Assessment of intracranial aneurysm rupture risk using a point cloud-based deep learning model

Area of Science:

• Neurosurgery
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate prediction of intracranial aneurysm rupture is crucial for clinical management.
• Existing methods may not fully capture complex aneurysm morphology.
• Deep learning offers a potential avenue for improved risk assessment.

Purpose of the Study:

• To develop and validate a deep learning framework using point clouds for predicting intracranial aneurysm rupture.
• To assess the efficacy of this novel approach compared to existing methods.

Main Methods:

• A dataset of 623 intracranial aneurysms (211 ruptured, 412 unruptured) was utilized.
• 3D aneurysm models were converted to point clouds for deep learning (PointNet++ architecture).
• Two models (dome and cut1) were trained and validated against a LASSO regression model.

Main Results:

• The 'cut1' model achieved a higher AUC (0.85) than the 'dome' model (0.81) in internal validation.
• The 'cut1' model outperformed the LASSO regression model (AUC 0.82 vs. 0.71) in external validation.
• The point cloud approach demonstrated better generalizability than the LASSO model.

Conclusions:

• Point cloud representation effectively captures 3D morphological features of intracranial aneurysms.
• Including parent vessel segments enhances model performance in rupture risk prediction.
• The deep learning model shows potential but requires further research to address generalizability challenges.