ESTRO 2025 - Abstract Book

S1278

Clinical – Lower GI

ESTRO 2025

[4].

For NAR classification, 77 patients (36 female, median age 62) were analysed. The best model, developed with DecisionTreeClassifier [4], achieved an AUC of 0.81 (SD 0.04) using demographic, clinical, and radiomic features.

Conclusion: Novel classification ML models based on CT radiomics were developed to predict 3-year OS and binary classification of the NAR score in rectal cancer patients treated with TNT. The inclusion of radiomic features significantly improved predictive performance, highlighting the potential of preoperative CT radiomics for personalized treatment planning in rectal cancer.

Keywords: Radiomics, rectal cancer, machine learning