Dosimetric effects of anatomical changes during fractionated photon radiation therapy in pancreatic cancer patients

Abstract Pancreatic tumors show large interfractional position variation. In addition, changes in gastrointestinal gas volumes and body contour take place over the course of radiation therapy. We aimed to quantify the effect of these anatomical changes on target dose coverage, for the clinically used fiducial marker‐based patient position verification and, for comparison, also for simulated bony anatomy‐based position verification. Nine consecutive patients were included in this retrospective study. To enable fraction dose calculations on cone‐beam CT (CBCT), the planning CT was deformably registered to each CBCT (13–15 per patient); gas volumes visible on CBCT were copied to the deformed CT. Fraction doses were calculated for the clinically used 10 MV VMAT treatment plan (with for the planning target volume (PTV): D98% = 95%), according to fiducial marker‐based and bony anatomy‐based image registrations. Dose distributions were rigidly summed to yield the accumulated dose. To evaluate target dose coverage, we defined an iCTV +5 mm volume, i.e., the internal clinical target volume (iCTV) expanded with a 5 mm margin to account for remaining uncertainties including delineation uncertainties. We analyzed D98%, Dmean, and D2% for iCTV +5 mm and PTV (i.e., iCTV plus 10 mm margin). We found that for fiducial marker‐based registration, differences between fraction doses and planned dose were minimal. For bony anatomy‐based registration, fraction doses differed considerably, resulting in large differences between planned and accumulated dose for some patients, up to a decrease in D98% of the iCTV +5 mm from 95.9% to 85.8%. Our study shows that fractionated photon irradiation of pancreatic tumors is robust against variations in body contour and gastrointestinal gas, with dose coverage only mildly affected. However, as a result of interfractional tumor position variations, target dose coverage can severely decline when using bony anatomy for patient position verification. Therefore, the use of intratumoral fiducial marker‐based daily position verification is essential in pancreatic cancer patients.


| INTRODUCTION
For pancreatic cancer, radiation therapy is typically planned using a single (three-or four-dimensional) computed tomography (CT) scan.
Between acquiring this planning CT and the start of radiation therapy, as well as during radiation therapy, anatomical changes may take place, including interfractional target position variation and changes in body contour and gastrointestinal gas. Such anatomical changes will introduce differences between planned and delivered dose, possibly compromising target dose coverage. To assess the necessity of monitoring and/or compensating for anatomical changes over the course of radiation therapy, the robustness of treatment plans against these anatomical changes must be investigated.
The dosimetric effects of changes in body contour, e.g., due to weight change, have been studied for head and neck 1,2 and prostate cancer patients. 3,4 For tumor sites in the upper abdomen, including the pancreas, no studies were found.
Dosimetric studies into the effect of gastrointestinal gas changes are scarce. In five-fraction stereotactic body radiation therapy (SBRT) in pancreatic cancer patients, variation in gastric gas volume substantially changed the dose to the planning target volume (PTV). 5 However, when radiation therapy consists of more fractions, the effects of gastrointestinal gas on dose are more likely to average out. For intensity-modulated radiation therapy (IMRT) plans, the dosimetric impact of gastrointestinal gas in pancreatic cancer patients was found to be limited to a few percent, even with large variations in gas volume; 6 only weekly cone-beam CT (CBCT) scans were used, though, preventing analysis of the full effect on accumulated target dose.
Interfractional position variation of the target relative to the bony anatomy can be considerable in pancreatic cancer patients; [7][8][9] in an earlier study, we found in 13 patients a three-dimensional vector displacement of >10 mm in 116 out of 300 fractions. 10 Intratumoral fiducial markers may be used for daily position verification to mitigate this positional uncertainty. The dosimetric effects of interfractional position variation have been investigated in pancreatic cancer patients. [11][12][13] However, in most of these studies, only three or five CBCT or CT scans per patient were used, which may not be representative for a 3-week 15-fraction treatment schedule. 11,12 One study (10 patients) used 107 daily CT scans obtained with an in-room CT-on-rails to characterize anatomical changes and study the dosimetric advantages of online replanning, but did not quantify gastrointestinal gas changes. 13 The aim of our study was to quantify for fractionated photon radiation therapy of pancreatic cancer patients the effect on target dose coverage as a result of changes in body contour, gastrointestinal gas, and tumor position. To do so, we accumulated fraction doses, calculated based on 13-15 daily CBCTs per patient, and analyzed target dose coverage.

2.A | Patient population
We included nine consecutive patients with (borderline) resectable pancreatic cancer in this retrospective treatment planning study.
Inclusion criteria were: intratumoral fiducial markers visible on the planning CT (pCT) and daily CBCTs; patient contour visible over the extent of each CBCT. Implantation of fiducial markers is standard clinical care for (borderline) resectable pancreatic cancer patients in our institute; the fiducial markers (2-4 per patient) are used for daily online CBCT-based position verification. 10,14 Patients were treated between July 2013 and October 2014 with a planned therapy scheme of 13 9 3.0 Gy (patient 7) or 15 9 2.4 Gy (all others; these patients participated in the PREO-PANC trial 15 ). The daily CBCT scans were obtained in free breathing prior to irradiation (Synergy system; Elekta Oncology Systems, Crawley, UK); a total of 132 daily CBCTs (13-15 per patient) were available.

2.B | Target volumes and treatment planning
In our study we used the target volumes and treatment plans as used for these patients in the clinic. The target volumes had been delineated on the pCT, which was either a three-dimensional (3D)-CT (patient 1) or a 4D-CT consisting of ten respiratory phase scans.
The contoured gross target volume (GTV) included suspicious lymph nodes and fat infiltration and had been expanded to an internal GTV (iGTV) that encompassed the GTV on each respiratory phase scan. A margin of 5 mm for microscopic tumor extension had been applied to the iGTV to create the internal clinical target volume (iCTV). 16 For patient 1, the CTV (i.e., the GTV plus a 5 mm margin) as defined on the 3D-CT was used. Application of a 10 mm margin to the iCTV or CTV resulted in the PTV.
We used the 10 MV single-arc volumetric modulated arc therapy (VMAT) plans as used in the clinic, for which the dose to 99% of the PTV had to be at least 95% of the prescribed dose (36 or 39 Gy), the mean kidney dose could not exceed 18 Gy and the mean liver dose had to be ≤30 Gy. With prescription doses of 36 and 39 Gy, constraints for spinal cord, stomach and bowel were irrelevant. Stomach and duodenum were delineated for the purpose of this study by one physician experienced in delineation in pancreatic cancer patients.

2.C | CBCT dose calculations and dose accumulation
As CBCT Hounsfield units (HUs) are inaccurate, performing dose calculations directly on CBCT was not feasible. Therefore, using a method previously applied by our group, 17,18 we deformably regis-  Using the clinical plan, we calculated the dose distribution for the pCT (for the 4D-CTs on the average scan) and each of the CBCTs (i.e., deformed pCTs) in Oncentra (Oncentra 4.3, Elekta, Stockholm, Sweden) for the clinically used patient position (i.e., the online fiducial marker-based image registration). To investigate the effect of interfractional target position variation when fiducial markers would not be used, these calculations were also performed for a shift of the isocenter to simulate bony anatomy-based position verification. For this, we used the bony anatomy-based image registration that in the clinic is always performed prior to the fiducial marker-based registration (XVI, version 4.5, Elekta, Sweden).
For dose accumulation, the fraction doses can be accumulated either deformably or rigidly. To use the deformable image registration to deform the dose distribution back to the planning CT, a voxel-by-voxel accuracy of the deformable image registration would be required. 19 On CBCT, which suffers from poor soft tissue contrast and image artifacts, the target volumes as well as the organs at risk (OARs) are not well visible. Consequently, for these volumes, the deformable image registration is not accurate enough for dose accumulation. Therefore, we accumulated the dose rigidly using Velocity, exploiting the fact that for each patient all CBCTs had been rigidly registered to the pCT based on the fiducial markers. Thus, the fraction dose distributions were all in the same frame of reference as the pCT and could therefore be summed. Such rigid dose accumulation implicitly assumes all volumes remain in a fixed configuration with respect to the intratumoral fiducial markers. For the target volumes, this is a reasonable approach, as marker migration is minimal or absent. 10 For the OARs we do not have such an independent measure of position; hence, rigid dose accumulation will likely not yield a reliable estimate of delivered OAR dose.

2.D | Dose evaluation parameters
For each patient, we obtained DVHs in Velocity, for the planned dose distribution as well as for the two accumulated dose distributions (fiducial marker-based and bony anatomy-based). To account for differences in prescribed dose, dose distribution, and number of CBCTs between patients, we normalized the doses: for each patient, the planned dose DVH was scaled such that for the PTV D 98% = 95% (i.e., 98% of the PTV volume receives 95% of the prescribed dose); 20 this factor was then applied to all accumulated dose DVHs of that patient as well.
As the iCTV-to-PTV margin also accounts for delineation uncertainties, using the iCTV to assess target dose coverage would be inadequate. To evaluate dose coverage, we therefore determined the mean dose (D mean ), near maximum dose (D 2% , highest dose to at least 2% of the volume), and near minimum dose (D 98% ) not only for the PTV but also for the iCTV +5 mm , defined as the iCTV expanded with a 5 mm margin (based on a margin = 2.5 Σ, 21 with an estimated systematic error Σ of 2 mm due to delineation uncertainties).
For the surrounding OARs, rigid dose accumulation will not yield the actual delivered dose per voxel. However, as a coarse measure for (changes in) dose to surrounding healthy tissues, we did analyze the rigidly accumulated dose and determined D mean and D 2 cc (highest dose to at least 2 cm 3 of the volume) for the nearest OARs, i.e., duodenum and stomach.
For each DVH parameter, we tested the difference in group mean over the nine patients between planned and accumulated dose, for fiducial marker-based and bony anatomy-based image registration (two-sided Wilcoxon signed-rank test; a = 0.05). All analyses were done in R. 22

2.E | Parameters of anatomical change
We determined for each CBCT the interfractional tumor displacement, defined as the difference in table shift as obtained from the fiducial marker-based and the bony anatomy-based image registration. Per patient, we calculated the mean and range of the displacements in right-left (RL), posterior-anterior (PA), and superior-inferior To gauge the extent of the changes in gastrointestinal gas, we determined for each pCT and CBCT the relevant gas volume, i.e., the overlap between the gastrointestinal gas and the irradiated volume; as irradiated volume we selected the volume encompassed by the 20% isodose surface. For the CBCTs, this was done for the fiducial marker-based registration only. We used a two-sided Wilcoxon signed-rank test to determine for each patient whether there was a difference in relevant gas volume between CBCTs and pCT (significance level a = 0.05).

| RESULTS
The interfractional position variation varied considerably between patients ( Table 1). The largest mean displacements were found in the SI direction, with a mean displacement of 6.5 mm in inferior direction for patient 2 and of 6.0 mm in superior direction for patient 3.
The mean relevant gas volume on pCT was 150 cm 3 (Table 1).
Over all CBCTs, the range of relevant gas volume was 5-669 cm 3 , with for 74% (98/132) of CBCTs a relevant gas volume <100 cm 3 (Fig. 2). For 83% of CBCTs, the relevant gas volume was smaller than the relevant gas volume on the respective pCT. Shape, position, and volume of gastrointestinal gas could vary considerably between fractions, as illustrated in Fig. 1, which shows for patient 8 the gas volume on pCT [237 cm 3 ; Fig. 1(b)] and on CBCT for fraction 9 [503 cm 3 ; Fig. 1(d)] and fraction 11 [21 cm 3 ; Fig. 1(f)].
For six out of nine patients, the relevant gas volume differed significantly between CBCTs and pCT (Table 1); for these six patients, the mean relevant gas volume was smaller on CBCT than on pCT.
For patient position verification based on fiducial markers, the accumulated dose distribution differed very little from the planned distribution (Table 2)      For each patient, all DVHs are scaled with a single factor such that for the planned dose distribution the D 98% of the PTV = 95%. VAN  center of mass between twice delineated target volumes of 3.0 mm, 23 while another found a mean overall observer variation of 8.0 mm for the delineated iGTV for four patients. 24 However, studies are often limited in size and report different parameters and, in addition, the observer variation is not uniform over the surface of the volume. Therefore, in our study, we used the iCTV +5 mm volume to report target dose coverage, with the margin of 5 mm based on a (low) estimate of the delineation uncertainty (systematic error Σ) of 2 mm. 21 The limited dosimetric effects of changes in gas and body con- When looking at the group mean, dose coverage appears sufficient (i.e., D 98% >95%; However, daily imaging that also visualizes target volume deformations (e.g., MRI or CT) may enable more accurate dose accumulation.
A consequence of the application of fixed CTV and PTV margins and direct copying of the gastrointestinal gas volumes from CBCT to the deformed CT is that the delineated gas can overlap with the iCTV and/or PTV. As gas will result in a decreased calculated dose in that part of the iCTV/PTV, a change in overlap possibly affects target dose coverage for that fraction. For 110 of the 132 fractions (83%), the overlap of gas and iCTV +5 mm was <1% of the iCTV +5 mm volume (fiducial marker-based position verification). The mean overlap over the fractions was ≤0.5% for seven out of nine patients; for patients 2 and 6 the mean overlap was 0.8% and 1.7%, respectively.
With the overlap being small, and not for every fraction at the same location, the overall effect of the overlap on the accumulated dose can be expected to be minimal, as can also be inferred from the limited discrepancies between planned and marker-based accumulated dose for iCTV and PTV. Applying a density override of 1 to the overlap may partially circumvent this issue.
The relevant gas volume was smaller on CBCT than on pCT for the majority of CBCTs. Whether this is a consequence of the radiation therapy is unknown; no apparent time trends in volume were observed (Fig. 2). In particle therapy, changes in gastrointestinal gas can severely affect target dose coverage, as established for carbon ion and proton therapy in earlier studies of our group. 17,27 The observed limited effects of changes in gastrointestinal gas in our current study illustrate the inherent robustness of photon radiation therapy.
With the use of an iCTV or ITV, respiration-induced motion is incorporated into this internal target volume and not into the CTVto-PTV margin. Changes in respiration-induced motion during the course of radiotherapy, which are likely to take place, 28 can affect target dose coverage; an increase in motion with respect to the planning CT may compromise target coverage. Intrafractional motion, however, is outside the scope of our current study, in which we focused on the effects of interfractional changes in gastrointestinal gas, body contour and target position.
In our study, the CTV-to-PTV margin was 10 mm. This margin is used clinically for the fiducial marker-based daily online position verification, but may also be used for bony anatomy-based position verification. 15 Ideally, the applied margin is determined based on calculations incorporating quantified uncertainties determined specifically for the patient population and applied treatment technique. In practice, however, this is not always the case, due to a lack of data on the various uncertainties or the choice made by the clinician in the tradeoff between target dose coverage and dose to healthy tissue. The decrease in target dose coverage for large interfractional position variations when using bony anatomy-based position verification could unquestionably be remedied by the use of a CTV-to-PTV margin larger than 10 mm. This, however, would also increase dose to surrounding organs, especially as the required margins may be >20 mm when taking measured interfractional displacements 10 and delineation uncertainties 24 into account. Therefore, the use of fiducial marker-based position verification is preferred for pancreatic cancer patients.

CONFLI CT OF INTEREST
The authors declare that they have no conflicts of interest.

R E F E R E N C E S
1. Barker JL Jr, Garden AS, Ang KK, et al. Quantification of volumetric and geometric changes occurring during fractionated radiotherapy