Rupture Risk Assessment for Cerebral Aneurysm Using Interpretable Machine Learning on Multidimensional Data.

Area of Science:

• Neurosurgery
• Medical Informatics
• Biostatistics

Background:

• Cerebral aneurysm rupture risk assessment is critical yet challenging.
• Existing machine learning models for rupture risk often use limited data and lack interpretability.
• Clinical application of predictive models is hindered by a lack of transparency in their decision-making processes.

Purpose of the Study:

• To develop interpretable machine learning models for cerebral aneurysm rupture risk assessment.
• To evaluate the performance of these models against conventional methods using multidimensional patient data.
• To enhance model interpretability using SHAP analysis for clinical utility.

Main Methods:

• Collected demographic, medical history, lifestyle, lipid profile, and morphological data from 374 aneurysms.
• Developed prediction models using Support Vector Machine, Artificial Neural Network, XGBoost, and logistic regression.
• Employed Shapley Additive Explanations (SHAP) for interpretability of the best-performing model.

Main Results:

• The XGBoost model achieved an AUC of 0.882, significantly outperforming logistic regression (0.779) and the PHASES score (0.758).
• Key predictors identified include aneurysm location, size ratio, and triglyceride levels.
• SHAP analysis provided clear insights into model predictions, demonstrated through case studies.

Conclusions:

• Machine learning models, particularly XGBoost, offer superior performance for cerebral aneurysm rupture risk assessment compared to conventional methods.
• Interpretable AI, via SHAP analysis, enhances the clinical feasibility of machine learning in neurovascular applications.
• Multidimensional data integration with interpretable ML holds significant promise for improving patient care and outcomes.