3D cardiac shape analysis with variational point cloud autoencoders for myocardial infarction prediction and virtual heart synthesis

Marcel Beetz, Abhirup Banerjee, Lei Li, Julià Camps, Blanca Rodríguez, Vicente Grau

Abstract: Highlights•Multi-class Point VAE: novel geometric DL for 3D cardiac shape & function analysis.•Point VAE shows sub-pixel reconstruction accuracy on 10,000+ cardiac point clouds.•Point VAE learns realistic, disentangled features of cardiac shape & function.•Latent space outperforms clinical benchmarks in MI prediction and survival.•In silico simulations of electrophysiology show realistic cardiac variability.