Hemodynamic predictors of cerebral aneurysm rupture: A machine learning approach

Cerebral aneurysms, a common yet silent condition, affect many people worldwide. Proper treatment selection is crucial because the disease's severity guides the course of treatment. An aneurysm in the Circle of Willis is particularly concerning due to its potential for rupture, leading to severe consequences. This study aims to predict the rupture status of cerebral aneurysms using a comprehensive dataset of clinical and hemodynamic data from blood flow simulations in real three-dimensional geometries from past patients. The Carreau–Yasuda model was used to capture the effects of shear thinning, considering blood as a non-Newtonian fluid that affects the hemodynamic properties of each patient. This research provides insights to aid treatment decisions and potentially save lives. Diagnosing and predicting aneurysm rupture based solely on brain scans is challenging and unreliable. However, statistical methods and machine learning (ML) techniques can help physicians make more confident predictions and select appropriate treatments. We used five ML algorithms trained on a database of 708 cerebral aneurysms, including three clinical features and 17 hemodynamic parameters. Unlike previous studies that used fewer parameters, our comprehensive prediction approach improved prediction accuracy. Our models achieved a maximum accuracy and precision of 0.79 and a recall rate of 0.92. Given the condition's critical nature, recall is more vital than accuracy and precision, and this study achieved a fair recall score. Key features for predicting aneurysm rupture included aneurysm location, low shear area ratio, relative residence time, and turnover time, which significantly contributed to our understanding of this complex condition.