Intracranial Aneurysm Rupture Risk Estimation Utilizing Vessel-Graphs and Machine Learning

摘要

Intracranial aneurysms frequently cause subarachnoid hemorrhage-a life-threatening condition with a high mortality and morbidity rate. State-of-the-art methods combine demographic, clinical, morphological, and computational fluid dynamics parameters. We propose a method combining morphological radiomics features, gray-level radiomics features, and a novel aneurysm site location encoding via directed graphs on the vessel tree. Some of the gray-level features seem to be good proxies for blood flow within the vessel and the aneurysms. Furthermore, our proposed method shows improved F2-scores and accuracy across various models fed with the aneurysm site encoding. A K-nearest neighbors method shows the best results during our model selection with an F2-score of 0.7 and an accuracy of 0.73 on the relatively small private test set with 22 individuals and 30 aneurysms.