Accurate estimation of wall shear stress (WSS) in abdominal aortic aneurysms (AAA) is essential for understanding hemodynamic risk, yet conventional computational fluid dynamics (CFD) pipelines remain computationally intensive and difficult to deploy at scale. We investigated the applicability of a template-based deep learning framework (hereafter referred to as WSSNet), originally developed for thoracic aortic WSS inference, to the anatomically heterogeneous AAA. We propose a preprocessing and mesh-alignment workflow that maps patient-specific computed tomography angiography reconstructions and CFD-derived hemodynamic features onto a unified template mesh, enabling consistent spatial correspondence across cases without manual intervention. Using a cohort of patient-specific AAA geometries, we generated paired CFD–deep learning (DL) datasets and designed twelve controlled experimental scenarios to quantify the influence of (i) input spatial resolution, (ii) neighborhood distance configuration, (iii) temporal data splitting to assess generalization, and (iv) synthetic Gaussian noise to emulate clinical acquisition variability. Model performance was evaluated with mean absolute error (MAE), structural similarity index, relative error, and Pearson correlation, complemented by linear regression and Bland–Altman analyses. WSSNet achieved high agreement with CFD references in AAA, and specific input configurations—particularly balanced neighborhood distances and higher mesh resolutions—significantly improved accuracy and robustness, while noise studies delineated tolerance margins relevant for clinical imaging. The proposed pipeline demonstrates a viable path to near–real-time, patient-specific WSS assessment in AAA and offers practical guidance on input design for deploying DL surrogates in vascular hemodynamics.
You et al. (Wed,) studied this question.