ESTRO 2025 - Abstract Book

S3360

Physics - Machine learning models and clinical applications

ESTRO 2025

Fig.1: Analysis workflow.

Results: Fig.2 illustrates the AUC performance across different timepoints for 74 Gy and 60 Gy groups. In the separate analyses, both groups showed the highest AUCs in training and test sets at week 1, indicating strong early predictive capability (AUCs 0.73 – 0.83). Cumulative analysis revealed increasing AUCs for the 74 Gy group up to week 3, after which it plateaued (AUCs 0.86 (training) and 0.77 (test)), suggesting that extending the model beyond this point offers no significant benefit. Conversely, for the 60 Gy group, week 1 was selected as the optimal timepoint based on its treatment timing and generalisability.