lable at ScienceDirect

Pancreatology 13 (2013) 570e575

Contents lists avai

Pancreatology

journal homepage: www.elsevier .com/locate/pan

Original article

Computed tomography staging of pancreatic cancer: A validationstudy addressing interobserver agreement

L. Loizou a,b, N. Albiin a,b, C. Ansorge a,c, M. Andersson d, R. Segersvärd a,c, B. Leidner a,b,A. Sundin e,f, L. Lundell a,c, N. Kartalis a,b,*

aDepartment of Clinical Science, Intervention and Technology at Karolinska Institutet, Division of Medical Imaging and Technology,14186 Stockholm, SwedenbDepartment of Radiology, Karolinska University Hospital, Huddinge, 14186 Stockholm, SwedencDepartment of Surgery, Karolinska University Hospital, Huddinge, 14186 Stockholm, SwedendDepartment of Radiology, Sahlgrenska University Hospital, 41345 Göteborg, SwedeneDepartment of Radiology, Karolinska University Hospital, Solna, 17176 Stockholm, SwedenfDepartment of Molecular Medicine and Surgery, Karolinska Institutet, 17176 Stockholm, Sweden

a r t i c l e i n f o

Article history:Received 17 June 2013Received in revised form29 September 2013Accepted 30 September 2013

Keywords:Blood vesselsClassificationDiagnostic imagingPancreatic ductal carcinomaSurgery

* Corresponding author. Division of Medical Imagment of Clinical Science, Intervention and Technologtutet, Karolinska University Hospital, SE-14186 Sto767615006; fax: þ46 87114840.

E-mail address: nikolaos.kartalis@karolinska.se (N

1424-3903/$ e see front matter Copyright � 2013, IAhttp://dx.doi.org/10.1016/j.pan.2013.09.004

a b s t r a c t

Background/objectives: Ductal adenocarcinoma in the head of the pancreas (PDAC) is usually unresect-able at the time of diagnosis due to the involvement of the peripancreatic vessels. Various preoperativeclassification algorithms have been developed to describe the relationship of the tumor to these vessels,but most of them lack a surgically based approach. We present a CT-based classification algorithm forPDAC based on surgical resectability principles with a focus on interobserver variability.Methods: Thirty patients with PDAC undergoing pancreaticoduodenectomy were examined by using astandard CT protocol. Nine radiologists, representing three different levels of expertise, evaluated the CTexaminations and the tumors were classified into four categories (AeD) according to the proposedsystem. For the interobserver agreement, the Intraclass Correlation Coefficient (ICC) was estimated.Results: The overall ICC was 0.94 and the ICCs among the trainees, experienced radiologists, and expertswere 0.85, 0.76, and 0.92, respectively. All tumors classified as category A1 showed no signs of vascularinvasion at surgery. In category A2, 40% of the tumors had corresponding infiltration and requiredresection of the superior mesenteric vein/portal vein (SMV/PV). One of two tumors in category B2 andtwo of three in category C required SMV/PV resection. All six patients in category D had both arterial andvenous involvement.Conclusion: There is almost perfect agreement among radiologists with different levels of expertise inregards to the local staging of PDAC. For tumors in a more advanced preoperative category, an increasedrisk for vascular involvement was noticed at surgery.Copyright � 2013, IAP and EPC. Published by Elsevier India, a division of Reed Elsevier India Pvt. Ltd. Allrights reserved.

1. Introduction

Pancreatic ductal adenocarcinoma (PDAC) is a disease withdismal prognosis and the only therapy that can offer a cure isradical surgical excision. However, only 10e20% of patients withPDAC can be offered surgery with curative intent [1e3]. In addition

ing and Technology, Depart-y (CLINTEC), Karolinska Insti-ckholm, Sweden. Tel.: þ46

. Kartalis).

P and EPC. Published by Elsevier I

to the biological characteristics of the tumor, the anatomical loca-tion of the tumor can have an important bearing on the chances forcurative treatment. Most cancers in the head of the pancreas areunresectable at the time of diagnosis due to metastatic spread and/or involvement of adjacent major vessels. Therefore, preoperativelocal staging algorithms have been developed mainly to describethe relationship of the tumor to the major peripancreatic vessels.These various algorithms have been developed at various in-stitutions and have been evaluated for their clinical usefulness, butthis has often been without the uniform and surgically aggressivetherapeutic approach that is a prerequisite for the proper evalua-tion of these systems [4e6]. Moreover, none of these classificationsystems have been validated concerning interobserver agreement

ndia, a division of Reed Elsevier India Pvt. Ltd. All rights reserved.

Table 1CT protocol.

Preparation Fasting for 4 hOral contrast 1000 mL tap water, 30 min prior to examinationIntravenous contrast Iomeron 400 or Visipaque 320a. Dose 0.75 g Iodine/kg body weight (�10%)b. Injection time 25 sc. Flushing 50 mL NaCl at a rate of 5 mL/s

Scan range and starta. NCP Upper abdomenb. PPP Upper abdomen; 20 s after bolus tracking reached

160 HU (Hounsfield Units) in the aorta at the levelof the first lumbar vertebra

c. PVP Upper abdomen and pelvis; 25 s after the end ofthe PPP scan

Scanning parametersa. (section) Collimation 0.625 mmb. (table) Increment 0.6 mmc. Tube current 120 kVd. PPP Pitch: 0.5; NI (noise index): 22e. PVP Pitch: 1; NI: 36

Reconstructions Axial, coronal, and sagittala. Slice thickness 5 mm (and 3 mm with narrower FOV focused

on the pancreas)b. Interval 2.5 mm (and 1.5 mm with narrower FOV focused

on the pancreas)

FOV, field-of-view; NCP, Non-contrast phase; PPP, Pancreatic parenchymal phase;PVP, Portal venous phase.

L. Loizou et al. / Pancreatology 13 (2013) 570e575 571

or the impact of the level of experience of the readers on the ac-curacy of the assessment.

Accurate preoperative staging is especially important in cases ofPDAC with vascular involvement that requires vessel resection and/or reconstruction. This is supported by a recently published study,showing a similar outcome for patients undergoing visceral veinresection as for those who did not [7]. Moreover, with the advent ofneo-adjuvant therapeutic methods in pancreatic cancer the needfor precise preoperative grading and staging is growing exponen-tially. In addition to this, the use of effective downstaging radio-chemotherapy regimens requires valid techniques not only forstaging but also ultimately for evaluation of therapeutic responses.

The development of a classification system of tumor extent inthe pancreatic head should be based on information that has animpact on the immediate therapeutic management. The extent ofinfiltration of the tumor into and around the vessels in and aroundthe pancreatic bed is one such piece of necessary information. Thedevelopment of modern high-quality pancreatic-protocol MDCT(multidetector computed tomography) and MRI (magnetic reso-nance imaging) techniques make it possible to predict the resect-ability of the tumor based on its relation to the major surroundingvascular structures [3].

The two main parameters traditionally used for evaluating tu-mor vessel involvement are the grade of the circumferentialinvolvement and the length that the tumor is contiguous with thevessel [5,8]. The importance of both parameters probably differsbetween arteries and veins in the pancreatic region. For instance, aless than 180� circumferential involvement of a short vein segmentis easy to deal with for the trained surgeon, whereas correspondinginvolvement of arteries makes it more difficult to achieve an R0resection. It is also possible that periarterial stranding has to betaken into account [8], and venous narrowing or completeobstruction might jeopardize surgical resection [9].

In this work we present a CT classification system for PDAC thatis based on surgical resectability principles and has been developedthrough a close collaboration between dedicated radiologists andpancreatic surgeons. The first step in the comprehensive validationprocess of the system for grading PDAC is also presented. We havefocused on the interobserver variability among nine radiologistsrepresenting three different levels of expertise and relate theseobservations to the final surgical findings.

2. Methods

2.1. Patient population

The study population was recruited retrospectively fromOctober 2006 to December 2010 from a prospectively collectedregister of 277 consecutive patients undergoing attempt forWhipple’s procedure for various pathologies (PDAC, ampullarycarcinoma, distal cholangiocarcinoma, pancreatitis, neuroendo-crine tumors, cystic pancreatic lesions, GISTs, duodenal adenomas/carcinoma, pancreatic metastases, sarcoma), originating in thehead of the pancreas. Of these 277 patients, 30 fulfilled all the in-clusion criteria, which were: i. having a CT examination accordingto the standard preoperative protocol in our department, ii. nopreoperative chemo- or radiochemotherapy, iii. decision at themultidisciplinary conference of having a potentially resectable tu-mor in the pancreatic head, and iv. histopathological confirmationof PDAC diagnosis. The study was approved by the regional ethicalreview board. All operations were performed by experienced andfully trained pancreatic surgeons familiar with the techniques ofmajor vessel resection and/or reconstruction. The mean time be-tween the standard preoperative CT examination and surgery was23 days (the median time was 19 days and the time interval ranged

from 2 to 52 days; in 15 patients the interval was shorter than fourweeks and in 15 longer than four weeks). The study cohort con-sisted of 18 men and 12 womenwith a mean age of 68 years (range42e78 years).

2.2. CT technique

All CT examinations were performed with a 64-channel MDCTscanner (LightSpeed VCT or LightSpeed VCT XTE, GE Healthcare,Milwaukee, WI). Intravenous contrast medium was administeredfollowed by a saline flush. A triple phase protocol, with a non-contrast phase (NCP), a pancreatic parenchymal phase (PPP), anda portal venous phase (PVP), was used. To reduce patient-to-patientvariability, SmartPrep (GE Healthcare, Milwaukee, WI) real-timetracking of contrast in the aorta was used at the level of the firstlumbar vertebra (L1), the scan start threshold was set at 160Hounsfield units (HU), and the PPP was obtained with a 20 s delay.The PVP was obtained after a further 25 s delay. All contrasteenhanced images were reconstructed in 0.6 mm slices that werereformatted in the transversal, coronal, and sagittal planes with5 mm slice thickness and 2.5 mm interval. Another set of imageswith thinner sections (3 mm) and interval (1.5 mm) and a narrowerfield-of-view (focused on the pancreas), were created in all threeplanes. Details of the protocol are included in Table 1.

2.3. Image analysis

The nine radiologists were grouped as trainees, experiencedradiologists, and experts (three in each group) based on the numberof years they had practiced gastrointestinal radiology (1e2 years,5e6 years, and >10 years, respectively). Each reader was asked toretrospectively evaluate each individual preoperative CT examina-tion, and all readers were blinded to the surgical outcomes andpathology reports. At the end of the study, all radiologists hadevaluated all CT examinations. Image evaluation was performedusing a standard picture archiving and communicating systemworkstation (Sectra, Linköping, Sweden). The evaluation time wasunlimited and the participants were free to change the windowsettings and the zoom level at their preference.

Table 3The new classification system based on vascular involvement.

Vein Artery

A1 None NoneA2 SMV/PV < 180� NoneB1 SMV/PV >180� and <2 cm NoneB2 Any of the above (A1, A2, or B1) and HA <180� and <2 cmC SMV/PV >180� and >2 cm and HA <180� and <2 cm

SMV/PV >180� and >2 cm and/or CA and/or SMA <180�

and <2 cmD Total occlusion of SMV/PV or

encasement of mesentericbranching

and/or HA or CA or SMA >180�

and/or >2 cm orencasement of mesentericbranching

SMV, superior mesenteric vein; PV, portal vein; HA, hepatic artery; CA, celiac artery;SMA, superior mesenteric artery.

L. Loizou et al. / Pancreatology 13 (2013) 570e575572

Each examination was evaluated according to a predefinedprotocol. The readers were asked to evaluate the tumor in relationto the major peripancreatic arteries [celiac artery (CA), hepatic ar-tery (HA), and superior mesenteric artery (SMA)] and veins [su-perior mesenteric vein/portal vein (SMV/PV)]. They were asked tograde tumor involvement of each vessel addressing 1) the maximalpercentage of the circumference that was encased by the tumor and2) the maximal length (in cm) that was contiguous to the tumor(Table 2). In order to limit measurement error, the readers wereadvised to use the thinner slices (3 mm).

2.4. Classification

Based on the circumferential and/or longitudinal involvement ofthe major peripancreatic vessels as judged by the readers, each in-dividual tumor was classified by one of the authors (Loizou, L.) intoone of four categories (AeD). Categories A and B were further sub-categorized into A1 and A2 and B1 and B2, respectively (Table 3).The tumors in category A1 did not attach to any major vessel and itwas considered unlikely that any major vessel resection would berequired. This was in contrast to the tumors in category A2 thatattached to a limited segment of SMV/PV and potentially requiredvein resection. Tumors in categories B1 and B2 had vessel infiltra-tion most probably involving the SMV/PV and/or HA and potentiallymandating an SMV/PV and/or a limited HA resection. Category Ctumors were those considered to be borderline resectable and forwhich the multidisciplinary therapy conference would most likelyrecommend neo-adjuvant therapy with the intent of downstagingand increasing the likelihood of achieving an R0 resection. Finally,category D consisted of locally advanced and unresectable tumorsfor which patients are often offered only palliative therapy.

2.5. Correlation to surgery, R0 rate and median survival

The CT classification of the tumor was correlated to the resect-ability rate and to the surgical procedures that were performed interms of vein and/or arterial resection according to the surgicalreports. For this final validation, a joint category score was estab-lished that incorporated the results of all nine readers. When six ormore readers classified the tumors in the same category, that jointimage score was used for the subsequent comparison with thesurgical outcome. In the remaining patients, reevaluation of the CTexaminations was performed in consensus between three of thenine readers (one from each group). Finally, the categories of the CTclassification were grouped e due to the small sample size e andcorrelated to the R0 rate (%) andmedian survival (group 1, primarilyresectable tumors with no contact to major peripancreatic vessels,i.e. category A1; group 2, primarily resectable tumors with limitedcontact to major peripancreatic vessels, i.e. categories A2, B1 andB2; and group 3, primarily non-resectable tumors. i.e. categories Cand D).

Table 2Grading of vascular involvement.

1. Vessel circumference (�)1 02 1e893 90e1794 180e2695 �2702. Vessel length (cm)1 02 0.1e0.93 1.0e1.94 2.0e2.95 �3

2.6. Statistical analysis

The interobserver agreement was analyzed according to themethod presented and described by Shrout & Fleiss [10] and Bland& Altman [11], which yields inter- and intraclass correlations (ICC).The ICC can be interpreted as follows: a score of 0e0.2 indicatespoor agreement, 0.3e0.4 indicates fair agreement, 0.5e0.6 in-dicates moderate agreement, 0.7e0.8 indicates strong agreement,and more than 0.8 indicates a very strong and almost perfectagreement.

3. Results

At the time of surgical exploration, there was no sign of vesselinfiltration in 16 of the 30 patients (Fig. 1(a),(c),(e) and Fig. 2(a),(b)).Ten patients underwent SMV/PV resection due to tumor involve-ment (Figs. 1(b),(d) and 2(c),(f)). Involvement of the HA wasobserved in one patient who had a resection and subsequentreconstruction. One patient had a concomitant HA and PV resec-tion, but the vessel involvement at the preoperative decision con-ference was judged to be due to local inflammation and not tumorinfiltration. Two patients had locally advanced disease, i.e. anunresectable tumor (Fig. 2(b) and (e)).

3.1. Interobserver agreement

The overall ICC among the nine readers was 0.94. The corre-sponding ICCs among the trainees, the experienced radiologists,and the experts were 0.846, 0.759, and 0.919, respectively. Table 4shows the distribution of the interobserver agreement for eachpatient divided by the three groups. It should be noted that no B1case was included in the final analysis (Tables 4 and 5).

3.2. Correlation to surgical findings, R0 rate and median survival

In 22 of the 30 patients, six or more readers classified the tu-mors in the same category (Figs. 1 and 2(b)e(f)). In the remainingeight patients, a consensus was scored and used for the finalevaluation (Fig. 2(a) and (d)). The tumor distribution in thedifferent categories after consensus is shown in Table 5. In all ninetumors classified as category A1, the preoperative findings corre-lated perfectly with the surgical assessments that showed no signof vascular invasion (Fig. 1(a)). However, in four out of the ten(40%) tumors classified as category A2, a corresponding infiltrationof the SMV/PV was observed and a resection of the respectivevessel was found to be necessary (Fig. 1(b) and (d)). In theremaining six tumors (60%) classified as category A2, no infiltra-tion of the SMV/PV was observed at surgery (Fig. 1(c) and (e)). In

Fig. 1. CT images at the pancreatic parenchymal phase of three different patients, patient 15 (a), patient 17 (b,d), and patient 6 (c,e) (see also Table 5). The white arrow indicates thetumor, the open arrow indicates the major vein, and the arrowhead indicates the major artery. In (a), all nine readers classified the tumor in category A1, which correlated well withthe surgical findings because the tumor was excised without any vessel resection. In (b,d) and (c,e), all nine readers classified the tumors in category A2. In the former, the tumorwas excised with a simultaneous wedge vein resection, and in the latter the tumor was excised without any vessel resection. The overestimation of vein engagement during imagingwas probably due to inflammatory changes that were secondary to the commonly seen obstructive pancreatitis that was also seen around the tail area [black arrows in (c)]. Images(a,d,e) are on the coronal and (b,c) on the transversal plane.

L. Loizou et al. / Pancreatology 13 (2013) 570e575 573

category B2, one of the two tumors (50%) required SMV/PVresection. Two out of three tumors (67%) classified in category Chad corresponding vein resections. All six patients with tumorsclassified as category D had signs of extensive vascular involve-ment during the following surgical exploration. Four of them (67%)underwent vein resection (Fig. 2(c) and (f)), two of which alsorequired an en-bloc arterial resection. Two patients (33%) had alocally advanced tumor judged to be unresectable (Fig. 2(b) and(e)). In all patients with an interval between the CT-examinationand surgery longer than 4 weeks, the surgical findings did notreveal a tumor stage higher than expected by using the CT-classification system (data not shown).

Finally, the R0 rate was 33% (group 1), 25% (group 2), and 0%(group 3) and the median survival was 20 months (group 1), 22months (group 2), and 12 months (group 3).

Fig. 2. CT images at the pancreatic parenchymal phase of three different patients, patientindicates the tumor, the open arrow indicates the major vein, and the arrowhead indicateregarding tumor classification that ranged from A2 to D. At the consensus reading, the tumorthe tumor was excised without any vessel resection. In (b,e), all nine readers classified the ttumor was unresectable. In (c,f), all nine readers classified the tumor in category D, butsimultaneous vein resection. Even here, the overestimation of vein engagement at imaging w(a,b,c) are on the transversal and (d,e,f) on the coronal plane.

4. Discussion

CT staging of pancreatic cancer occupies a central position in theclinical management of these patients. Based on the pivotal role ofsurgery in the curative treatment of PDAC, it is of critical impor-tance that the preoperative radiological examinations providerelevant and accurate information in order to reach the optimaltherapeutic decision for each patient. In this study, we developed aCT-based classification system for PDAC originating in the head ofthe pancreas. This system has been developed as a joint venturebetween dedicated radiologists and surgeons. The focus of this, aswell as of other related systems [5,8], has been on the criticalanatomical structures in and around the pancreatic gland becausethe involvement of these structures will have immediate conse-quences for the ensuing therapeutic and surgical strategies.

2 (a,d), patient 13 (b,e), and patient 3 (c,f) (see also Table 5). Again, the white arrows the major artery. In (a,d), there was considerable disagreement among the readerswas classified in category B2, which did not correlate well with the surgical findings as

umor in category D. Findings at imaging correlated well with findings at surgery as thethis did not correlate with the surgical findings as the tumor was resectable withas probably due to inflammatory changes secondary to obstructive pancreatitis. Images

Table 4Scoring of each individual reader for each patient enrolled in the study.

Patient Category

A1 A2 B1 B2 C D

T e E T e E T e E T e E T e E T e E

1 3 2 3 12 1 3 1 1 1 1 13 3 3 34 2 3 3 15 3 2 3 16 3 3 37 1 1 2 3 28 3 3 39 2 3 3 110 1 2 1 1 1 1 211 2 2 3 1 112 1 3 3 1 113 1 2 3 314 1 1 1 2 3 115 3 3 316 1 2 2 1 2 117 3 3 318 1 2 1 1 1 2 119 2 2 3 1 120 2 2 2 1 1 121 1 1 2 1 1 1 222 1 1 1 2 1 323 2 2 1 2 1 124 3 2 2 1 125 1 3 3 226 2 1 3 1 1 127 1 3 2 1 1 128 1 1 1 3 2 129 2 3 2 1 130 1 1 1 2 1 2 1

T, trainees; e, experienced radiologists; E, experts.

L. Loizou et al. / Pancreatology 13 (2013) 570e575574

In this study, we aimed to investigatewhether radiologists coulduse our classification system to accurately, and with high agree-ment, assess the tumor involvement of the major peripancreaticblood vessels. Moreover, we wanted to know whether such anagreement was dependent on the reader’s level of experience. TheICC values of 0.85, 0.76, and 0.92 for trainees, experienced radiol-ogists, and experts, respectively, suggest strong agreement be-tween radiologists independently of their level of expertise. Thesefindings indicate the potential for our classification system to beclinically very useful. Our data are quite unique compared to thosefrom validation studies that have been performed in similar fieldsof diagnostic and therapeutic medicine [12e14]. Although a slightvariation was seen in the ICC values between the three groups ofradiologists with different level of expertise, the variation was notclinically significant. The outcome among the trainees is of partic-ular importance because it suggests that our novel PDAC classifi-cation system is easy to learn and use.

As the classification goes from A2 to C, the degree of vascularinvolvement becomes more and more pronounced and this

Table 5Radiological staging as related to subsequent vascular resection.

Category Total number of patients Surgical report

No vessel resection Vein rese

A1 9 9A2 10 6 4B1 0B2 2 1 1C 3 1 2D 6 2

increases the probability that en-bloc vascular resection will beneeded to achieve local radicality. In this context, we propose asubgrouping of the A and B categories based on the conceptualassumption that these categories lead to important alterations inthe therapeutic strategy planning. As seen in Table 4, we were ableto demonstrate a good concordance between the different readers,but the number of cases in each subgroup was limited and, in fact,category B1 does not contain a single patient. Therefore, more in-formation needs to be obtained from a larger patient population todetermine the relevance and accuracy of this subgrouping. In aparallel manner, i.e. going from group 1 to 3, the R0 rate as well asthe median survival decreases. Due to the relatively small samplesize in this study, no statistical comparison between the differentcategories or groups could be performed, and thus, the resultsregarding these two parameters are indicative and not conclusive.Further studies on larger cohorts are needed to evaluate the impactof the proposed CT classification on both R0 rate and mediansurvival

A comprehensive and reliable CT-based staging system willbecome even more important as neo-adjuvant regimens in thetreatment of PDAC are explored and implemented, either todecrease the need for vascular resections or to increase the tumor-free surgical margins (i.e. the R0 rate) and in that way offer a morepersonalized therapeutic approach. In all clinical trials designedand developed to test such regimens, there will be a need for aunified and generally accepted radiologic method to describe,characterize, and stage each individual PDAC patient enrolled inthese studies. The outcome of this study suggests that the clinicalresearch community may now have a tool that will allow forstandardization between various studies in the important field ofpreoperative local staging and permit their in-between safe andreliable comparison regarding outcome data.

One potentially problematic area within the current classifica-tion system is confined to the C and, to a greater extent, D cate-gories. Today and even more so in the near future, an increasingnumber of these patients will be offered preoperative downstagingtreatment. Evaluation of the response to induction chemo- orradiochemotherapies will significantly increase the radiologicalcomplexity when it comes to assessment. These and related issueshave to be addressed within the framework of well-designedclinical trials. It is also necessary to explore whether complemen-tary CT- or MRI-based technologies (e.g. perfusion imaging) are thediagnostic pathways that should be pursued in this system.

The present classification system was based entirely on therelationship between the tumor mass and the adjacent majorvessels. This does not mean that the tumor’s relation to theremaining three-dimensional anatomic structures in the area areprognostically and/or therapeutically irrelevant. Future studies,especially when combined with dedicated, highly specialized pa-thology reporting [15], have to address these issues and theirclinical significance. Based on the results of such studies, furtherrefinement and adjustments of the present PDAC classificationsystemmight be required. Furthermore, this classification may, at a

ction Artery resection Artery þ vein resection Unresectable

1 1 2

L. Loizou et al. / Pancreatology 13 (2013) 570e575 575

first glance, look somehow complex, consisting of different inter-connecting elements. However, the ambition is to reflect reality andbasically cover the clinically relevant combinations of venous and/or arterial involvements. Currently each vessel is evaluated sepa-rately and when this assessment is completed, each patient is fittedin the respective category (from A1 to D).

In our attempt to have the most homogenous study materialpossible with high quality examinations, only 30 patients wereeligible for inclusion. We were also limited in our attempts to findpatients with more advanced tumor stages because neo-adjuvanttherapy with chemo- or radiochemotheraphy can alter the tumorcharacteristics and, therefore, the results of the radiological imag-ing. Accordingly, only 3 of the 30 patients (10%) were graded incategory C following the consensus image reading. There were alsoonly a few patients classified in category D. The latter limitationmay introduce a selection bias that makes the soundness of ourconclusions regarding the evaluation of patients in categories C andD less robust. This issue, i.e. the evaluation of the proposed CTclassification system in patients belonging to the more advancedcategories needs to be addressed in larger prospective trialsincluding sufficient numbers of such patients. Furthermore, con-cerns could be raised about the CT technique used during the studyperiod between October 2006 and December 2010 and the possi-bility of using a more recent patient group in order to have CT ex-aminations of higher quality bearing in mind this rapidly evolvingtechnology. As stated in Methods Section, all patients included inthe present study were examined with the same CT imaging pro-tocol at the same 64-channel MDCT that was available at ourRadiological Department from the beginning of 2006. This protocolis still considered state-of-the-art and with only minor modifica-tions is currently used at our institution as our standard pancreaticCT protocol [16]. In that way, it is highly unlikely that the inclusionof a more recent group of patients would provide with CT-examinations of higher quality.

In conclusion,wehave foundanalmost perfect agreement amongradiologists with different levels of expertise in the local staging ofPDAC based on a modern high-quality multidetector computed to-mography pancreatic protocol. In the more advanced preoperativecategories, increased levels of vascular involvement were noticedduring surgery. This categorization system now needs to be pro-spectively tested on a large, consecutively collected patient series.

Author declaration

LLo, NA, RS, BL, AS, LLu and NK conceived and designed thestudy; LLo, NA, CA, MA, RS, BL, AS, LLu and NK collected, analyzedand interpreted the data; all authors drafted and critically revisedthe article for important intellectual content.

Conflict of interest

All authors declare they have no conflict of interest.

Acknowledgments

The authors would like to thank LenaMHallberg, Elin Bacsovics-Brolin and Farouk Hashim for their contribution.

References

[1] Zakharova OP, Karmazanovsky GG, Egorov VI. Pancreatic adenocarcinoma:outstanding problems. World J Gastrointest Surg 2012 May 27;4(5):104e13.

[2] Bond-Smith G, Banga N, Hammond TM, Imber CJ. Pancreatic adenocarcinoma.BMJ 2012 May 16;344:e2476.

[3] Vincent A, Herman J, Schulick R, Hruban RH, Goggins M. Pancreatic cancer.Lancet 2011 Aug 13;378(9791):607e20.

[4] O’Malley ME, Boland GW, Wood BJ, Fernandez-del Castillo C, Warshaw AL,Mueller PR. Adenocarcinoma of the head of the pancreas: determination ofsurgical unresectability with thin-section pancreatic-phase helical CT. AJR AmJ Roentgenol 1999 Dec;173(6):1513e8.

[5] Lu DS, Reber HA, Krasny RM, Kadell BM, Sayre J. Local staging of pancreaticcancer: criteria for unresectability of major vessels as revealed by pancreatic-phase, thin-section helical CT. AJR Am J Roentgenol 1997 Jun;168(6):1439e43.

[6] Phoa SS, Reeders JW, Stoker J, Rauws EA, Gouma DJ, Laméris JS. CT criteria forvenous invasion in patients with pancreatic head carcinoma. Br J Radiol 2000Nov;73(875):1159e64.

[7] Turrini O, Ewald J, Barbier L, Mokart D, Blache JL, Delpero JR. Should the portalvein be routinely resected during pancreaticoduodenectomy for adenocarci-noma? Ann Surg 2013 Apr;257(4):726e30.

[8] Varadhachary GR, Tamm EP, Abbruzzese JL, Xiong HQ, Crane CH, Wang H,et al. Borderline resectable pancreatic cancer: definitions, management, androle of preoperative therapy. Ann Surg Oncol 2006 Aug;13(8):1035e46.

[9] Chun YS, Milestone BN, Watson JC, Cohen SJ, Burtness B, Engstrom PF, et al.Defining venous involvement in borderline resectable pancreatic cancer. AnnSurg Oncol 2010 Nov;17(11):2832e8.

[10] Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater reliability.Psychol Bull 1979 Mar;86(2):420e8.

[11] Altman DG, Bland JM. Measurement in medicine: the analysis of methodcomparison studies. J R Stat Soc Series D 1983;32:307e17.

[12] Ronot M, Bahrami S, Calderaro J, Valla DC, Bedossa P, Belghiti J, et al.Hepatocellular adenomas: accuracy of magnetic resonance imaging andliver biopsy in subtype classification. Hepatology 2011 Apr;53(4):1182e91.

[13] Besselink MG, van Santvoort HC, Bollen TL, van Leeuwen MS, Laméris JS, vander Jagt EJ, et al., Dutch Acute Pancreatitis Study Group. Describing computedtomography findings in acute necrotizing pancreatitis with the Atlanta clas-sification: an interobserver agreement study. Pancreas 2006 Nov;33(4):331e5.

[14] Gerke H, Jaffe TA, Mitchell RM, Byrne MF, Stiffler HL, Branch MS, et al.Endoscopic ultrasound and computer tomography are inaccurate methods ofclassifying cystic pancreatic lesions. Dig Liver Dis 2006 Jan;38(1):39e44.

[15] Verbeke CS, Menon KV. Redefining resection margin status in pancreaticcancer. HPB (Oxford) 2009 Jun;11(4):282e9.

[16] Tamm EP, Bhosale PR, Vikram R, de Almeida Marcal LP, Balachandran A. Im-aging of pancreatic ductal adenocarcinoma: state of the art. World J Radiol2013 Mar;5(3):98e105.