Diagnostic Accuracy of Artificial Intelligence for Detecting ST-Elevation Myocardial Infarction on 12-Lead Electrocardiograms: A Systematic Review and Meta-Analysis

Artificial intelligence (AI) may improve electrocardiographic detection of ST-elevation myocardial infarction (STEMI), but available studies address distinct diagnostic questions. We systematically reviewed studies evaluating AI-based STEMI detection from 12-lead electrocardiograms and separately synthesized AI performance against a clinical reference standard, AI and clinician performance against a clinical reference, and AI agreement with physician electrocardiogram (ECG) interpretation. PubMed, Embase, Web of Science, Scopus, Cochrane Library, and ClinicalTrials.gov were searched from January 2010 through March 10, 2025. Bivariate random-effects models generated pooled sensitivity, specificity, and hierarchical summary receiver operating characteristic area under the curve (AUC). Twenty-six studies comprising 539,876 ECGs were included. In 20 studies comparing AI with a clinical reference standard, pooled sensitivity was 0.95 (95% confidence interval [CI], 0.91-0.97), specificity 0.95 (0.90-0.98), AUC 0.98 (0.97-0.99), positive likelihood ratio 19.2 (9.3-39.5), and negative likelihood ratio 0.05 (0.03-0.10). In 8 studies reporting both AI and clinician performance against a clinical reference, separate syntheses showed sensitivity of 0.91 (0.86-0.94) for AI versus 0.78 (0.69-0.85) for clinicians, specificity of 0.91 (0.82-0.96) for both, and AUCs of 0.95 and 0.90, respectively. In 7 studies using physician ECG interpretation as the comparator, AI sensitivity was 0.85 (0.70-0.94), specificity 0.99 (0.96-1.00), and AUC 0.98 (0.96-0.99). Sensitivity analyses excluding case-control designs, influential studies, and restricting analyses to deep-learning models or raw-waveform ECGs produced similar findings.