CT-Based Radiomics Model for Predicting Acute Skin Toxicity after Breast Cancer Radiotherapy: A Single-Center Retrospective Study
Wei Ye, Lu Wang, Xiaofeng Wan, Ling Gui, Jindi Peng, Ruijun Niu, Hongbo ZuoBackground
Radiotherapy for breast cancer frequently leads to skin toxicity, but current prediction methods rely on subjective clinical assessments and a narrow range of dosimetric parameters, lacking personalized precision. This study aimed to develop a radiomics model utilizing CT imaging to predict the severity of skin toxicity prior to its clinical manifestation.
Methods
This retrospective study included 200 breast cancer patients who received radiotherapy after radical mastectomy. Radiomic features were extracted from CT images using dose-grid-defined regions of interest, including isodose volumes corresponding to 100%, 105%, and 108% of the prescribed dose, as well as skin regions at 20 Gy, 30 Gy, and 40 Gy isodose levels. Six machine learning algorithms were compared, and logistic regression was selected as the final model. Model performance and clinical utility were evaluated using the AUC, calibration analysis, and decision curve analysis, while SHAP analysis was used to assess feature contributions.
Results
Eight key radiomic features were selected to construct the final model. The logistic regression model achieved an AUC of 0.993 on the test set, with a sensitivity of 0.969 and a specificity of 0.963. Decision curve analysis confirmed its potential clinical net benefit. SHAP analysis revealed that radiomic features extracted from the 105% and 108% high-dose regions were the most influential predictors.
Conclusion
This preliminary CT-based radiomics model provides a non-invasive tool for predicting radiation-induced skin toxicity. However, its clinical applicability warrants further validation in larger, independent, multicenter cohorts.