Computational Hemodynamics of Left Ventricular Flow Inferred from Two-Dimensional Echocardiography

This study integrates deep learning and computational fluid dynamics (CFD) for cardiac diagnosis. Using the CAMUS dataset, a U-Net architecture segmented the left ventricle from 2D echocardiograms. These geometries provided boundary conditions for transient simulations in OpenFOAM, comparing a healthy baseline and a dilated cardiomyopathy patient. Results show the healthy ventricle develops a highly energetic, coherent central vortex, preserving momentum and ensuring efficient blood washout. Conversely, the dilated ventricle exhibits a sluggish flow with dispersed, low-energy vortex structures, leading to poor washout and energy dissipation. This AI-CFD pipeline demonstrates great potential for predicting cardiovascular risk.