Deep Shape Features for Predicting Future Intracranial Aneurysm Growth.

Area of Science:

• Neurosurgery
• Medical Imaging
• Artificial Intelligence

Background:

• Intracranial aneurysms (IAs) pose a significant rupture risk, often leading to fatal outcomes.
• Aneurysm growth is a key indicator of rupture risk, necessitating accurate prediction methods.
• Current management strategies for IAs often involve careful follow-up imaging.

Purpose of the Study:

• To develop and evaluate artificial intelligence (AI) prediction models for future intracranial aneurysm growth.
• To utilize baseline aneurysm morphology for computer-aided treatment decision support.
• To assess the efficacy of deep shape features in predicting aneurysm growth.

Main Methods:

• Extraction of vascular surface meshes from CT angiography (CTA) and MRA angiograms.
• Characterization of aneurysm shape using established morphologic and novel deep shape features.
• Prediction of future aneurysm growth using univariate thresholding, random forest, MLP, and deep shape learning (PointNet++).

Main Results:

• The deep shape feature learning method achieved high accuracy (0.82) with excellent sensitivity (0.96).
• Multivariate learning (accuracy up to 0.68) and univariate thresholding (accuracy up to 0.63) were less effective.
• The AI model demonstrated superior performance in classifying growing intracranial aneurysms.

Conclusions:

• Deep shape feature learning offers a high-performing approach for classifying future growing intracranial aneurysms.
• This AI-driven method can serve as a crucial tool for managing rupture risk in patients undergoing follow-up imaging.
• The findings support the integration of AI in the non-treatment paradigm for managing intracranial aneurysms.