Background :
Valvular heart diseases (VHD) contribute significantly to mortality, yet early detection remains
challenging. Artificial intelligence (AI) applied to ECG has emerged as a promising screening
tool.
Study aim:This study evaluates the diagnostic accuracy of AI-based ECG algorithms for
detecting VHD. Secondary objectives include assessing the effect of disease severity on
diagnostic performance, comparing between Deep Learning (DL) and Traditional Machine
Learning (ML) models, and evaluating 12-lead modality versus single-lead ECG modality.
Methods:We retrieved 1030 cardiology-related records from PubMed, Scopus, and Web of
Science, and selected 19 studies focusing on VHD for full-text review. After applying exclusion
criteria and adding four studies from reference mining, 17 studies were reviewed, but only 11
studies have been entered in the quantitative analysis. The primary outcome is the pooled Area
Under the Curve (AUC).
Results: The pooled AUC was 0.85 for Aortic Stenosis and Mitral Regurgitation, and 0.80 for
Aortic Regurgitation. Severe AR was detected with higher accuracy (AUC 0.88) than moderateto-severe AR (AUC 0.78). DL models outperformed ML models (AUC 0.87 vs 0.72), and 12-
lead ECGs performed better than single-lead recordings.
Conclusion : AI-enabled ECG can reliably screen for significant VHD, particularly severe
cases. Integrating these algorithms into primary care may allow earlier referrals and improve
patient outcomes |
