Scar-Related Ventricular Arrhythmia Prediction from Imaging Using Explainable Deep Learning

Description

The aim of this study is to create an automatic frameworkfor sustained ventricular arrhythmia (VA) prediction using cardiac com-puted tomography (CT) images. We built an image processing pipelineand a deep learning network to explore the relation between post-infarctleft ventricular myocardium thickness and previous occurrence of VA.Our pipeline generated a 2D myocardium thickness map (TM) from the3D imaging input. Our network consisted of a conditional variationalautoencoder (CVAE) and a classifier model. The CVAE was used tocompress the TM into a low dimensional latent space, then the classifierutilised the latent variables to predict between healthy and VA patient.We studied the network on a large clinical database of 504 healthy and182 VA patients. Using our method, we achieved a mean classificationaccuracy of 75%±4 on the testing dataset, compared to 71%±4 from theclassification using the classical left ventricular ejection fraction (LVEF).

Abstract

International audience

Additional details