Program Information
Improving RapidPlan Model by Knowledge-Based Reoptimizing Its Component Plans
Meijiao Wang, Sha Li , Yibao Zhang*, Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education/Beijing), Department of Radiation Oncology, Peking University Cancer Hospital & Institute, Beijing 100142 China. *Corresponding author.
Presentations
TU-RPM-GePD-J(B)-3 (Tuesday, August 1, 2017) 3:45 PM - 4:15 PM Room: Joint Imaging-Therapy ePoster Lounge - B
Purpose: It was reported that RapidPlan knowledge-based treatment planning system improves plan quality. This study aims to investigate if RapidPlan can improve model performance by re-optimizing the manual plans that were used to configure the model (closed-loop evolution).
Methods: Eighty-one best-effort manual plans were used to configure a RapidPlan model (M0) for pre-surgical rectal cancer. The component plans (P0) were reoptimized using M0, yielding P1 (closed-loop). Improved P1 (P1+) were used to replace corresponding P0, yielding M1. Models were evaluated statistically and dosimetrically on 30 patients (Pv) independent from model library (open-loop).
Results: Reoptimization using M0 produced 75 P1+ out of 81 P0, which reduced the mean dose to femoral head (FH), urinary bladder (UB) and small bowel (SB) by 16.11%, 7.89% and 3.16% respectively. Incorporating P1+ into M1 improved coefficients of determination (R²) for FH and UB from 0.411 and 0.356 to 0.777 and 0.579 respectively. Marginally worsened R² for SB was observed (from 0.561 to 0.552). The amount of potential statistical outliers (geometric, dosimetric and influential points) were also reduced from 4 to 3 (FH), 4 to 3 (UB) and 6 to 1(SB) respectively. By comparing the applications of M1 than M0 to Pv, open-loop validation suggested significant reduction of FH mean dose (by 16.15%). Reduced mean dose to FH (by 7.92%) and SB (by 3.16%), V40Gy to FH (by 100%) and UB (by 19.53%), V45Gy to UB (by 6.53%), and V35Gy to SB (by 15.23%) respectively were also observed, although statistically insignificant.
Conclusion: RapidPlan model can improve partial if not most of the component plans in a closed-loop reoptimization. By incorporating the plans of better quality, the performance of RapidPlan model enhanced statistically and dosimetrically as validated by independent open-loop reoptimization. This closed-loop model improving process might be automated via API scripting in the future.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by Beijing Natural Science Foundation (7172048), National Natural Science Foundation of China (11505012), and Beijing Municipal Administration of Hospitals' Youth Programme (QML20151004).
Contact Email: