 3%. Two of theSBRT patients had DQA errors between 4%–5% that wasalso reflected in the reconstructed dose (Patients 4 and 5 ofthe SBRT). The difference is caused by a larger error inthe average LOT as shown in the results. The exact reasonfor the error is not know but correlates to both the long gan-try rotation period and relatively short LOT. The combina-tion may be challenging for the particular tomotherapymachine.

Previous researchers also found the correlation betweenaverage LOTand DQA point dose difference such that lowermean LOT yields larger point dose differences betweenplanned and measured dose (20). Our results from theDQA plans agree with the observation partially that forHN patients with shortest average LOT, the gamma erroris also greater than two other types of patients. In this studyon a heterogenous group of patients with significantly largersamples (347 treatment fractions analyzed, compared with 6in the previous study), the correlation is not significant whenall patients were considered collectively. The difference inobservation indicates that the average LOT is only one ofthe contributing factors to the dosimetric error. The impor-tance of average LOTwas singled out in the previous studywith the same patient being replanned with different pitchesbut similar MLC patterns. The role of average LOT can beovershadowed in the current study by the heterogeneitiesin MLC patterns between patients.

We found that the dose delivery error is also correlated tothe gantry rotation period for all patients and the modulationfactor for non-SBRT treatment. The correlation between themodulation factors and the error in mean PTV doses was notobserved for SBRT patients, possibly because a single SBRTtreatment is intentionally divided into several passes so thatthe gantry rotation period can be less than 60 seconds, a hard

Table 5. Root mean square errors of the dose verification doses for PTV and OARs of spinal SBRT patients compared with theplanning dose

Structure

Average RMS % error (range)

Mean dose Minimum dose Maximum dose

PTV 2.47 (#3.67 to 3.53) 2.43 (#5.39 to 2.48) 3.69 (#7.21 to 5.31)Spinal Cord 0.98 (#1.73 to 1.71) 2.08 (#5.18 to 2.41) 1.05 (#2.86 to 0.89)

Abbreviations: PTV = planning target volume; RMS = root mean square; SBRT = stereotactic body radiation therapy.

3D Dose verification on tomotherapy d K. SHENG et al. 5

Fig. 3. The planning target volume (PTV) mean dose error as a function of the leaf opening time (LOT) error for fourpatient groups. (a). Head and neck patients, (b) all prostate patients, (c) prostate patients with gantry rotation periodsshorter than 27 seconds, (d) prostate patients with gantry rotation period longer than 29 seconds, (e) spinal stereotacticbody radiation therapy patients. MLC = multileaf collimator.

6 I. J. Radiation Oncology d Biology d Physics Volume -, Number -, 2011

limit posed by the machine. In the process, LOT shorter thana threshold is deleted from the plan, resulting in a change inthe actual modulation factor.

For an individual patient, the scatter plot of average MLCleaf opening error and the dosimetric error tends to distributeon a straight line and aggregate together. The slope of theline is strongly determined by the gantry rotation period.The result indicates that for plans with a long gantry rotationperiod, the accuracy of dosimetry is relatively insensitive tothe absolute leaf opening error. This result indicates that longgantry rotation period plans result in longer average LOT,and so the same leaf opening error in millisecond resultsin a smaller percent of the total LOT.

Despite the similar methods for dose reconstruction forEPID-based conventional linacs and CT detector array-based HT systems, there are several distinct differences.HT uses fan beam geometry, which is not susceptible to thenoise from cross-plane scatter photons. The detector arrayfor megavoltage CT reconstruction (21) in HT consists ofion chambers filled by pressurized Xenon gas. Comparedwith solid-state flat panel detectors, ion chambers haveawider dynamic range and quicker response time.Megavolt-age CT also allows a more accurate estimate of the electrondensity than cone beam kV CT.

The CT ion chamber array is more hardy than a flat paneldetector and able to withstand years of high-dose radiation,making it an ideal exit dose measurement instrument. Infact, the exit fluence is collected for essentially every fractionof treatment on HT. In vivo dosimetry on HT does not affectregular treatment, whereas on a conventional linac, the de-ployment of the EPID MV detector may limit the treatmentpositions of certain patients. However, HT detector QA isnot impervious to noise, which can come from the leakagedose through the MLC, crosstalk between detector channels,in-plane scatter photons, and tongue-and-groove effects (15).Improved accuracy is expected with better characterizationof these variables to better model them in the DV tool.

Overall, the DV tool appears to be robust enough forclinical implementation of continuous monitoring of theIMRTdose delivered byHT.Among the 51dosimetric param-eters compared in Tables 3–5, 49 of themwere under 5%, and38 were under 3%. The HN doses were systematicallyoverestimated by 2% for unknown reasons. After correctionfor the systematic error, all dosimetric metrics were lessthan 5% from the expected values. Overall, these values arewell within the current standard of accepted threshold of&5% for Food and Drug Administration–approved andcommercially available thermoluminescent dosimeters,diodes, metal-oxide-semiconductor field-effect transistors,and implantable dosimeters. Another advantage of this doseverification methodology is that this requires no additionalhardware to be purchased, and other than the cost of thesoftware, there should be little additional costs to patient treat-ments. Finally, this methodology does not require additionaltherapist training, and no special tasks, such as diodeplacement and measurement, need to be performed. Thus,there is no impact on clinical efficiency, and the process isdoes not affect the patient.

Controversy may arise as to whether the HT-based DV isan independent check of the IMRT plan. Because the CT

Fig. 4. Inverse of the slopes from Fig. 3 vs. gantry rotation period.

Table 6. Correlation coefficients between the mean PTVdose error and the modulation factor and the gantry rotation

period for all patients

Disease Site

Correlation coefficient

Modulationfactor

Gantryperiod

Head and neck PTV meandose % error

#0.73 #0.49

Prostate(Per. >28 s)

PTV meandose % error

#0.58 #0.79

Prostate(Per. <28 s)

PTV meandose % error

#0.79 #0.61

Spinal SBRT PTV meandose % error

0.04 #0.72

Abbreviations: Per = Gantry period; PTV = planning target vol-ume; SBRT = stereotactic body radiation therapy.

Fig. 5. Average leaf opening time (LOT) vs. root mean square errorin the mean planning target volume doses for all patients. HN =head and neck.

3D Dose verification on tomotherapy d K. SHENG et al. 7

detector is not used in the planning process, exit fluence andMVCT can be regarded as independent. However, the samedose calculation engine was used in forward-projecting thedose onto the patient; hypothetically, a flaw in the algorithmcould affect both the planning and verification doses. There-fore, the program in its current form serves as an indepen-dent check of the hardware but not the dose computationalsoftware. A separate dose calculation based on different al-gorithms should be the next step of investigation.

As mentioned earlier, patient anatomy changes and defor-mation are taken into the consideration of the radiologicalpathlength but not the DV and derivation of the DVH fromoriginal contours. To account for the later, the organs needto be recontoured automatically or manually on the basis

of the MVCT, which is still a topic of ongoing research.This study was dedicated to machine QA alone.

CONCLUSIONS

A DV tool based on the radiological pathlength estimatedfromMVCT images obtained on anHTunit was evaluated on24 patients and a total of 347 treatment fractions.Good agree-ment was observed between the ion chamber measurementand the reconstructed dose in phantomDQA plans. In patienttreatments, the agreement in doses to the PTVand OARs waswithin 5% for the cumulative treatment course doses. The do-simetric error strongly depends on the error in MLC LOTwith a sensitivity correlating to the gantry rotation period.

REFERENCES

1. Mackie TR, Holmes T, Swerdloff S, et al. Tomotherapy: A newconcept for the delivery of dynamic conformal radiotherapy.Med Phys 1993;20:1709–1719.

2. Yu CX. Intensity-modulated arc therapy with dynamic multi-leaf collimation: An alternative to tomotherapy. Phys MedBiol 1995;40:1435–1449.

3. Beyer GP, Scarantino CW, Prestidge BR, et al. Technical eval-uation of radiation dose delivered in prostate cancer patients asmeasured by an implantable MOSFET dosimeter. Int J RadiatOncol Biol Phys 2007;69:925–935.

4. Scarantino CW, Ruslander DM, Rini CJ, et al. An implantableradiation dosimeter for use in external beam radiation therapy.Med Phys 2004;31:2658–2671.

5. McNutt TR, Mackie TR, Paliwal BR. Analysis and conver-gence of the iterative convolution/superposition dose recon-struction technique for multiple treatment beams andtomotherapy. Med Phys 1997;24:1465–1476.

6. Renner WD, Norton K, Holmes T. A method for deconvolutionof integrated electronic portal images to obtain incident fluencefor dose reconstruction. J Appl Clin Med Phys 2005;6:22–39.

7. Yeo IJ, Jung JW,ChewM, et al. Dose reconstruction for intensity-modulated radiation therapy using a non-iterative method andportal dose image. Phys Med Biol 2009;54:5223–5236.

8. Mans A, Remeijer P, Olaciregui-Ruiz I, et al. 3D Dosimetricverification of volumetric-modulated arc therapy by portal do-simetry. Radiother Oncol 2010;94:181–187.

9. van Elmpt W, Ollers M, van Herwijnen H, et al. Volume orposition changes of primary lung tumor during (chemo-)radio-therapy cannot be used as a surrogate for mediastinal lymphnode changes: The case for optimal mediastinal lymph nodeimaging during radiotherapy. Int J Radiat Oncol Biol Phys2010;79:89–95.

10. van Elmpt W, Nijsten S, Mijnheer B, et al. The next step inpatient-specific QA: 3D dose verification of conformal and

intensity-modulated RT based on EPID dosimetry and MonteCarlo dose calculations. Radiother Oncol 2008;86:86–92.

11. van Elmpt WJ, Nijsten SM, Dekker AL, et al. Treatment veri-fication in the presence of inhomogeneities using EPID-basedthree-dimensional dose reconstruction. Med Phys 2007;34:2816–2826.

12. Chen J, Morin O, Aubin M, et al. Dose-guided radiationtherapy with megavoltage cone-beam CT. Br J Radiol 2006;79(Spec No 1):S87–S98.

13. Jarry G, Graham SA, Moseley DJ, et al. Characterization ofscattered radiation in kVCBCT images usingMonte Carlo sim-ulations. Med Phys 2006;33:4320–4329.

14. Zhu L, Xie Y, Wang J, et al. Scatter correction for cone-beamCT in radiation therapy. Med Phys 2009;36:2258–2268.

15. Kapatoes JM, Olivera GH, Ruchala KJ, et al. A feasible methodfor clinical delivery verification and dose reconstruction in to-motherapy. Med Phys 2001;28:528–542.

16. Kapatoes JM, Olivera GH, Balog JP, et al. On the accuracy andeffectiveness of dose reconstruction for tomotherapy. PhysMed Biol 2001;46:943–966.

17. Kapatoes JM, Olivera GH, Reckwerdt PJ, et al. Delivery veri-fication in sequential and helical tomotherapy. Phys Med Biol1999;44:1815–1841.

18. Kissick MW, Fenwick J, James JA, et al. The helical tomother-apy thread effect. Med Phys 2005;32:1414–1423.

19. Kapatoes JM, Olivera GH, Ruchala KJ, et al. On the verifica-tion of the incident energy fluence in tomotherapy IMRT.Phys Med Biol 2001;46:2953–2965.

20. Westerly DC, Soisson E, Chen Q, et al. Treatment planning toimprove delivery accuracy and patient throughput in helical to-motherapy. Int J Radiat Oncol Biol Phys 2009;74:1290–1297.

21. Ruchala KJ, Olivera GH, Kapatoes JM, et al. Megavoltage CTimage reconstruction during tomotherapy treatments. PhysMed Biol 2000;45:3545–3562.

8 I. J. Radiation Oncology d Biology d Physics Volume -, Number -, 2011