MRI-Based Radiomics Ensemble Model for Predicting Radiation Necrosis in Brain Metastasis Patients Treated with...

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Radiation therapy is crucial for brain metastasis but can cause radiation necrosis, leading to neurological deficits.
• Accurate prediction of radiation necrosis is essential for optimizing patient treatment and outcomes.

Purpose of the Study:

• To develop and validate an ensemble machine learning model incorporating radiomic features to predict radiation necrosis.
• To improve the accuracy and reliability of predicting post-radiotherapy necrosis in brain metastasis patients.

Main Methods:

• Retrospective analysis of MRI images and clinical data from 209 stereotactic radiosurgery sessions.
• Development and validation of an ensemble model (gradient boosting, random forest, decision tree, SVM) using radiomic and clinical features.
• Performance evaluation using AUC, sensitivity, specificity, NPV, and PPV; model interpretability via SHAP and LIME analyses.

Main Results:

• The ensemble model demonstrated superior performance, achieving the highest Area Under the Curve (AUC) in the validation cohort.
• The model consistently outperformed individual machine learning algorithms and a stacking ensemble model.
• SHAP and LIME analyses identified key predictive factors, enhancing understanding of radiation necrosis dynamics.

Conclusions:

• The developed ensemble model shows high accuracy and robustness in predicting radiation necrosis.
• This radiomics-based model can serve as a valuable tool to support radiotherapy planning for brain metastasis.
• The findings contribute to personalized treatment strategies, aiming to mitigate treatment-induced complications.