Using clinical and radiographic variables to predict intracranial aneurysm rupture status with machine learning

Area of Science:

• Neurosurgery
• Medical Imaging
• Artificial Intelligence

Background:

• Artificial intelligence (AI) and machine learning (ML) are increasingly used to analyze clinical and radiographic data for intracranial aneurysm (IA) rupture prediction.
• Previous studies have explored various ML techniques to correlate patient-specific features with IA rupture status.

Purpose of the Study:

• To apply and compare the performance of multiple ML models, including random forest (RF), XGBoost (XGB), support vector machines (SVM), and multi-layer perceptron (MLP), in predicting IA rupture status.
• To identify key clinical and radiographic features that contribute most significantly to IA rupture prediction.

Main Methods:

• A dataset of 178 IAs with 53 features was analyzed, with highly correlated features removed to reduce redundancy.
• Hyperparameter tuning was performed using grid search, and models were evaluated via 5-fold cross-validation over five iterations.
• Performance metrics including accuracy, precision, recall, F1-score, and area under the curve (AUC) were calculated. The Wilcoxon signed-rank test compared AUC scores.

Main Results:

• Random forest (RF) achieved the highest accuracy (85%) and superior AUC (0.85) compared to XGBoost (0.76), SVM (0.69), and MLP (0.65) models (P < 0.05).
• Fractal dimension was identified as the most crucial feature across all models.
• Three-dimensional (3D) shape features constituted 8 of the top 15 most important features driving model performance.

Conclusions:

• RF models demonstrate high accuracy and balanced precision/recall for predicting IA rupture.
• 3D geometric features are critical predictors of IA rupture status, underscoring their importance in clinical assessment and AI-driven analysis.