Quantitative Output Factor Predictions for Small Irregular Fields with Fluence Features: A Feasibility Study

摘要: Accurate measurement of small fields, particularly irregular small field output factors (OFs), remains challenging in stereotactic radiotherapy due to limitations in experimental measurements affected by lateral charged particle disequilibrium, partial source occlusion, and detector volume averaging. On the basis of an analysis of fluence features characterizing beam dosimetry at irregular small field scenarios, we developed a machine learning (ML)-based OF prediction method and tested its feasibility with data simulated at various irregular small field scenarios. A dataset of 1178 beam fluence maps and corresponding OFs was simulated. Radiomics features extracted from fluence maps were used to train three ML algorithms: Random Forest (RF), Decision Tree Regression with AdaBoost (ADA), and Gradient Boosting Regression (GBR). The models were tested on independent datasets containing subcircular and complex clinical fields. Results demonstrated high predictive accuracy across all models, with excellent performance in handling irregular field geometries. This framework shows potential for future extension to experimental measurement data, improving its precision, and reducing measurement workload.