International evaluation of an artificial intelligence-powered electrocardiogram model detecting acute coronary occlusion myocardial infarction

Abstract

Aims: A majority of acute coronary syndromes (ACS) present without typical ST elevation. One-third of non-ST-elevation myocardial infarction (NSTEMI) patients have an acutely occluded culprit coronary artery [occlusion myocardial infarction (OMI)], leading to poor outcomes due to delayed identification and invasive management. In this study, we sought to develop a versatile artificial intelligence (AI) model detecting acute OMI on single-standard 12-lead electrocardiograms (ECGs) and compare its performance with existing state-of-the-art diagnostic criteria.

Methods and results: An AI model was developed using 18 616 ECGs from 10 543 patients with suspected ACS from an international database with clinically validated outcomes. The model was evaluated in an international cohort and compared with STEMI criteria and ECG experts in detecting OMI. The primary outcome of OMI was an acutely occluded or flow-limiting culprit artery requiring emergent revascularization. In the overall test set of 3254 ECGs from 2222 patients (age 62 ± 14 years, 67% males, 21.6% OMI), the AI model achieved an area under the curve of 0.938 [95% confidence interval (CI): 0.924-0.951] in identifying the primary OMI outcome, with superior performance [accuracy 90.9% (95% CI: 89.7-92.0), sensitivity 80.6% (95% CI: 76.8-84.0), and specificity 93.7 (95% CI: 92.6-94.8)] compared with STEMI criteria [accuracy 83.6% (95% CI: 82.1-85.1), sensitivity 32.5% (95% CI: 28.4-36.6), and specificity 97.7% (95% CI: 97.0-98.3)] and with similar performance compared with ECG experts [accuracy 90.8% (95% CI: 89.5-91.9), sensitivity 73.0% (95% CI: 68.7-77.0), and specificity 95.7% (95% CI: 94.7-96.6)].

Conclusion: The present novel ECG AI model demonstrates superior accuracy to detect acute OMI when compared with STEMI criteria. This suggests its potential to improve ACS triage, ensuring appropriate and timely referral for immediate revascularization.

Keywords: Acute coronary syndrome; Artificial intelligence; Electrocardiogram; Myocardial infarction; NSTEMI; Occlusion myocardial infarction.

Structured Graphical Abstract

Figure 1

A PRISMA flow chart showing data sources and study populations. Suspect acute coronary syndrome patients identified, exclusions (in grey), and the final study population split into a model development set (in green), EU internal test set (in blue), and US external test set (in red). ECG, electrocardiogram; ACS, acute coronary syndrome; pts, patients; CAG, coronary angiography; MI, myocardial infarction; OMI, occlusion myocardial infarction.

Figure 2

Artificial intelligence model performance on the overall testing data set. The receiver operating characteristic curve of the occlusion myocardial infarction artificial intelligence model (red) and the sensitivity and specificity of the occlusion myocardial infarction artificial intelligence model optimal threshold (red X), STEMI criteria (green dot), and electrocardiogram experts (purple cross) on combined EU and US testing cohorts. The AUC is 0.938 [n = 2263 contacts (21.61% occlusion myocardial infarction)]. OMI, occlusion myocardial infarction; AI, artificial intelligence; STEMI, ST-elevation myocardial infarction.

Figure 3

A subgroup analysis of the sensitivity and specificity of the occlusion myocardial infarction artificial intelligence model. The vertical dashed red line represents the overall artificial intelligence model sensitivity and specificity across all electrocardiograms in the testing data set. ECG, electrocardiogram; STEMI, ST-elevation myocardial infarction; AF, atrial fibrillation; VH, ventricular hypertrophy; LBBB, left bundle branch block; RBBB, right bundle branch block; LAD, left anterior descending artery; RCA, right coronary artery; LCx, left circumflex artery.

Figure 4

A real-world demonstration of an occlusion myocardial infarction artificial intelligence true-positive electrocardiogram downloaded from Twitter. (A) The original electrocardiogram posted to Twitter by Brooks Walsh, MD (https://twitter.com/BrooksWalsh, emergency physician at the Bridgeport Hospital, Bridgeport, CT, USA) with the occlusion myocardial infarction artificial intelligence model interpretation (above the optimal threshold); (B) the occlusion myocardial infarction artificial intelligence electrocardiogram model interpretation (above optimal threshold) with model explainability; (C) the angiogram of the occluded proximal left circumflex culprit artery and high-sensitivity troponin T evolution for this case.

Figure 5

A real-world demonstration of occlusion myocardial infarction artificial intelligence true-negative electrocardiogram downloaded from Twitter. (A) The original electrocardiogram posted to Twitter by Pendell Meyers, MD (https://twitter.com/PendellM, emergency physician at the Carolinas Medical Centre, Charlotte, NC, USA). Both the automated diagnostic statements and the attending physician misinterpreted this electrocardiogram, subsequently triggering a false-positive ST-elevation myocardial infarction cathlab activation; (B) the automatically digitized electrocardiogram with a very low occlusion myocardial infarction artificial intelligence model output (below the optimal threshold) and model explainability; (C) the echocardiography, catheterization, and laboratory report for this case.