A Computational Framework for Rapid, Zero-Cost Analysis of Hemodynamic and Structural Factors in Cerebral Aneurysms

Overview

PROBLEM Cerebral aneurysm rupture causes subarachnoid hemorrhage with a 50% mortality rate. The current clinical detection tool, the PHASES score, has an AUC of 0.64 and misclassifies 70-80% of rupture-risk patients as low risk. Though inaccurate, PHASES is used because it requires no cost or computing time. ML models have improved accuracy, but require expensive software, long computing times, dedicated personnel, and provide no patient-specific explanations behind predictions. These limitations prevent them from being implemented in real clinical settings. GOAL Our goal was to design a zero-cost computational framework that predicted cerebral aneurysm rupture risk while balancing accuracy, time, cost, scalability, and patient-specific explainability. METHOD We obtained 100 cerebral aneurysm meshes from three public databases. CFD simulations were automated via the SimScale API to extract hemodynamic features (pressure and neck wall shear stress). Next, a custom-built ParaView macro translated the CFD pressure field into FEA boundary conditions, extracting wall displacement as a structural factor. Twelve morphological factors were reduced to three using t-tests and correlation tests to verify statistical independence across all six factors. A Random Forest classifier was trained on these factors, with SHAP applied to produce a ranked, patient-specific explanation of which factors drove each decision. RESULTS Our model achieved an AUC of 0.834 and a sensitivity of 80.0%, outscoring PHASES and remaining competitive with high-end ML models. CONCLUSION We demonstrated that rupture prediction was possible while balancing accuracy, explainability, speed, and cost, ultimately offering a realistic clinical alternative to PHASES and existing ML models.