This article appeared in a journal published by Elsevier. The attachedcopy is furnished to the author for internal non-commercial researchand education use, including for instruction at the authors institution

and sharing with colleagues.

Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies areencouraged to visit:

http://www.elsevier.com/authorsrights

Author's personal copy

Physics Contribution

Variability of Target and Normal StructureDelineation Using Multimodality Imagingfor Radiation Therapy of Pancreatic CancerEntesar Dalah, PhD,* Ion Moraru, PhD,* Eric Paulson, PhD,*,y

Beth Erickson, MD,* and X. Allen Li, PhD*

Departments of *Radiation Oncology and yRadiology, Medical College of Wisconsin, Milwaukee,Wisconsin

Received Oct 3, 2013, and in revised form Feb 10, 2014. Accepted for publication Feb 24, 2014.

Summary

This work explores the po-tential of using variousmagnetic resonance imaging,positron emission tomogra-phy, and computed tomogra-phy techniques to definetreatment targets and organsat risk for radiation therapyof pancreatic cancer. Signif-icant differences exist in thegross tumor volumes definedby these imaging modalitiesand sequences. Furtherstudies are required toestablish reliable imagingmodalities and/or techniquesfor the accurate target delin-eation, which will be crucialparticularly in dose escala-tion and/or dose painting forpancreatic cancer.

Purpose: To explore the potential of multimodality imaging (dynamic contrasteenhanced magnetic resonance imaging [DCE-MRI], apparent diffusion-coefficientdiffusion-weighted imaging [ADC-DWI], fluorodeoxyglucose positron emission to-mography [FDG-PET], and computed tomography) to define the gross tumor volume(GTV) and organs at risk in radiation therapy planning for pancreatic cancer. Delin-eated volumetric changes of DCE-MRI, ADC-DWI, and FDG-PET were assessed incomparison with the finding on 3-dimensional/4-dimensional CT with and withoutintravenous contrast, and with pathology specimens for resectable and borderlineresectable cases of pancreatic cancer.Methods and Materials: We studied a total of 19 representative patients, whoseDCE-MRI, ADC-DWI, and FDG-PET data were reviewed. Gross tumor volumeand tumor burden/active region inside pancreatic head/neck or body were delineatedon MRI (denoted GTVDCE, and GTVADC), a standardized uptake value (SUV) of 2.5,40%SUVmax, and 50%SUVmax on FDG-PET (GTV2.5, GTV40%, and GTV50%).Volumes of the pancreas, duodenum, stomach, liver, and kidneys were contoured ac-cording to CT (VCT), T1-weighted MRI (VT1), and T2-weighted MRI (VT2) for 7 pa-tients.Results: Significant statistical differences were found between the GTVs from DCE-MRI, ADC-DW, and FDG-PET, with a mean and range of 4.73 (1.00-9.79), 14.52(3.21-25.49), 22.04 (1.00-45.69), 19.10 (4.84-45.59), and 9.80 (0.32-35.21) cm3

for GTVDCE, GTVADC, GTV2.5, GTV40%, and GTV50%, respectively. The meandifference and range in the measurements of maximum dimension of tumor onDCE-MRI, ADC-DW, SUV2.5, 40%SUVmax, and 50%SUVmax compared withpathologic specimens were �0.84 (�2.24 to 0.9), 0.41 (�0.15 to 2.3), 0.58(�1.41 to 3.69), 0.66 (�0.67 to 1.32), and 0.15 (�1.53 to 2.38) cm, respectively.

Reprint requests to: X. Allen Li, PhD, Department of Radiation

Oncology, Medical College of Wisconsin, 8701 Watertown Plank Rd,

Milwaukee, WI 53226. Tel: (414) 805-4362; E-mail: ali@mcw.edu

This work is partially supported by MCW Cancer Center Meinerz

Foundation and by Elekta Inc.

Conflict of interest: none.

Int J Radiation Oncol Biol Phys, Vol. 89, No. 3, pp. 633e640, 20140360-3016/$ - see front matter � 2014 Elsevier Inc. All rights reserved.

http://dx.doi.org/10.1016/j.ijrobp.2014.02.035

Radiation OncologyInternational Journal of

biology physics

www.redjournal.org

Author's personal copy

The T1- and T2-based volumes for pancreas, duodenum, stomach, and liver weregenerally smaller compared with those from CT, except for the kidneys.Conclusions: Differences exists between DCE-, ADC-, and FDG-PETedefinedtarget volumes for RT of pancreatic cancer. Organ at risk volumes based on MRIare generally smaller than those based on CT. Further studies combined with path-ologic specimens are required to identify the optimal imaging modality or sequenceto define GTV. � 2014 Elsevier Inc.

Introduction

The local relapse rates after definitive radiation therapy(RT) for locally advanced pancreatic carcinoma (LAPC)patients are high, ranging from 42% to 68% (1), and may beeven more of a challenge than controlling distant metasta-ses, at least for certain groups of patients (2). This poorlocal control is thought to be dose dependent, with insuf-ficient standard doses of 45-50 Gy imposed by the toleranceof the surrounding organs at risk (OARs). With thepancreas in close proximity to many critical structures,there is an increased need for improved accuracy in grosstumor volume (GTV) delineation to define and escalatedose to the GTV while selectively sparing the adjacentOARs.

Inadequate tumor control from current RT regimes,combined with recent indications for radiation dose esca-lation (3), is providing strong motivation for growing in-terest in dose escalation protocols in patients withunresectable LAPC (4, 5). A recent trial, however, did notshow an improvement in survival over chemotherapy alone,but doses of only 53-54 Gy were used and thus part of thelimitation of the study (6). Advanced RT techniques (eg,respiration motion management [7] and adaptive RT stra-tegies [8, 9]) improve the safe delivery of higher doses,with tighter margins for better normal tissue sparing.

The optimal imaging modality to accurately define theGTV for RT of pancreatic cancer is unknown. Withcomputed tomography (CT), primary pancreatic tumorsoften appear as hypoattenuating lesions with poorly definedborders. Computed tomography scans with specific contrastprotocols, however, offer the best possible visualization toassess tumor resectability according to the vascular anat-omy/tumor interface. Recently evidence has emergedshowing a discrepancy between pancreatic tumor mea-surements obtained from pathologic specimens and CT(10), as well as dynamic contrasteenhanced (DCE) T1-weighted gradient echo magnetic resonance imaging(MRI) (11). In principle, the use of multimodality imagingis intended to synthesize and integrate information pro-vided by individual imaging modalities, aiming to providemore accurate tumor volume delineation that could enabledose escalation and eventually more durable definitive localcontrol. In this work, we address the agreement between theGTVs defined with fluorodeoxyglucose positron emissiontomography (FDG-PET) using 3 different thresholds, DCE,

apparent diffusion-coefficient (ADC) maps obtained fromdiffusion-weighted (DW) MRI, and 3-dimensional (3D)/4-dimensional (4D) CT with and without intravenouscontrast. Additionally, differences between OAR volumesdetermined on MRI and CT are assessed. The divergencebetween tumor-defined maximum dimension (MD) from allimaging modalities is assessed against pathologic speci-mens in resectable and borderline resectable cases.

Methods and Materials

Patient characteristics

Imaging and pathologic data for a total of 19 patients withhistologic confirmation of pancreatic adenocarcinoma wereretrospectively reviewed. Patient staging was classified into3 categories according to whether tumor was surgicallyresectable, borderline resectable, or locally advanced. Ofthese 19 patients, 89.5% of the patients had tumors in thepancreatic head/neck and 10.5% in the pancreatic body. All19 patients had DCE, ADC, and FDG-PET.

Image acquisition

CT imagingAbdominal 3D/4D CTs were collected using a scanner(LightSpeed, GE, Waukesha, Wisconsin, USA) for all pa-tients before the start of RT. Additional CTs with oral and/or intravenous contrast were acquired on a subset of pa-tients at the discretion of the radiation oncologist to assistwith OAR delineation and assessment of vasculatureinvolvement. Images were acquired with 120 kVp, auto-mAs (range, 182-285), 2.5-mm slice thickness, and <1-mmpixel size. The 4D CTs were retrospectively sorted andbinned into 10 respiratory phase images using vendor-provided software (Advantage4D; GE).

PET imagingFree-breathing PET studies were performed using a PET/CT system (Discover Loadstone; GE). Patients wereinstructed to fast for at least 6 hours before acquisition ofthe PET study, which diminished FDG uptake in normalbackground organs due to reduced physiologic glucose useand insulin levels. For imaging of the pancreatic mass,approximately 10-19 mCi (370-703 MBq) of FDG wasinjected intravenously. Images from the base of the skull

Dalah et al. International Journal of Radiation Oncology � Biology � Physics634

Author's personal copy

through the mid-thigh level were acquired after a delayperiod of 45-60 minutes using a whole-body protocol.Nonecontrast-enhanced axial CT was obtained over thesame geometric prescription for attenuation correction ofthe PET images. All PET images were reconstructed usingthe ordered subset expectation maximization algorithm toan average slice thickness of 4.25 mm and pixel spacing of3.91 � 3.91 mm2.

MR imagingMagnetic resonance imaging simulation was performed on a70-cm bore, 3T scanner (Verio, Siemens, Erlangen, Ger-many), with patient setup in treatment position on a customG9 fiberglass flat tabletop insert. The integrated body radi-ofrequency (RF) coil was used for signal transmission, andthe combination of a spine phased-array RF coil and 2flexible phased-array RF coils was used for signal reception.The MR protocol consisted of respiratory-triggeredmorphologic T2 and DWI sequences, as well as breath-held DCE sequences. Respiratory triggering was applied atthe 50% phase of the respiratory cycle, corresponding to endof expiration. To reduce differences caused by the respira-tion, the breath-holds were also performed at the end ofexpiration. The residual difference in the tumor location andshape between the 50% phase and the breath-hold isgenerally <2 mm, and mostly <1 mm for the cases studied.A 2-dimensional single-shot, twice-refocused spin-echo,echo-planar imaging sequence was used for DWI acquisi-tion along with the following parameters: pixel size,1.4 � 1.4 mm2; matrix (MAT), 256 � 160; time to echo(TE), 60 ms; time to repetition (TR), 10,000 ms; Gene-Ralized Autocalibrating partially Parallel acquisition(GRAPPA) (RZ2); number of excitations (NEX), 4; slicethickness (TH), 8 mm; b Z 0, 500, 1000 s/mm2. A 3Dvolumetric interpolated breath-held examination sequencewas used for DCE acquisition along with the following pa-rameters: pixel size, 1.19 � 1.19 mm2; MAT, 320 � 260;TE, 1.4 ms; TR, 3.9 ms; GRAPPA (RZ2); TH, 3 mm. A 4-phase DCE image series was obtained, including precontrast,late arterial-phase, venous-phase, and portal venous-phaseimages. Before the late arterial-phase acquisition, a 0.1-mmol/kg bolus injection of MultiHance (Braccodiagnostics, Monroe Township, NJ) was administered intra-venously at 3 mL/s via power injector after a delay deter-mined using an initial timing bolus. All DCE and DWIimages were corrected for gradient nonlinearity-inducedgeometric distortion using the vendor-provided 3D distor-tion correction algorithm.

GTV delineation

The 40%, 50%, and 60% phase CTs were loaded onto atreatment planning system (XiO; Elekta, Sweden), and theGTV contours for each of the 3 phases were manuallydelineated by an experienced radiation oncologist. An in-ternal tumor volume (ITV) was constructed from the

individual GTV contours drawn on the 3 CT phase images.The volume of the ITV was then used as the GTVCT.

The PET and PET-CT images were transferred offline toa research workstation running OsiriX (version 5.6; Pix-meo, Geneva, Switzerland). Native PET images were con-verted to standardized uptake value (SUV) on a voxelwisebasis using the built-in algorithm in Osirix. Custom soft-ware was created to threshold the SUV image at a value of2.5, 40%SUVmax, and 50%SUVmax (11). The GTV2.5,GTV40%, and GTV50% contours were then manuallygenerated, representing the surviving SUV region confinedwithin the pancreas, determined by fusing the thresholdedSUV image with the PET-CT image.

The native DWI images and 4-phase (precontrast, latearterial, venous, portal venous) DCE images were trans-ferred offline to the Osirix research workstation used forPET GTV delineation. Custom software was used togenerate ADC parameter maps from the native DWI using anoise threshold of 2%. A montage of the b Z 0 s/mm2,b Z 1000 s/mm2, and ADC maps was displayed, andGTVADC was manually constructed from regions of ADChypointensity confined within the pancreas. The 4-phaseDCE images were then displayed as a montage andcompared in terms of signal enhancement within thepancreas. A GTVDCE contour was manually constructedfrom regions of delayed contrast enhancement within thepancreas appearing on the late arterial-phase DCE image.To eliminate window level and width variability fromaffecting GTV delineation, the mean DCE and ADC win-dow level and window width were calculated after opti-mization on a patient-specific basis.

Tumor maximum dimension

The maximum dimension (MD) of GTV contour for eachimaging modality, sequence, or thresholding method wasmeasured by drawing a cord through the GTV volume onan axial slice with maximal surface area.

Delineation of organs at risk

Organs at risk, including normal pancreas, duodenum,stomach, liver, and kidneys, were manually delineated onCT (VCT), T1-weighted MRI (VT1), and T2-weighted MRI(VT2). The T1 images were acquired using a breath-held 3Dvolumetric interpolated breath-held examination sequence.The T2 images were acquired using a respiratory-gated 2-dimensional half-Fourier acquisition single-shot turbospin-echo sequence (field of view, 380 � 225 mm2; MAT,320 � 190; TE, 96 ms; TR, 2000 ms; GRAPPA [RZ2];TH, 5 mm). Differences in OAR volumes were assessed incomparison with the CT by calculating VT1eVCT, VT2eVCT,and the results were averaged over the entire patient cohort.Similarly, volumetric differences between VT1 and VT2 werealso calculated and averaged over the entire patient cohort.

Volume 89 � Number 3 � 2014 Multimodality imaging for pancreas RT 635

Author's personal copy

Statistical analysis

The GTVs obtained from CT, DCE, ADC, PET-SUV2.5,PET40%-SUVmax, and PET-50%-SUVmax for the patientsstudied were transferred to Prism (version 6; GraphPadSoftware, La Jolla, CA) for statistical analysis. A Kruskal-Wallis test, along with a Dunns post test, was used todetermine whether significant differences existed betweenGTV volumes. Relative GTV volumes were created bysubtracting off the GTVCT volumes from the other volumes.AWilcoxon signed rank test was used to determine whethersignificant differences existed between the relative GTV,VT1eVCT, VT2eVCT, and VT1e VT2 volumes and zero. For allstatistical tests, PZ.05 was used for significance.

Results

Characteristics of GTVs

The mean difference (range) in the measurements of MD ofGTV from DCE, ADC, SUV2.5, 40%SUVmax, and 50%SUVmax compared with pathologic specimens for a totalof 8 patients were �0.84 (�2.24 to 0.9), 0.41 (�0.15 to2.3), 0.58 (�1.41 to 3.69), 0.66 (�0.67 to 1.32), and 0.15(�1.53 to 2.38) cm, respectively. Generally, 50%SUVmaxand ADC grossly agreed with pathologic specimens ifcompared with DCE, SUV2.5, and 40%SUVmax.

A graphic comparison of the differences in GTV sizesobtained from DCE, ADC, SUV2.5, 40%SUVmax, and50%SUVmax for 19 patients is displayed in Figure 1. Themean and standard deviation of GTVs obtained from CT,DCE, ADC, PET with fixed SUV (ie 2.5) threshold, andPET with relative SUV (ie 40% of SUVmax) threshold forall the patients studied are presented in Figure 2A.Numerically, the mean (range) of GTVs for GTVDCE,GTVADC, GTV2.5, GTV40%, and GTV50% were 4.73 (1.00-9.79), 14.52 (3.21-25.49), 22.04 (1.00-45.69), 19.10 (4.84-45.59), and 9.80 (0.32-35.21) cm3, respectively. The meandifferences and difference ranges between the GTV of

DCE, ADC, SUV2.5, 40%SUVmax, or 50%SUVmax andthe GTV from 3D/4D CT were �37.26 (�76.20 to �5.32),�31.24 (�79.40 to �6.40), �18.62 (�52.20 to 2.62),�23.18 (�56.49 to 42.17), or �32.50 (�69.84 to 25.55)cm3, respectively, and are plotted in Figure 2B. The sta-tistically significant differences are indicated in Figure 2 byasterisks (*).

The DCE-derived GTV was found to be the smallest,with smaller variance compared with ADC and PET. Grosstumor volume contours delineated on DWI, ADC, DCE,and SUV2.5, 40%SUVmax, and 50%SUVmax are shownin Figure 3. Variability seen on the GTV contours inFigure 3 is in part due to tumor motion and image regis-tration errors. Variability produced in thresholded PETimaging using the fixed value of SUV2.5 or a percentage ofSUVmax are shown in Figure 4. Figure 5 shows the overlapof GTV contours obtained from DCE, ADC, SUV2.5, 40%SUVmax, 50%SUVmax, and CT ITV reconstructed fromwhole 10 respiratory phases. All images were rigidly cor-egistered, one by one, to the CT image of the PET/CTmodality, followed by manual translations for reasonablygood fusion match. Each image was color washed beforecoregistration, then contours were delineated manually onthe coregistered image.

Because the MRI, PET, and CT were often acquired atdifferent times/days, we assessed the impact of thisimaging-time difference on the difference of the GTVsdefined by these images, although the actual GTV may notchange substantially during the imaging period. The datafor a group of 8 patients who had PET and MRI acquired onthe same day were analyzed. The mean (variation ranges)of the GTVs defined by the different imaging and/orthresholding methods, GTVDCE, GTVADC, GTV2.5,GTV40%, and GTV50%, were 4.686 (2.13-8.76), 11.54(5.59-25.48), 24.04 (1.00-45.69), 17.97 (4.84-36.5), and9.27 (1.51-20.19) cm3, respectively. These differences be-tween these GTVs were found to be statistically significant(PZ.0001). Furthermore, the differences between the2 MR methods as well as between the 3 PET techniques,which were obtained at the same patient setup, were also

Fig. 1. Comparison of gross tumor volumes (GTVs) obtained from late arterial-phase dynamic contrast-enhanced (DCE),apparent diffusion-coefficient (ADC), positron-emission tomography standardized uptake value (PET-SUV)2.5 (SUV2.5), PET-40% SUVmax (PET40%SUVm), and PET-50% SUVmax (PET50%SUVm) for each of the patients included in the study.

Dalah et al. International Journal of Radiation Oncology � Biology � Physics636

Author's personal copy

statistically significant. This confirms that the observedGTVs differences were not primarily due to the differentimaging times.

To account for intraobserver variability, GTVs fromDCE, ADC, SUV2.5, 40%SUVmax, and 50%SUVmaxwere delineated and repeated multiple times by 1 observerfor a randomly selected patient. The variation, mean GTV� 1 standard deviation, were 2.49% � 0.69%,2.67% � 2.69%, 3.35% � 0.90%, �0.12% � 5.28%, and�0.14% � 4.16% for GTVDCE, GTVADC, GTV2.5,GTV40%, and GTV50%, respectively, indicating that intra-observer variation is small.

Delineation of OARs

The T1- and T2-MRI based contours were generallysmaller than those based on CT, except for the kidneys. The

average volumetric changes between VT1eVCT, VT2eVCT,and VT1eVT2 for duodenum, stomach, and kidneys were notstatistically significant. However, significant differenceswere found in normal pancreas volumes between T2-MRIand CT (PZ.0313) as well as between T1- and T2-MRIvolumes (PZ.0156), as shown in Figure 6A. In addition,significant differences in liver volumes were found betweenT1-MRI and CT (PZ.0156) and between T2-MRI and CT(PZ.0156), as shown in Figure 6B.

Discussion

Identifying imaging methods to accurately define GTV forRT of LAPC are desirable (5, 12). The rationale behindchoosing DCE and ADC methods together with FDG-PETwas based on the yield of additional information eachmethod offers about the GTV. The DCE images

Fig. 2. (A) Mean and standard deviation of gross tumor volumes (GTVs) from 3-dimensional/4-dimensional CT, latearterial-phase dynamic contrast-enhanced (DCE), apparent diffusion-coefficient (ADC), positron-emission tomographystandardized uptake value (PET-SUV)2.5, PET-40% SUVmax, and PET-50% SUVmax for the patents studied. (B) Mean andstandard deviation of differences in DCE, ADC, and PET GTV relative to CT GTV (B). )Statistically significant differences.

Fig. 3. Illustration of gross tumor volume contours on diffusion-weighted imaging (DWI) (b Z 1000 s/mm2) (top left),apparent diffusion-coefficient (ADC) map generated from DWI (top right), late arterial-phase dynamic contrast-enhanced(DCE) (bottom left), and positron-emission tomography (PET) using standardized uptake value (SUV)2.5 (yellow), 40%SUVmax (magenta), and 50% SUVmax (cyan) (bottom right). The upper right corner inset in each panel is an enlarged view.A color version of this Figure is available at www.redjournal.org.

Volume 89 � Number 3 � 2014 Multimodality imaging for pancreas RT 637

Author's personal copy

demonstrate high spatial resolution, high contrast resolu-tion, low geometric distortion, reduced partial volume ef-fects (PVEs) (13), and are linked to the intense-fibroticnature of the tumor (14). Using this method, pancreaticcancer appears hypointense relative to normal pancreas onlate arterial-phase DCE images. The decreased enhance-ment visible on the late arterial-phase or early venous-phase images is due to the delayed gadolinium uptake inthe reduced extracellular space, reflecting the desmoplasticnature of this cancer.

Diffusion-weighted MRI is quantitative, noninvasive,requires no contrast agent, is easily incorporated intoroutine patient evaluations, and is associated with lesionaggressiveness in oncology (15). However, despite theseadvantages, DWI can be confounded by T2 shine-througheffects and perfusion in vascular-rich tissue. Through theuse of higher b-values (ie, �500 s/mm2) (16) and genera-tion of ADC parametric maps, these issues are largelyovercome. Similar to DCE, lesions of pancreatic carcinomaappear hypointense on ADC maps, demonstrating theconstrained Brownian motion of water related to the highcellularity of pancreatic cancer.

Tissue metabolic information using FDG-PET demon-strates an increased accuracy of tumor detection (17, 18).However, the ability of metabolic imaging of FDG-PET toaccurately define the GTV has been limited by the loss of

accuracy of PET in assessing tumor larger than 4 cm,partially related to the low metabolic rates in portions oflarger tumors (19). However, for small (ie, <2 cm) hyper-metabolic tumors, the sensitivity of PET is superior to thatachieved with CT, yet GTV accuracy is challenging owingto the PVE associated with feature sizeedependent 3Dblurring effects. Another drawback using PET is the vari-ability created using different threshold values on the GTVmeasurement. Generally, applying 40% of the SUVmax tothreshold PET images yields a larger GTV than the onegenerated using SUV2.5 or 50%SUVmax. Choosing 40%SUVmax yields a threshold value that is smaller than 2.5,thus more voxels are included to constitute the GTV, whichsomewhat contradicts the diagnostic principle of usingSUV2.5 to distinguish normal from abnormal lesions in thefirst place.

Recently, studies (10, 11) demonstrated median 7- and4-mm underestimation of tumor MD compared with grosspathologic specimen on CT and late arterial-phase DCE-MRI, respectively. The GTVs defined from DCE-MRIseemed to lead to increasing underestimation for large tu-mors (11). In our study, GTVs from DCE, ADC, PET-SUV2.5, PET-40%SUVmax, and PET-50%SUVmaxdemonstrated approximate mean differences of �8, 4, 6,7, and 2 mm, respectively, in tumor MDs compared withthe pathologic specimen (the minus sign implying

Fig. 4. Illustration of thresholded positroneemission tomography images based on standardized uptake value (SUV)2.5(top left), 40% SUVmax (top right), 50% SUVmax (bottom left), and gross tumor volume contours from SUV2.5 (yellow),40% SUVmax (magenta), and 50% SUVmax (cyan) (bottom right). The upper right corner inset in each panel is an enlargedview. A color version of this Figure is available at www.redjournal.org.

Dalah et al. International Journal of Radiation Oncology � Biology � Physics638

Author's personal copy

underestimation). Interestingly, PET-50%SUVmax andADC seemed to show better agreement with pathology thanDCE, PET-SUV2.5, and PET-40%SUVmax. Generally,DCE-defined GTV was shown to be smaller than ADC(Fig. 1), reflecting the diminished extracellular space andhigh tissue density in pancreatic tumor vasculature. On theother hand, PET-SUV2.5 seemed to encompass ADC plusartifacts from PVE, tumor motion effect, and perhaps thefalse signal coming from the presence of backgroundpancreatitis.

Significant organ deformations exist between T2-MRIand CT, and T1- and T2-MRI for normal pancreas and inthe T1, T2-MRI, and CT for the liver. For the pancreas, thebreath-hold method used in T1-MRI seemed to showsimilar volumetric measurements compared with

respiratory-gated methods used in T2-MRI. A recent study(20) on analyzing the daily CTs collected during pancreasRT reported that pancreatic head deformed with Dicesimilarity coefficient reduced to 75% comparing betweendaily and planning CTs. These data, along with otherpublished work (21), indicate that a deformable registrationis preferred between the images acquired at different times.

To the best of our knowledge there are no studiesdirectly comparing these different methods in the context ofRT planning that requires accurate GTV definition. Weconclude that there is no apparent consensus about theoptimal method/sequence to define the GTV for pancreaticcancer, and further investigations combined with volu-metric information of pathologic specimens are required.

There are several limitations of our study. A largersample size is needed to a draw a firm conclusion, partic-ularly in the comparisons of tumor MD with the pathologicspecimens. However, despite the sample size, statisticallysignificant differences existed among the GTV volumesaccording to different imaging modalities. Another limita-tion may be the use of a fixed SUV value 2.5 or a per-centage of the SUVmax to the threshold PET images.However, no other methods have been proven to be supe-rior, particularly in the pancreas. Additionally, it is wellknown that DWI, obtained using conventional imagingmethods, can be prone to regions of local geometricdistortion. This geometric distortion translates directly tothe corresponding ADC maps, which in turn can translateinto distorted GTV definition using ADC. Furthermore, wereported only comparison of the MD of the pathologicspecimen because 3D information is hard to obtain atpresent. Finally, other factors influencing the GTV mea-surement from DCE, ADC, and PET include the imageregistration error (eg, deformation ignored, different slicethickness), intraobserver variability, and different respira-tion motion management techniques.

Conclusions

This work explores the potential of using various MRI,PET, and CT techniques to try to define treatment targetsfor RT of pancreatic cancer. Significant differences exist in

Fig. 6. Mean and standard deviation of organ at risk volume differences in pancreas (A) and liver (B) between T1-weightedMRI, T2-weighted MRI, and CT. )Statistically significant differences.

Fig. 5. Overlap of gross tumor volume contours delin-eated on dynamic contrast-enhanced map (yellow),apparent diffusion-coefficient map (black), standardizeduptake value (SUV)2.5 (green), 40% SUVmax (blue), 50%SUVmax (red), and CT internal tumor volume recon-structed from whole 10 respiratory phases (white). Allimages were rigidly coregistered with the CT image of thepositron-emission tomography/CT modality, followed bymanual registration to obtain best matching. The upperright corner inset is an enlarged view. A color version ofthis Figure is available at www.redjournal.org.

Volume 89 � Number 3 � 2014 Multimodality imaging for pancreas RT 639

Author's personal copy

the GTVs defined by these imaging modalities and se-quences. Further imaging studies along with pathologiccorrelation are required to establish reliable imagingmodalities and/or techniques for the accurate delineationof tumor target, which will be crucial particularly in doseescalation and/or dose painting for RT of pancreaticcancer.

References

1. Jemal A, Siegel R, Xu J, et al. Cancer statistics, 2010. CA Cancer J

Clin 2010;60:277-300.

2. Aref A, Berri R. Role of radiation therapy in the management of

locally advanced pancreatic cancer. J Clin Oncol 2012;30:1564-1565.

3. Moraru IC, Tai A, Erickson B, Li XA. Radiation dose responses for

chemoradiation therapy of pancreatic cancer: An analysis of compiled

clinical data using biophysicalmodels.Pract Radiat Oncol 2014;4:13-19.

4. Willett CG, Fernandez Del Castillo C, Shih HA. Long-term results of

intraoperative electron beam irradiation (IOERT) for patients with

unresectable pancreatic cancer. Ann Surg 2005;241:295-299.

5. Chang DT, Shellenberg D, Shen J, et al. Stereotactic radiotherapy for

unresectable adenocarcinoma of the pancreas. Cancer 2009;115:665-

672.

6. Hammel P, Huguet F, Van Laethem JL, et al. Comparison of chemo-

radiotherapy (CRT) and chemotherapy (CT) in patients with locally

advanced pancreatic cancer (LAPC) controlled after 4 months of

gemcitabine with or without erlotinib: Final results of the international

phase III LAP 07 study (Abstr.). J Clin Oncol 2013;31(Suppl.).

7. Li X A, Stepaniak C, Gore E. Technical and dosimetric aspects of

respiratory gating using a pressure-sensor motion monitoring system.

Med Phys 2006;33:146-156.

8. Ahunbay EE, Peng C, Chen GP, et al. An on-line replanning scheme

for interfractional variations. Med Phys 2008;35:3607-3615.

9. Li X A, Liu F, Tai A, et al. Development of an online adaptive solution

to account for inter-and intra-fractional variations. Radiother Oncol

2011;3:370-400.

10. Arvold ND, Niemierko A, Mamon HJ, et al. Pancreatic cancer tumor

size on CT scan versus pathologic specimen: Implications for radiation

treatment planning. Int J Radiat Oncol Biol Phys 2011;80:1383-1390.

11. Hall WA, Mikell JL, Mittal P, et al. Tumor size on abdominal MRI

versus pathologic specimen in resected pancreatic adenocarcinoma:

Implications for radiation treatment planning. Int J Radiat Oncol Biol

Phys 2013;86:102-107.

12. Eccles CL, Vallis KA, McKenna WG, et al. Comparison of target

definition methods using 18FDG-PET for radiotherapy planning of

locally advanced pancreatic cancer. Int J Radiat Oncol Biol Phys

2011;81:S732.

13. Koong AC, Le QT, Ho A, et al. Phase I study of sterotactic radio-

surgery in patients with locally advanced pancreatic cancer. Int J

Radiat Oncol Biol Phys 2004;58:1017-1021.

14. Rofsky NM, Lee VS, Pollack MA, et al. Abdominal MR imaging with

a volumetric interpolated breath-hold examination. Radiology 1999;

212:876-884.

15. Wang Y, Miller FH, Chen ZE, et al. Diffusion-weighted MR imaging

of solid and cystic lesions of the pancreas. Radiographics 2011;31:

E47-E65.

16. Padhani AR, Liu G, Koh DM, et al. Diffusion-weighted magnetic

resonance imaging as a cancer biomarker: Consensus and recom-

mendations. Neoplasia 2009;11:102-125.

17. Topkan E, Parlak C, Kotek A, et al. Predictive value of metabolic

18FDG-PET response on outcomes in patients with locally advanced

pancreatic carcinoma treated with definitive concurrent chemo-

radiotherapy. BMC Gastroenterol 2011;11:1-9.

18. Delbeke D, Rose MD, Chapman WC, et al. Optimal interpretation of

FDG PET in the diagnosis, staging and management of pancreatic

carcinoma. J Nucl Med 1999;40:1784-1791.

19. Kalra MK, Maher MM, Boland GW, et al. Correlation of positron

emission tomography and CT in evaluating pancreatic tumors: Technical

and clinical implications. AJR Am J Roentgenol 2003;181:387-393.

20. Liu F, Ericson B, Peng C, et al. Characterization and management of

international anatomic changes for pancreatic cancer radiotherapy. Int

J Radiat Oncol Biol Phys 2012;83:e423-e429.

21. Lu XQ, Shanmugham LN, Mahadevan A, et al. Organ deformation

and dose coverage in robotic respiratory-tracking radiotherapy. Int J

Radiat Oncol Biol Phys 2007;71:281-289.

Dalah et al. International Journal of Radiation Oncology � Biology � Physics640