Author: Asma Amjad, Renae Conlin, Beth A. Erickson, William Hall, Eric S. Paulson, Christina M. Sarosiek 👨🔬
Affiliation: Department of Radiation Oncology, Medical College of Wisconsin 🌍
Purpose: The adapt-to-shape (ATS) workflow on the MR-Linac involves manual contour edits followed by treatment plan reoptimization on daily pre-beam MRIs. A verification image is acquired after plan optimization to assess the dose distribution with respect to baseline drift, organ motion, or bulk patient motion using the pre-beam contours. We introduce here a method to automatically detect hotspots in the GI structures, prior to treatment, by updating the organ-at-risk (OAR) contours based on intrafraction motion occurring during ATS planning for pancreatic cancer.
Methods: The contours and dose distribution created on the daily pre-beam MR are first propagated to the verification MR. A deep learning(DL)-based contour refinement tool designed to improve propagated abdominal OAR contours is applied and the D0.03cc is updated. If the D0.03cc was above the maximum dose tolerance (34Gy for duodenum and 32Gy for stomach/colon/small bowel) then a hotspot is detected. We tested the method on six pancreatic cancer patients treated to 35Gy-40Gy/5Fx with the ATS workflow. Pre-beam and verification images were acquired with a fat-suppressed 3D-Vane btFE sequence. Manually drawn contours were added to the verification images as ground truth(GT). The propagated and DL-refined contour quality was assessed against the GT contours with Dice similarity coefficient (DSC) and mean distance to agreement (MDA).
Results: The median DSC and MDA was 0.641/0.737 and 5.34mm/3.77mm for the propagated/DL-refined contours. The sensitivity/specificity for detecting a hotspot was 0.82/0.69, resulting in accurate categorization (hotspot/no hotspot) in five patients for all organs except colon. Accurately detected hotspots showed D0.03cc as high as 4109cGy/3516cGy/3300cGy/3418cGy in the duodenum/stomach/colon/small bowel. The D0.03cc using the corresponding GT/propagated contours were 3516cGy/3292cGy, 3355/2901cGy, 3355cGy/2382cGy, 3482cGy/3072cGy.
Conclusion: The proposed method detects out-of-tolerance hotspots in OARs due to motion during ATS planning for pancreatic treatments and can aid in the clinical decision to re-optimize the plan on the verification MR.