Abstract 4676: MRI-based radiomics ensemble model for predicting radiation necrosis in brain metastasis patients treated with stereotactic radiosurgery and immunotherapy

Abstract This study aims to apply an ensemble model integrating MRI-based radiomics and clinical information as a reliable tool for precisely predicting radiation necrosis, a severe consequence of radiation therapy for brain metastases that requires accurate early prediction to improve patient management. We retrospectively collected and analyzed MRI images and clinical information from 209 stereotactic radiosurgery sessions involving 130 patients with brain metastasis. Radiomic features were extracted from MRI using PyRadiomics and selected via L2 regularization and coefficient analysis. An ensemble model integrating gradient boosting, random forest, decision tree, and support vector machine as base regressors was developed using a soft voting approach to generate the final prediction of the likelihood of necrosis. Performance was assessed and compared with other machine-learning algorithms using metrics including the area under the curve (AUC), sensitivity, specificity, negative predictive value (NPV), and positive predictive value (PPV). SHapley Additive exPlanations (SHAP) and local interpretable model-agnostic explanations (LIME) analyses were applied to explain the model's prediction. The soft-voting ensemble model demonstrated strong performance in the validation cohort, achieving the highest AUC of 0.873 (95%CI: 0.672 - 1.000). It consistently outperformed individual models and the stacking ensemble model, exhibiting superior accuracy, generalizability, and reliability in predicting radiation necrosis. Both univariate and multivariate logistic regression analyses were performed and confirmed the model as the strongest predictor. SHAP and LIME analyses were employed to interpret the predictive model for radiation necrosis, identifying metastasis volume and the radiomic feature, log-sigma-1-mm_glcd_ldmn, as key predictors. Both analyses highlighted similar significant factors, enhancing the understanding of prediction dynamics. This MRI-based radiomics ensemble model exhibited high accuracy and robustness in predicting radiation necrosis. It has the potential to serve as a novel and valuable tool to facilitate radiotherapy for patients with brain metastasis. Citation Format: Yijun Chen, Corbin A. Helis, Christina K. Cramer, Michael T. Munley, Fei Xing, Qing Lyu, Christopher T. Whitlow, Jeffrey Willey, Michael D. Chan, Yuming Jiang. MRI-based radiomics ensemble model for predicting radiation necrosis in brain metastasis patients treated with stereotactic radiosurgery and immunotherapy [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 4676.