Interpretable Machine Learning Identifies Key Inflammatory and Morphological Drivers of Intracranial Aneurysm Rupture...

Area of Science:

• Neurosurgery
• Biomedical Engineering
• Data Science

Background:

• Traditional statistical methods for intracranial aneurysms (IAs) identify group-level associations but miss patient-specific nonlinear interactions.
• Accurate rupture risk stratification is crucial for effective patient management and treatment decisions.

Purpose of the Study:

• To develop and validate an interpretable machine learning (ML) pipeline for predicting IA rupture status.
• To identify key predictors, including inflammatory markers and morphological features, that influence IA rupture risk at the patient level.

Main Methods:

• Retrospective analysis of 35 saccular IAs, incorporating demographic, morphological, and inflammatory/immune markers.
• Evaluation of five ML algorithms (DT, AdaBoost, GBM, XGBoost, RF) using stratified five-fold cross-validation and class balancing.
• Primary model (Random Forest) tuned with Optuna and explained using global feature importance and LIME for interpretability.

Main Results:

• The tuned Random Forest model achieved a cross-validation ROC-AUC of 0.98 and maintained a test ROC-AUC of 0.92.
• Key predictors included inflammatory markers (CRP, C3, C4) and morphological features (neck width, irregular shape).
• LIME analysis revealed specific patterns: lower CRP/complement favored 'Not-Broken,' while higher inflammation with smaller, irregular necks indicated 'Broken' status.

Conclusions:

• An interpretable ML pipeline successfully captured nonlinear interactions between inflammation and IA geometry.
• This approach enhances rupture risk stratification and enables patient-level rationale for personalized management.
• Integrating explainable ML with conventional statistics holds significant potential for improving IA treatment quality.