Machine learning models demonstrated good diagnostic performance for detecting regional wall motion abnormalities on echocardiography, with a pooled C-statistic of 0.88 (95% CI 0.81-0.95).
Meta-Analysis (n=36,000)
Do machine-learning algorithms accurately detect regional wall motion abnormalities on transthoracic echocardiography?
Machine learning models show good diagnostic performance for detecting regional wall motion abnormalities on echocardiography, though clinical readiness is currently limited by heterogeneity and lack of external validation.
Effect estimate: Pooled C-statistic 0.88 (95% CI 0.81-0.95)
BACKGROUND: Regional wall motion abnormality (RWMA) assessment is fundamental in transthoracic echocardiography (TTE) for diagnosing ischaemic heart disease, yet visual interpretation is subjective and variable. Machine-learning (ML) models may offer a more objective and reproducible RWMA evaluation, but their diagnostic accuracy has not been comprehensively synthesized. OBJECTIVE: To systematically evaluate the diagnostic performance of ML algorithms for detecting RWMA on TTE. METHODS: PubMed (MEDLINE) and EMBASE were searched from inception to 7 December 2025 for studies applying ML to RWMA detection using two-dimensional TTE. The primary outcomes were C-statistics, sensitivity and specificity of ML models. RESULTS: Eight studies comprising ≈36,000 echocardiographic examinations were included. RWMA prevalence ranged from 9 % to 75 %, with ground truth definitions primarily based on expert consensus. Reported C-statistics ranged between 0.67-0.99, reflecting substantial heterogeneity (I² = 98 %). The pooled C-statistic was 0.88 (95 % CI 0.81-0.95). Internally validated models demonstrated a pooled C-statistic of 0.90 (95 % CI 0.82-0.97), and externally validated models 0.88 (95 % CI 0.84-0.92). Across studies reporting diagnostic data, pooled sensitivity and specificity were 0.83 (95 % CI 0.64-0.93) and 0.84 (95 % CI 0.75-0.91) respectively. CONCLUSIONS: ML models demonstrate good diagnostic performance for RWMA detection on TTE, approaching the accuracy of human readers in existing studies. However, marked heterogeneity, limited external validation, and methodological limitations currently restrict clinical readiness. Future research should prioritize multicentre external validation, improved reference standards, and adherence to TRIPOD-AI framework to support safe clinical integration.
Koh et al. (2026) conducted a meta-analysis in Regional wall motion abnormalities (n=36,000). Machine learning algorithms vs. Expert consensus (ground truth) was evaluated on C-statistics, sensitivity and specificity of ML models (Pooled C-statistic 0.88, 95% CI 0.81-0.95). Machine learning models demonstrated good diagnostic performance for detecting regional wall motion abnormalities on echocardiography, with a pooled C-statistic of 0.88 (95% CI 0.81-0.95).