Machine Learning Classification of Cerebral Aneurysm Rupture Status with Morphologic Variables and Hemodynamic...

Area of Science:

• Neurosurgery
• Biomedical Engineering
• Radiology

Background:

• Cerebral aneurysms pose a significant risk of rupture and hemorrhage.
• Accurate prediction of rupture status is crucial for clinical decision-making.
• Morphologic and hemodynamic factors are implicated in aneurysm instability.

Purpose of the Study:

• To develop machine learning (ML) models for classifying cerebral aneurysm rupture status.
• To assess the predictive importance of morphologic variables and hemodynamic parameters.
• To identify key features discriminating ruptured from unruptured aneurysms.

Main Methods:

• Retrospective analysis of 226 cerebral aneurysms using computational fluid dynamics (CFD).
• Application of a random forest ML algorithm to create three classification models: morphologic only, hemodynamic only, and combined.
• Evaluation of model accuracy and feature importance for rupture status prediction.

Main Results:

• The combined model achieved the highest accuracy (78.3%), outperforming morphologic (77.0%) and hemodynamic (71.2%) models.
• Morphologic features like projection ratio, irregular shape, and size ratio were significant predictors.
• Hemodynamic parameters such as low shear area ratio and oscillatory shear index were also important.

Conclusions:

• ML models effectively classify cerebral aneurysm rupture status.
• Morphologic features, particularly projection ratio, irregular shape, and size ratio, are critical for identifying ruptured aneurysms.
• Integrating morphologic and hemodynamic data enhances predictive accuracy for aneurysm rupture.