Machine Learning-Based Prediction of Overall Survival After Lung SBRT Using Clinical, Dosimetric, and Radiomics Features: a multicenter study

Radiomics combined with clinical and dosimetric data may improve outcome prediction after lung stereotactic body radiotherapy (SBRT). This study aimed to develop machine learning models to predict overall survival (OS) after lung SBRT for primary and metastatic lung tumors using planning CT-derived features. Clinical and dosimetric data were retrospectively collected from six centers. Radiomics features were extracted from planning CT scans using double tumor segmentation and selected using statistical filtering. Models were trained using 180 features composed of 124 radiomics, 45 clinical and 11 dosimetric features. A total of 163 patients treated between 2016 and 2018 were included. Patients were divided into primary (n=105) and metastatic (n=58) cohorts. Two XGBoost models were trained to predict OS for either primary or metastatic cohorts, using nested stratified cross-validation and Bayesian hyperparameter optimization. The 20 most influential features identified by SHAP analysis were retained. Death occurred in 26.7% of patients in the primary cohort and 48.3% in the metastatic cohort. The models achieved high predictive performance, with ROC-AUC values of 0.89 and 0.87, respectively. These models provide accurate and well-calibrated OS predictions after lung SBRT, supporting individualized clinical decision-making.