Toward Grading Subarachnoid Hemorrhage Risk Prediction: A Machine Learning-Based Aneurysm Rupture Score

Area of Science:

• Neurosurgery
• Medical Artificial Intelligence
• Biostatistics

Background:

• Current methods for predicting subarachnoid hemorrhage (SAH) lack accuracy and fail to quantitatively compare risk factors.
• Existing scores like PHASES and UIATS have limited predictive power for SAH.

Purpose of the Study:

• To evaluate the Population, Hypertension, Age, Size, earlier Subarachnoid hemorrhage, and Location (PHASES) and Unruptured Intracranial Aneurysm Treatment Score (UIATS).
• To develop an Artificial Intelligence-based Aneurysmal Rupture Criticality Prediction (ARCP) score for predicting 5-year and lifetime rupture risk.
• To quantitatively compare SAH risk factors using data-driven insights.

Main Methods:

• Developed location-specific and ensemble learning models for lifetime rupture risk prediction.
• Utilized longitudinal data and linear regression for aneurysm growth prediction.
• Employed the Apriori algorithm to identify strong SAH risk factors.
• Integrated Apriori and machine learning outputs to create the ARCP score.

Main Results:

• PHASES and UIATS scores demonstrated low sensitivities (22% and 35%) and moderate specificities (76% and 79%).
• Location-specific models achieved high precision and recall for key arteries (e.g., 93% precision, 90% recall for Middle Cerebral Artery).
• The ARCP score outperformed a control group of neurosurgeons in validation, identifying 61 risk factor combinations with varying severity.

Conclusions:

• PHASES and UIATS are identified as weak predictors of SAH.
• The developed ARCP score effectively identifies and grades risk factors associated with SAH.
• AI-driven risk factor analysis offers a significant advancement in predicting aneurysmal rupture.