Motivation: The increasing prevalence of cardiovascular diseases (CVDs) and the demand for precise diagnostic tools underscore the need for effective left ventricular (LV) myocardial segmentation methods in cardiac Magnetic Resonance Elastography (MRE). Goal(s): To assess the feasibility of deep learning-based LV myocardium segmentation in cardiac MRE. Approach: This study compared seven DL models (e.g., U-Net, 3D U-Net) trained on 41 cardiac MRE datasets. Training involved three data preparation approaches, including individual and averaged MRE images. Results: The U-Net architecture, trained on averaged magnitude images, achieved the highest segmentation accuracy (Dice score: 0.77 ± 0.07), highlighting DL's potential for reliable LV segmentation in MRE. Impact: This study demonstrates deep learning's capability to automate left ventricular myocardium segmentation in cardiac MR elastography, enabling faster and more consistent myocardial stiffness assessments. Such advancements could enhance cardiovascular disease diagnostics, paving the way for improved clinical decision-making in cardiology.
Atamaniuk et al. (Tue,) studied this question.