Mesh Convolutional Neural Networks for Wall Shear Stress Estimation in 3D Artery Models

Computational fluid dynamics (CFD) is a valuable tool for personalised, non-invasive evaluation of hemodynamics in arteries, but its complexity and time-consuming nature prohibit large-scale use practice. Recently, the deep learning rapid estimation CFD parameters like wall shear stress (WSS) on surface meshes has been investigated. However, existing approaches typically depend hand-crafted re-parametrisation mesh to match convolutional neural network architectures. In this work, we propose instead networks that directly operate same finite-element as used CFD. We train evaluate our method two datasets synthetic coronary artery models with without bifurcation, using ground truth obtained from simulation. show flexible model can accurately predict 3D WSS vectors mesh. Our processes new less than 5 [s], consistently achieves normalised mean absolute error $\leq$ 1.6 [%], peaks at 90.5 [%] median approximation accuracy over held-out test set, comparing favourably previously published work. This demonstrates feasibility surrogate modelling hemodynamic parameter models.