Author: Michael Cummings, Olga M. Dona Lemus, Hana Mekdash, Tyler Moran, Alexander R Podgorsak, Sean M. Tanny, Matthew J. Webster, Lexiang Yang, Dandan Zheng, Yuwei Zhou, Xiaofeng Zhu 👨🔬
Affiliation: Department of Radiation Oncology, University of Rochester, University of Miami, Inova Schar Cancer Institute, University of Rochester 🌍
Purpose: oART is revolutionizing radiotherapy by allowing treatment plans to be adjusted based on daily imaging, improving targeting precision and potentially enhancing patient outcomes. However, its implementation is hindered by the substantial clinician effort required to create and evaluate multiple adaptive plans, increasing workload and extending patient time on the treatment table. This study aims to address these challenges by developing a decision-support tool that leverages advanced statistical and machine learning techniques to identify when adaptive planning is truly necessary, optimizing resource use while maintaining oART benefits.
Methods: This proof-of-concept study analyzed 50 oART sessions from two post-prostatectomy patients, targeting the surgical bed and lymph nodes, using retrospective, anonymized CBCT imaging and plan data. Anatomical changes between reference CT at simulation and daily CBCT images were quantified using the Dice coefficients of the targets and OARs and correlated with dosimetric gains calculated from Signed Dose Difference scores based on physician-defined dose constraints between scheduled and adaptive plans. Correlations between Dice similarity and dosimetric gains were analyzed using segmented regression to determine thresholds where adaptation yielded negligible benefit.
Results: Our results revealed strong correlations between rectum anatomy similarity and dosimetric gains (R² = 0.8–0.9 with an exponential fit), indicating rectal anatomy as a key driver of the need for adaptation. A breakpoint of 0.76–0.78 Dice coefficient was identified, beyond which adaptation offered minimal dosimetric improvement. Bowel showed moderate correlations, while bladder and target regions demonstrated weaker or no correlation. These findings provide actionable insights into when adaptation is most beneficial, identifying critical breakpoints for decision-making during treatment.
Conclusion: This study demonstrates an actionable link between anatomical changes and dosimetric benefits in oART, establishing a decision-making threshold to guide adaptation. A prototype tool utilizing these findings was developed to reduce unnecessary adaptations, optimize workflows, and enhance patient and clinician experiences.