Location-Dependent Role of CFD-Based Hemodynamic Factors in Intracranial Aneurysm Rupture Prediction
Please login to view abstract download link
Blood flow analysis in intracranial aneurysms is highly dependent on patient-specific arterial geometry and boundary conditions. However, the extent to which hemodynamic factors derived from computational fluid dynamics (CFD) contribute to rupture prediction remains unclear. This study quantitatively evaluates the informational value of CFD-based hemodynamic features in machine-learning–based rupture prediction models by aneurysm location. Among 8,194 unruptured saccular aneurysms followed between 2003 and 2022, 700 aneurysms (including 44 ruptured aneurysms during follow-up) with a maximum size ≤10 mm and completed CFD analyses were included. Aneurysms were classified into the anterior communicating artery (ACA), internal carotid artery (ICA), middle cerebral artery (MCA), and vertebrobasilar artery (VABA). CFD analysis was conducted using patient-specific three-dimensional arterial geometries under an incompressible laminar, unsteady flow assumption. A pulsatile inflow rate was prescribed at the inlet, a fixed pressure of 0 Pa at the outlet, and rigid no-slip conditions at the vessel walls. Unstructured meshes with tetrahedral elements and prismatic near-wall layers were generated, and the finite volume method was employed. Hemodynamic factors such as wall shear stress were incorporated into a Random Forest–based rupture prediction model (Model A), while a model excluding CFD-based features (Model B) was constructed for comparison. Model A achieved a slightly higher overall harmonic mean than Model B (0.81 and 0.80, respectively), with notable improvements in sensitivity for ACA aneurysms (from 67% to 75%) and in specificity for ICA (from 81% to 87%) and MCA (from 79% to 85%) aneurysms. In ACA aneurysms, unilateral inflow configurations associated with underdeveloped anterior communicating arteries produced elevated wall shear stress, and rupture occurred even in small and low aspect ratio aneurysms. These findings demonstrate that the effectiveness of CFD-based hemodynamic information is not uniform but strongly depends on vascular geometry and inflow conditions, particularly in anatomically variable regions such as the ACA.
