Purpose/Objective(s) Technological advancements for locally advanced pancreatic cancer (LAPC) have enabled higher radiation doses to tumor and lower doses to nearby Organs at Risk (OAR). We present a single-institutional prospective phase-II study investigating the efficacy of MRI-defined moderate dose-escalated radiation (RT) in patients with LAPC after induction chemotherapy (IC). Materials/Methods Patients with LAPC after > 4 months of IC were eligible. All underwent CT and MRI simulation for accurate GTV definition. Doses of 50.4 Gy-51.15 Gy in 28-31 fractions (fx) were prescribed to an elective CTV, and a SIB was delivered at 2.25Gy/fx from 63 - 69.75Gy to the GTV, prescribing to the highest dose level (within protocol ranges) while prioritizing OAR constraints. Those enrolled after 2020 were treated on a 1.5T MR-LINAC. The coprimary endpoints were 2yr overall survival (OS) and median survival (MS). Secondary endpoints included Local Control (LC), Distant Metastasis-Free Survival (DMFS), 1yr OS, and toxicity. This study was designed with 80% power to detect a 15mo increased MS compared to historical controls with alpha = 0.05, requiring 23 patients. Descriptive statistics, Kaplan Meier and log-rank test were performed. Results This study enrolled 23 patients from 2016-2021. Five patients were treated to 63gy/28fx (22%), 8 to 65.25gy/29fx (34%), 5 to 67.5gy/30fx (22%), and 5 to 69.75gy/31fx (22%). Six patients (26%) were treated on an MR-LINAC, 5 of which were treated to the highest dose level and used an adapt-to-position workflow. All had cT4 disease, and 61% were cN0. Most lesions were in the head / body of the pancreas (88%). FOLFIRINOX (70%) or Gemcitabine/nab-paclitaxel (30%) represented an average of 7.9 (5-13) cycles of IC. All received concurrent gemcitabine (95%) or capecitabine (5%) with RT. All patients completed treatment per protocol. Median follow-up was 28.1 months. Acute grade 2 toxicity was seen in 35% of patients. There were no acute grade > 3 GI events. Late grade > 2 GI toxicity was seen in 35%, and 17% had grade 3 events. There were no late grade >4 GI events. Table 1 shows the study endpoints. Conclusion This study demonstrated favorable MS and OS for patients with inoperable LAPC after IC treated with moderate dose-escalated RT to an MRI-defined GTV, representing a statistically significant improvement from historical controls measuring from diagnosis. Treatment was well tolerated. Further studies may determine the optimal dose escalation strategy to further improve outcomes.

Manual delineation of pancreatic gross tumor volume (GTV) is generally time consuming and, sometimes, can be challenging, due to the heterogeneity of the pancreas composition and the ill-defined tumor boundary. In addition, fast and consistent segmentation of the GTV is essential in online adaptive radiation therapy (OART). To our knowledge, there is no automatic segmentation methods available for pancreatic tumor. The purpose of this work was to develop a fully-automatic pancreatic GTV segmentation method based on MRI using deep neural networks. We employed a square window convolutional neural network (CNN) architecture with four convolutional layer blocks. The input was a 3D patch with a transverse size of 35×35 pixels and a stack of 1, 3, 5, or 7 slices extracted from MRI along with the manually segmented pancreas and GTV by experienced radiologist and radiation oncologists. The output was the probability that the voxel centered on the patch belonging to the tumor. The probability map was subject to gauss blurring, thresholding, and 6-connected 3D morphology operations to obtain a binary map representing the GTV contour. The CNN model was trained with 88,131 normal pancreas patches and 64,754 tumor patches extracted from 36 T1 arterial phase MRI sets acquired in 27 patients. These images were bias corrected, normalized to the 99.9 percentile intensity of the pancreas tissue, and resampled to a fixed voxel size of 1×1×3 mm3. Patches were randomly subject to one of the eight transformations (0°, 90°, 180°, and 270° rotation, 0°, 45°, 90°, and 135° mirroring) to reduce overfitting. The trained CNN model was tested using 8 different MRI sets from 6 more patients, and the model-generated contours were compared with the manual contours (ground truth) based on dice similarity coefficient (DSC), specificity, sensitivity, Hausdorff distance (HD), and mean distance to agreement (MDA). Performance of the CNN model in terms of DSC increased with increasing patch slice number and saturated after 5 slices. The mean values and standard deviations of the performance metrics of the trained model for the testing dataset were: DSC = 0.72±0.8; Specificity = 0.94±0.4; Sensitivity = 0.79±0.16; HD = 7.7±3.3 mm; and MDA = 1.9±0.6 mm. The volumes of the ground truth GTVs were in the range of 0.35-34.4 cm2 (median 4.2 cm2) and had a moderate positive correlation with DSC and HD (correlation coefficient of 0.41 and 0.64, respectively). The time required to generate a GTV contour in pancreas was 34±12 sec, including pre-processing, CNN patch classification, and post-processing, using a 4-cores 3.4 GHz processor. We developed a CNN based pancreatic GTV segmentation method that is fast in execution and can be fully automated. Efforts are underway to improve its performance by using larger datasets and incorporating multimodal images, making the method robust for MRI-guided OART.

Changes in diffusion weighted image (DWI) parameters during chemotherapy (chemo) for pancreatic adenocarcinoma have been shown to correlate with clinical outcomes. It remains poorly understood if similar changes occur during a treatment course with radiation therapy (RT). The growing prevalence of integrated magnetic resonance imaging (MRI) and linear accelerators (MR-Linac) offers an opportunity to acquire advanced MRI sequences easily and routinely during a course of RT. We sought to determine the feasibility of acquisition and subsequent analysis of DWI using a 1.5 MR-Linac for patients undergoing RT for pancreatic cancer and investigate changes of mean tumor apparent diffusion coefficient (ADC) values during concurrent chemo-RT. Patients undergoing RT for pancreatic adenocarcinoma on conventional RT linear accelerators were enrolled in a prospective imaging trial and had fully quantitative MRI imaging with weekly DWI acquired using a 1.5 MR Linac. DWI was acquired using a free breathing single-shot spin echo EPI sequence (FOV: 380 mm2, matrix: 128x128, TE: 76 ms, TR: 3000 ms, ten b-values ranging from 0 to 800 s/mm). Fat suppression type, acceleration, and number of averages per b-value were optimized to maximize image quality and minimize geometric distortion. Tumors were contoured using imaging processing software and mean ADC values for each tumors were calculated. A total of 6 patients enrolled on this prospective imaging clinical trial (NCT03500081). The median patient age was 65 and the median tumor size was 37mm. Five of six patients were males and one was female. The resectability status included two resectable, one borderline resectable, one locally advanced type A, and two locally advanced type B patients. Each patient had a minimum of two MR scans acquired during their course of treatment, with a total of 20 MRI's with DWI images available. Multiple optimizations were made in the DWI scan parameters including SPAIR fat suppression, SENSE factor of 2.33, and b-value averages ranging between 2 and 7. These changes resulted in a mean DWI scan time of 312 seconds. The range of measured ADC values was 0.69 to 1.76. Measured changes in ADC values were variable; 3 patients demonstrated a measurable increase in mean ADC while 3 patients had stable ADC values with increasing radiation dose. We present the first series to demonstrate the feasibility of acquiring DWI MR images using a commercially available 1.5 MR Linac in patients with pancreatic adenocarcinoma. Our results reveal measurable changes in mean tumor ADC that occur in some patients during a course of RT for pancreatic adenocarcinoma. Such data presents a highly promising methodology for ADC-based biologically adaptive RT and future correlation with pathologic treatment response.

Identification of patients with pancreatic adenocarcinoma pre-treatment that optimally benefit from radiotherapy (RT) presents a common clinical challenge. We sought to identify pre-treatment MRI radiomics features associated with treatment response to chemoradiation therapy (CRT) by correlating pre-treatment MRI texture features with decline in the clinical biomarker CA19-9. The pre-CRT MRI data and pre- and post-CRT CA19-9 levels collected from 45 patients treated with pre-operative CRT for their resectable or borderline resectable pancreatic head cancers were analyzed. Pre-CRT MRIs included breath hold, multi-phase (pre, arterial, venous, portal-venous phase) dynamic contrast T1-weighted and ADC images. T1 images were standardized (bias corrected and normalized) prior to analysis. Registrations of these multi-parametric MRIs were used to delineate gross tumor volume (GTV) by experienced radiologist and radiation oncologists and to generate maps of signal enhancement ratio (SER), ratio of maximum intensity to pre-contrast intensity, and uptake rate difference (URD), difference in maximum intensity and pre-contrast intensity over time. A series of image texture features, including mean, standard deviation (SD), skewness, kurtosis, minimum, maximum, and entropy, were calculated from pre-contrast, arterial, venous, portal-venous, SER and URD images. Texture features were compared to pre-CRT CA19-9 levels using a two-tailed t-test to compare the cohort of patients with CA19-9 levels less than to greater than 35 U/mL. In addition, texture features were compared to percent changes of pre- and post-CRT using ROC analysis to find a threshold with a significant difference in textures and t-test. A two tailed t-test was then applied to investigate possible significant differences in the textures. A threshold was established based on the ROC of the data with a decrease in CA19-9 of 40% from pre-CRT level. The t-test showed using the normalized intensity of pre-contrast phase, there was a significant difference between a decrease greater than 40% and less than 40% in CA19-9 using pre-CRT texture metrics (entropy, p=0.016, mean, p=0.004, and skewness, p=0.008). Evaluating CA19-9 levels directly to texture features resulted in p-values of p= 0.001 for SD of normalized intensity of the pre-contrast phase, p= 0.001 for SD of normalized intensity of the arterial phase, p= 0.002 for entropy of SER maps. Image texture metrics from pre-treatment multiphase dynamic contrast MRI are associated with changes in CA19-9 levels over CRT of pancreatic cancer. These data present a potential promising methodology to identify patients pre-treatment that may optimally benefit from RT. Further comprehensive multivariate analysis is required to validate the presented pre-treatment MRI radiomic features as a biomarker for CRT of pancreatic cancer.

To assess treatment response using textures of uptake rate (UA) and signal enhancement (SE) maps generated from T1-weighted dynamic contrast enhanced (DCE) MRI, acquired before and after neoadjuvant chemoradiation therapy (nCRT), for pancreatic cancer vs. the pathological response. The MRI and pathological response data collected from 24 patients with resectable and borderline resectable pancreatic head cancers were analyzed. This set of patients all received nCRT before tumor resection. Pathological response to nCRT was graded from the surgical specimen as G0, G1, G2, and G3 for complete, near complete, partial, and poor or no response, respectively. The patient pool includes 6 patients for G1, 10 patients for G2, and 8 patients for G3 responses. Of the 24 patients, 20 have pre-nCRT MRIs and 19 have post-nCRT MRIs. The GTVs were contoured based on multi-parametric MRI, including multi-phase dynamic contrast and ADC images, on pre- and post-nCRT MRIs by an expert radiologist. The respiratory gated T1-weighted fat suppressed images were obtained pre-contrast, in the early arterial phase, in the late arterial phase, and the portal-venous phase. These images were used to generate the SE and UR maps. Various texture metrics (mean, standard derivation, minimum, maximum, entropy, skewness, and kurtosis) of the GTV were calculated based on the SE and UR maps using Matlab. A two-tailed t-test was used analyze association between the texture metrics and the pathological responses. The P-values for using the pre-nCRT texture metrics to predict pathological response between the complete (G0 and G1) and the partial-poor (G2 and G3) response groups were: entropy (P(SE) = 0.03, P(UR) = 0.21); and skewness (P(UR) = 0.13). For the post-nCRT metrics, entropy (P(SE) = 0.12, P(UR) = 0.26), SD (P(UR) = 0.03), minimum (P(UR) = 0.03), mean (P(UR) = 0.04), and maximum (P(UR) = 0.001). Looking at the change between pre- and post-nCRT, the strongest separation was found to be in the changes in entropy (P(SE) = 0.01, P(UR) = 0.18), minimum (P(UR) = 0.12), mean (P(UR) = 0.05), and maximum (P(UR) = 0.18). Certain texture metrics from pre- and post- nCRT multiphase dynamic contrast MRI, such as the entropy of pre-nCRT SE map and the change of entropy of SE maps between pre- and post-nCRT, can predict the complete versus partial-poor pathological responses following nCRT of pancreatic cancer. These data indicate that quantitative DCE images acquired prior to and post nCRT for pancreatic cancer can be a method to assess tumor response.

Purpose Accurate identification of the gross tumor volume (GTV) in pancreatic adenocarcinoma is challenging. We sought to understand differences in GTV delineation using pancreatic computed tomography (CT) compared with magnetic resonance imaging (MRI). Methods and materials Twelve attending radiation oncologists were convened for an international contouring symposium. All participants had a clinical and research interest in pancreatic adenocarcinoma. CT and MRI scans from 3 pancreatic cases were used for contouring. CT and MRI GTVs were analyzed and compared. Interobserver variability was compared using Dice's similarity coefficient (DSC), Hausdorff distances, and Jaccard indices. Mann-Whitney tests were used to check for significant differences. Consensus contours on CT and MRI scans and constructed count maps were used to visualize the agreement. Agreement regarding the optimal method to determine GTV definition using MRI was reached. Results Six contour sets (3 from CT and 3 from MRI) were obtained and compared for each observer, totaling 72 contour sets. The mean volume of contours on CT was significantly larger at 57.48 mL compared with a mean of 45.76 mL on MRI, P = .011. The standard deviation obtained from the CT contours was significantly larger than the standard deviation from the MRI contours (P = .027). The mean DSC was 0.73 for the CT and 0.72 for the MRI (P = .889). The conformity index measurement was similar for CT and MRI (P = .58). Count maps were created to highlight differences in the contours from CT and MRI. Conclusions Using MRI as a primary image set to define a pancreatic adenocarcinoma GTV resulted in smaller contours compared with CT. No differences in DSC or the conformity index were seen between MRI and CT. A stepwise method is recommended as an approach to contour a pancreatic GTV using MRI.

PURPOSE:Local recurrence is a common and morbid event in patients with unresectable pancreatic adenocarcinoma. A more conformal and targeted radiation dose to the macroscopic tumor in nonmetastatic pancreatic cancer is likely to reduce acute toxicity and improve local control. Optimal soft tissue contrast is required to facilitate delineation of a target and creation of a planning target volume with margin reduction and motion management. Magnetic resonance imaging (MRI) offers considerable advantages in optimizing soft tissue delineation and is an ideal modality for imaging and delineating a gross tumor volume (GTV) within the pancreas, particularly as it relates to conformal radiation planning. Currently, no guidelines have been defined for the delineation of pancreatic tumors for radiation therapy treatment planning. Moreover, abdominal MRI sequences are complex and the anatomy relevant to the radiation oncologist can be challenging. The purpose of this study is to provide recommendations for delineation of GTV and organs at risk (OARs) using MRI and incorporating multiple MRI sequences.METHODS AND MATERIALS:Five patients with pancreatic cancer and 1 healthy subject were imaged with MRI scans either on 1.5T or on 3T magnets in 2 separate institutes. The GTV and OARs were contoured for all patients in a consensus meeting.RESULTS:An overview of MRI-based anatomy of the GTV and OARs is provided. Practical contouring instructions for the GTV and the OARs with the aid of MRI were developed and included in these recommendations. In addition, practical suggestions for implementation of MRI in pancreatic radiation treatment planning are provided.CONCLUSIONS:With this report, we attempt to provide recommendations for MRI-based contouring of pancreatic tumors and OARs. This could lead to better uniformity in defining the GTV and OARs for clinical trials and in radiation therapy treatment planning, with the ultimate goal of improving local control while minimizing morbidity.

Purpose:It has been reported recently that radiation can induce CT number (CTN) change during radiation therapy (RT) delivery. In the effort to explore whether CTN can be used to assess RT response, we analyze the relationship between the pathological treatment response (PTR) and the changes of CTN, MRI, and PET before and after the neoadjuvant chemoradiation (nCR) for pancreatic adenocarcinoma.Methods:The preand post‐nCR CT, MRI, and PET data for a total of 8 patients with resectable, or borderline resectable pancreatic head adenocarcinoma treated with nCR were retrospectively analyzed. Radiographic characteristics were correlated to PTR data. The histograms, means and standard derivations (SD) of the CTNs in pancreatic head (CTNPH), the GTV defined by ADC (CTNGTV), and the rest of pancreatic head (CTNPH‐CTNGTV) were compared. Changes before and after nCR were correlated with the corresponding changes of ADC, lean body mass normalized SUV (SUVlb), and PTR using Pearson’ s correlation coefficient test.Results:The average mean and SD in CTPH for all the patients analyzed were higher in post‐nCR (53.17 ± 31.05 HU) compared to those at pre‐nCR (28.09 ± 4.253 HU). The CTNGTV were generally higher than CTNPH and CTNPH‐CTNGTV, though the differences were not significant. The post‐nCR changes of mean CTN, ADC, and SUVlb values in pancreatic head were correlated with PTR (R=0.3273/P=0.5357, R=−0.5455/P<0.0001, and R=0.7638/P=0.0357, respectively). The mean difference in the maximum tumor dimension measured from CTN, ADC, and SUVlb as compared with pathological measurements was −2.1, −0.5, and 0.22 cm, respectively.Conclusion:The radiation‐induced change of CTN in pancreas head after chemoradiation therapy of pancreatic cancer was observed, which may be related to treatment responses as assessed by biological imaging and pathology. More data are needed to determine whether the CTN can be used as a quantitative biomarker for response to neoadjuvant therapy.