Identifying potential (re)hemorrhage among sporadic cerebral cavernous malformations using machine learning.

Area of Science:

• Neurology
• Medical Informatics
• Biostatistics

Background:

• Sporadic cerebral cavernous malformations (CCM) pose a significant risk of hemorrhage.
• Predicting hemorrhage risk in CCM patients is crucial for effective management but remains challenging.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting potential hemorrhage in sporadic CCM patients.
• To identify key features contributing to hemorrhage risk prediction.

Main Methods:

• Utilized a dataset of 517 sporadic CCM patients with 5-year follow-up data.
• Constructed and compared Support Vector Machine (SVM), stacked generalization, and Extreme Gradient Boosting (XGBoost) models.
• Evaluated model performance using Area Under the Receiver Operating Characteristic Curve (AUROC) and Area Under the Precision-Recall Curve (PR-AUC).

Main Results:

• The XGBoost model demonstrated superior performance, achieving a mean AUROC of 0.87 during cross-validation.
• The All-Elements XGBoost model achieved an AUROC of 0.84 and PR-AUC of 0.49 in the testing set.
• A 4-Elements XGBoost model, using top SHAP-identified features, achieved an AUROC of 0.83 and PR-AUC of 0.40.

Conclusions:

• Machine learning models, particularly XGBoost, can accurately identify sporadic CCM patients at risk of hemorrhage within 5 years.
• These predictive models offer valuable insights for clinical decision-making in CCM patient management.
• The study highlights the potential of data-driven approaches in neurological disorder management.