Machine Learning Models can Detect Aneurysm Rupture and Identify Clinical Features Associated with Rupture

Area of Science:

• Neurosurgery
• Medical Imaging
• Artificial Intelligence

Background:

• Machine learning (ML) is increasingly utilized in neurosurgery.
• ML models can potentially distinguish ruptured from unruptured intracranial aneurysms.

Purpose of the Study:

• To evaluate ML models' ability to differentiate ruptured vs. unruptured aneurysms.
• To identify key aneurysm features associated with rupture.

Main Methods:

• Retrospective review of 845 intracranial aneurysms (2002-2018).
• Trained three ML models: random forest, linear SVM, and RBF kernel SVM.
• Derived predictor importance from the linear SVM model.

Main Results:

• Ruptured aneurysms were larger (6.51 mm vs. 5.73 mm) and more often in posterior circulation (20% vs. 11%).
• Random forest model achieved the highest AUC (0.81).
• Aneurysm location and size were the most significant predictors; specific locations like PCoA, ACoA, and PICA were associated with rupture.

Conclusions:

• ML models accurately distinguish ruptured from unruptured aneurysms.
• Aneurysm location and size are critical features associated with rupture risk.