Determination of Significant Three-Dimensional Hemodynamic Features for Postembolization Recanalization in Cerebral Aneurysms Through Explainable Artificial Intelligence

Abstract

BackgroundRecanalization poses challenge in coil embolization for cerebral aneurysms. Establishing predictive models for post-embolization recanalization is important for clinical decision-making. However, conventional statistical and machine learning (ML) models may overlook critical parameters during the initial selection process. MethodsIn this study, we automated the identification of significant hemodynamic parameters using a PointNet-based deep neural network (DNN), leveraging their inherent three-dimensional (3D) features. Further feature analysis was conducted utilizing saliency mapping, an explainable artificial intelligence (XAI) technique. The study encompassed the analysis of velocity, pressure, and wall shear stress (WSS) in both pre- and post-coiling models derived from computational fluid dynamic (CFD) simulations for 58 aneurysms. ResultsVelocity was identified as the most pivotal parameter, supported by the lowest P-value from statistical analysis and the highest area under the receiver-operating characteristic curves (AUROC)/precision-recall curves (AUPRC) values from DNN model. Moreover, visual XAI analysis revealed robust injection flow zones with notable impingement points in pre-coiling models, as well as pronounced interplay between flow dynamics and the coiling plane were important 3D features in identifying the recanalized aneurysms. ConclusionsThe combination of DNN and XAI was found to be an accurate and explainable approach not only at predicting post-embolization recanalization but also at discovering unknown features in the future.