Human versus artificial intelligence–assisted ECG analysis for the detection of acute coronary occlusion in acute coronary syndrome

Abstract Prompt recognition of acute coronary artery occlusion (ACO) is critical for myocardial salvage and cath-lab activation in acute coronary syndrome (ACS) but ECG patterns interpretation remains challenging. Artificial intelligence (AI) has shown promise, but direct comparisons with first-line clinicians in routine practice are needed. To compare the diagnostic performance of an AI-based ECG algorithm detection of ACO in ACS with that of human readers. Methods This retrospective bicentric study included consecutive patients admitted for ACS and planned to invasive strategy. 12 lead-ECGs recorded prior to angiography were independently classified as suggestive of ACO or not by 32 clinicians (senior cardiologists, cardiology residents, senior emergency physicians, and emergency residents) blinded to all clinical and angiographic data. The AI algorithm provided a binary output for ACO detection. Coronary angiography reviewed by two experienced practitioners served as the reference to define ACO using two prespecified definitions: strict ACO (TIMI flow ≤1) and an extended definition (TIMI flow ≤2). Standard ESC STEMI criteria for ECG analysis, reviewed by three independent cardiologists were also used to define ACO. The primary endpoint was the sensitivity for strict ACO detection comparing AI and human readers. Secondary endpoints included accuracy, AI performance assessed by receiver operating curves, intra-observer agreement among clinicians and AI reproducibility using Cohen’s κ. Results Among the 194 patients included, the prevalence of strict ACO (TIMI ≤1) after coronary angiography was 39.7% (77/194) and 45.4% (88/194) for the extended definition (TIMI ≤2). AI reclassified 11 of 88 ACO (TIMI ≤2) patients (12.5%) not identified by physician reviewers. The left circumflex was the culprit vessel in 45%. AI demonstrated higher sensitivity than clinicians under both predefined definitions of ACO: 68.8% for AI versus 59.7% for clinicians (p = 0.035) for TIMI ≤1 and 68.2% vs. 58.0% for TIMI ≤2, (p=0.013) with similar specificity (78% vs. 79% and 82.1% vs. 81.1%, respectively). Intra-observer agreement was markedly higher for AI (κ = 0.89) than for clinicians (κ = 0.44). ESC ECG STEMI criteria were observed in only 22.2% of the patients (43/194) and demonstrated a lower sensitivity (37% across both definitions) than AI and human readers. Across clinician specialties, diagnostic accuracy was not significantly different and ranged from 66.5% for emergency residents to 69.1% for senior cardiologists, but remained lower than AI (75.8%). Conclusion AI-assisted ECG analysis significantly improves sensitivity for detecting acute coronary artery occlusion compared with human readers and standard ESC STEMI ECG criteria, without compromising specificity, and by reducing interpretative variability. These findings support the integration of AI as a robust decision-support tool to optimise triage and early management in acute coronary syndromes.visual abstract 1/2visual abstract 2/2