ESTRO 2025 - Abstract Book

S3453

Physics - Machine learning models and clinical applications

ESTRO 2025

Results:

An example of an agent’s decision -making is shown in Figure 1b. Table 1 summarizes the reduction in OAR EUBED compared to uniform fractionation and the median number of fractions, alongside the theoretical maximum reduction, determined by a retrospective optimization using full anatomical knowledge. Key findings include: (1) The agents consistently achieved OAR sparing across all uncertainty scenarios attaining ~ 50-65% of the maximum achievable reduction. (2) The median number of fractions was 5-8 fractions fewer than optimal, even with tumor shrinkage. This suggests that the agent’s policy favors fewer fractions to create a buffer for unfavorable anatomies towards the end of treatment.