Abstract
BACKGROUND: Intracranial aneurysms at the posterior communicating artery (PCOM) are known to have high rupture rates compared to other locations. We developed and internally validated a statistical model discriminating between ruptured and unruptured PCOM aneurysms based on hemodynamic and geometric parameters, angio-architectures, and patient age with the objective of its future use for aneurysm risk assessment. METHODS: A total of 289 PCOM aneurysms in 272 patients modeled with image-based computational fluid dynamics (CFD) were used to construct statistical models using logistic group lasso regression. These models were evaluated with respect to discrimination power and goodness of fit using tenfold nested cross-validation and a split-sample approach to mimic external validation. RESULTS: The final model retained maximum and minimum wall shear stress (WSS), mean parent artery WSS, maximum and minimum oscillatory shear index, shear concentration index, and aneurysm peak flow velocity, along with aneurysm height and width, bulge location, non-sphericity index, mean Gaussian curvature, angio-architecture type, and patient age. The corresponding area under the curve (AUC) was 0.8359. When omitting data from each of the three largest contributing hospitals in turn, and applying the corresponding model on the left-out data, the AUCs were 0.7507, 0.7081, and 0.5842, respectively. CONCLUSIONS: Statistical models based on a combination of patient age, angio-architecture, hemodynamics, and geometric characteristics can discriminate between ruptured and unruptured PCOM aneurysms with an AUC of 84%. It is important to include data from different hospitals to create models of aneurysm rupture that are valid across hospital populations.