Room: Davidson Ballroom B
Purpose: Small bowel is a radiosensitive critical organ that limits radiotherapy dose to abdominal cancer hosting organs. This work addressed the demand to automatically depicting the complicated small bowel morphology, and provides an important step in personalize treatment adaptation and dose boosting in radiotherapy.
Methods: We first generate a small bowel likelihood map over the imaged abdominal region, combining both voxel intensity and a customized cross-scale flux filter for quantifying local tubularity to characterize the chance of each pixel belong to the small bowel. After that, local sections are formulated based on a state evolution process, using particle filtering as the control mechanism. Finally, to generate the complete small bowel course, a path optimization problem is formulated to find the section order such that each section is traversed exactly once. The ordering problem is then solved with efficient heuristics. The generated results were adjudicated by an experienced radiologist and an oncologist.
Results: Based on the radiologistâ€™s close inspection, the automatically generated small bowel course agrees with clinical judgement. The portion that is close to the duodenum also agrees with the radiation oncologistâ€™s daily contouring for the purpose of adaptive therapy.
Conclusion: We have developed the first automated full extraction and tracing system of the small bowel. This would enable precise understanding of per-fraction dose deposition, accurate spatiotemporal dose accumulation, and facilitate the design of adaptive treatment and potential dose boosting scheme in abdominal radiotherapy treatment.