Validation of Automated Topological LGE Thresholding for Peri-Infarct Substrate Characterization

Abstract

High-resolution late gadolinium enhancement imaging is a powerful tool for arrhythmia risk assessment post-myocardial infarction, but requires substantial operator time and expertise to analyze. To address this challenge, automated analysis is introduced to isolate and depict relevant image features corresponding to healthy myocardium, peri-infarct gray zone, and dense scar. Using two sets of manual epicardial and endocardial contours, weighted total variation denoising is used to correct for statistical noise, and persistent homology is used to stratify topological features of the image. K-means clustering was used to generate remote myocardium and dense scar signal intensities for automated FWHM thresholding.