Decision tree-based machine learning algorithm for prediction of acute radiation esophagitis.

Background: Radiation-induced esophagitis remains a significant challenge in thoracic and neck cancer treatment, impacting patient quality of life and potentially limiting therapeutic efficacy. This study aimed to develop and validate a decision tree-based model for predicting acute esophagitis grades in patients undergoing chemoradiotherapy. Methods: Data from 100 patients receiving thoracic and neck radiotherapy were analyzed. The dataset comprised 33 features, including demographic, clinical, and dosimetric parameters. A decision tree classifier was implemented for both binary (Grade ≥2 vs. <2) and multi-class (Grades 1, 2, and 3) classification. Model performance was evaluated using standard metrics including accuracy, precision, recall, and F1-score. Results: The binary classification model achieved 97 % accuracy in distinguishing acute esophagitis. The multi-class model demonstrated 98 % accuracy in predicting specific grades. Key predictive features included V40 (volume receiving 40 Gy), V60, and average esophageal dose. The model generated interpretable decision rules, with V60 ≥ 2.3 strongly indicating Grade 3 esophagitis. Conclusions: The decision tree model demonstrates high accuracy in predicting radiation-induced esophagitis grades while maintaining clinical interpretability. This approach offers potential for treatment optimization and personalized risk assessment in radiotherapy planning. The model's transparency and reliability make it a promising tool for clinical decision support in radiation oncology.