SU-E-T-177: Experience Based Predicition Model For Automated VMAT Planning: A Cervical Cancer Application

Abstract

Purpose: To develop a planning prediction model based on patient geometry and dose-volume endpoins from previously treated cases; to devise an automated volumetric modulated arc therapy (VMAT) planning for future cases. Methods: Thirty cervical patients were included for this retrospective study, including22 patients fortraining cohort and 8 for testing cohort. For each patient in the training cohort, a VMAT plan with two full arcs weregenerated using a clinical plan template. The relative volume of the selected organ-at-risks (OARs) within a specified margin of the PTV(Lx) were extracted from overlap volume histograms (OVHs). A prediction model at 2D dose-distance (Lx, Dx) grid were established using a linear regression model. For the testing cohort,the model predicted DVH endpoints were used as constraints to automatically generate a new VMAT plan, the new plans were evaluated against the original plans. Results: On average, the prediction doses of rectum, bladder and bowel were 1.13Gy, 2.09Gy, 0.81Gy lower than the manual VMAT plans at predicted DVH endpoints. The auto plan showed slightly lower PTV conformity,CI = 1.18(auto) vs CI =1.12(manual) and almost the same PTV uniformity (UI= 0.17(auto) vs UI=0.18(manual)). V40 of rectum and bladder and V30 of bowelfrom auto plan were reduced by 2.36%, 13.56% and 4.36%, respectively. Mean doses of the rectum, bladder and bowel reduced 0.35Gy, 2.17Gy and 1.08Gy, respectively, as compard with the manual plans. Conclusion: The experience-based prediction model has demonstrated the ability as a plan quality control tool to further improve OARs dose constrains setting in optimization process, butoffers a potential method for generating automatic VMAT plans which significantly improve theplanqualityand planning efficiency.