A novel morphologic and metabolic feature fused treatment response evaluation pipeline for pancreatic adenocarcinoma patients.

311 Background: Adaptive radiation therapy for pancreatic adenocarcinoma (PA) relies on accurate treatment response assessment. Traditional RECIST criteria poorly characterize tumors with complex morphological features, while PET imaging inefficiently detects tumors with intrinsically low standardized uptake value (SUV). Here, we performed regional comparisons of 3D intact PA surfaces pre and post chemoradiotherapy (CRT) utilizing surface measurements containing both morphological and metabolic features to better assess response. Methods: Twenty-one locally advanced PA patients with pre- and 6-8 week post-CRT 18F FDG-PET/CT scans were evaluated. Boundaries of initial and post-CRT tumors were manually defined on respective CT images. On each of the tumors, 3D meshes were generated, followed by surface based registration to achieve vertex-wise correspondence. For each surface vertex, a multivariate vector was formed from two components: anatomic (deformation tensors resulted from surface registration), and metabolic (regional SUV obtained from radius to surface projections). To assess tumor response, paired mahabanobis distance (Mdist) between pre- and post-CRT tumor surfaces with previously formed multivariate vectors were calculated for each patient. Mdist was evaluated using Cox analysis correlated with overall survival (OS) and compared with measurements based on serum CA19-9, volume, SUVmax and SUVmean. Results: Among all the tested parameters, Mdist is the best predictor of OS, with a hazard ratio of 0.437 (p = 0.036). Post-CRT versus pre-CRT ratios based on volume and SUVmax both reached borderline significance (p = 0.0769 and 0.0799, respectively), while CA19-9 and SUVmean failed in predicting OS in our small cohort of patients. Conclusions: We introduced a PET/CT-based novel morphologic and metabolic pipeline for post-CRT response evaluation in locally advanced PA. The fused Mdist outperformed traditional morphologic, metabolic, and physiological measurements in OS prediction. The presented fused model may serve as a new biomarker to better characterize the heterogeneity of tumor response to CRT and a predictive marker for surgical resection.