Room: Track 2
Purpose: The difficulties in assessing daily, online adaptive plan quality and safety in magnetic resonance imaging guided radiation therapy (MRgRT) serve as barriers to its widespread use. In this study, knowledge-based prediction models have been used to analyze plan quality and robustness in both 6°Co and linac based adaptive MRgRT. We explore online adaptive plan variability and quality relative to fully optimized, offline plans in order to scrutinize current MRgRT workflows, processes, and hardware.
Methods: Over 600 online adaptive plans from 84 patients undergoing adaptive MRgRT for pancreas cancer were included. Linear regression models developed using plans created offline after patient simulation were used to analyze and compare online plans that were adapted and reoptimized in real time right before treatment with patient on table. Similar models were also developed to compare 6°Co and linac online adaptive plans directly. A few plans that were identified by the moedels as having inferior plan quality were manually re-optimized to assess for potential improvement.
Results: Prediction model accuracy was 4-5% and 5-8% for V95 and D95, respectively. Roughly one third of the time 6°Co adapted plans were of inferior quality relative to similar offline plans, as measured by V95 and/or D95 model predictions. Model predictions for linac adapted plans were not statistically different from their clinically observed plan metrics. Inferior plans that were identified by the models were improved through retrospective, manual re-optimization, in some cases dramatically.
Conclusion: Prediction models for treatment plans in online adaptive MRgRT enable improved adaptive planning strategies and workflows through more informed plan optimization and evaluation in real time. The models succeed in identifying inferior plans that can be improved through further optimization. These results facilitate potential real-time improvement of future adaptive radiation therapy plans prior to their approval and use in patients.
Funding Support, Disclosures, and Conflict of Interest: Parag Parikh has received honoraria from ViewRay outside the current work. Research reported in this study was partially supported by the Agency for Healthcare Research and Quality (AHRQ) under award R01-HS022888. The content is solely the responsibility of the authors and doesn't necessarily represent the official views of the AHRQ.