Intracranial Aneurysm Rupture Prediction with Computational Fluid Dynamics Point Clouds

Intracranial aneurysms frequently cause subarachnoid hemorrhageâa life-threatening condition with a high mortality and morbidity rate. State-of-the-art methods of the rupture risk prediction combine demographic, clinical, morphological, and computational fluid dynamics based hemodynamic parameters. We propose a method of blending morphological features, computational fluid dynamics parameters, and patient demographic features. The shape and wall-shear-stress at each point of the aneurysm are encoded with a deep point cloud neural network and extended by additional location encodings of the aneurysm as well as age and sex of the patient. On this concatenated feature vector, an MLP infers the probability of rupture for a given cerebral aneurysm. The proposed network was trained on the CADA - rupture risk estimation challenge set of 109 aneurysms. The proposed method achieves an accuracy of 0.64 and an F2-score of 0.73 on the private CADA - rupture risk estimation challe nge test set of 30 aneurysms.