Predicting Cerebral Aneurysms Rupture From Hemodynamics

Cerebral aneurysms are relatively common vascular abnormalities with potentially severe consequences. Although most remain unruptured, a small number do rupture, sometimes leading to subarachnoid hemorrhage. Reliably identifying which aneurysms pose the greatest risk remains an unresolved challenge in clinical practice. In this work, we show that combining computational fluid dynamics (CFD) with statistical testing and basic classification models can help identify features linked to aneurysm rupture and assess rupture risk. Using simulations of 103 aneurysms, we extracted a wide set of hemodynamic metrics, combined with other available geometric, anatomical, and patient-specific information. Using PERMANOVA and simple classification models, we assessed the ability of different datasets to differentiate between ruptured and unruptured cases. Results show that monovariate analyses do not perform especially well, with a prediction accuracy of 65% at best, and that combining features from different categories improves rupture prediction significantly. More specifically, we showed that datasets using non-CFD data reached up to 67% accuracy, while adding flow-based features increased performance to nearly 79%. Some hemodynamic variables also showed stronger statistical relevancy during specific phases of the cardiac cycle, suggesting that transient flow effects may be key in uncovering rupture mechanisms. Overall, our findings demonstrate that combining hemodynamics with statistical techniques enables a robust assessment of rupture risk and offers insights into the underlying mechanisms of rupture. Given rapid advances in image-based hemodynamic analysis using CFD in clinical settings, we anticipate that our approach could be translated into clinical practice to support more informed and individualized rupture risk assessment.