Deep Learning-Based Echocardiography Wall Thickness Analysis for Heart Failure Detection
IEEE Access · 2026 · 14 · 36924-36932
DOI: 10.1109/access.2026.3670331 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Hsu Thiri Soe und 1 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 20 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Themen: Herzinsuffizienz
Abstract (Originalsprache)
Heart failure remains a major global health challenge, with survival rates declining dramatically from 97% in stage A to only 20% in stage D. Early detection, particularly at stage B in which patients remain asymptomatic but structural changes are already present, is critical for timely intervention. One of the most important markers of early myocardial infarction-related progression to heart failure is regional wall thinning, which is associated with wall motion abnormalities. Cardiologists typically assess wall thinning using the parasternal short-axis (PSAX) view, which allows evaluation of all three coronary artery territories. However, conventional measurements such as the interventricular septal diameter, left ventricular posterior wall diameter, and left ventricular mass typically fail to reflect localized thinning, particularly in the early stages. Furthermore, accurate assessment requires diastolic-phase analysis, which may be challenging and subject to inter-observer variability. In this study, we propose a deep learning-based multi-view echocardiographic wall thickness analysis system for early detection of heart failure. Although PSAX is considered the gold standard for wall thinning evaluation, we hypothesize that integrating complementary views, namely, parasternal long-axis (PLAX) and apical four-chamber (4CH), can improve automated detection by capturing structural and functional information not fully captured in a single view. The proposed system leverages a convolutional neural network trained on 6,000 patient data points with expert annotations, performing both automated wall thickness measurements and heart failure classification. Our multi-view approach achieved 89.3% accuracy in early-stage heart failure detection with 87.1% sensitivity, 91.8% specificity, and an F1-score of 0.89. Regarding individual view performance, PSAX, PLAX, and 4CH achieved accuracy rates of 84.2%, 79.6%, and 76.8%, respectively. The proposed approach considerably improves the accuracy of computer-aided diagnosis, particularly for early heart failure detection, providing a valuable tool for identifying asymptomatic patients at risk of progression to symptomatic heart failure.
Zitieren
APA
Soe, H. T., & Iwata, H. (2026). Deep Learning-Based Echocardiography Wall Thickness Analysis for Heart Failure Detection. IEEE Access, 14, 36924-36932. https://doi.org/10.1109/access.2026.3670331
Vancouver
Soe HT, Iwata H. Deep Learning-Based Echocardiography Wall Thickness Analysis for Heart Failure Detection. IEEE Access. 2026;14:36924-36932. doi:10.1109/access.2026.3670331
BibTeX
@article{Soe2026,
author = {Soe, Hsu Thiri and Iwata, Hiroyasu},
title = {Deep Learning-Based Echocardiography Wall Thickness Analysis for Heart Failure Detection},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {36924--36932},
doi = {10.1109/access.2026.3670331}
}
EUVIMED