What question did this study set out to answer?

The study aims to evaluate an AI model's capability to identify occluded myocardial infarctions using 12-lead ECGs.

March 26, 2026Open Access

Detecting Occlusion Myocardial Infarction with an AI-Powered ECG Model: A Retrospective Cohort Study

Key Result

An AI-powered ECG model identified occlusion myocardial infarction with 89.4% accuracy, 90.0% sensitivity, and 93.2% specificity, outperforming standard STEMI criteria.

Key Points

The study aims to evaluate an AI model's capability to identify occluded myocardial infarctions using 12-lead ECGs.
Conducted a retrospective cohort study analyzing 12-lead ECGs of patients suspected to have occlusions.
Verified myocardial infarction through angiographic evidence of coronary artery stenosis.
Examined AI performance metrics such as accuracy, sensitivity, and specificity.
AI model achieved an overall accuracy of 89.4% in identifying confirmed occlusions.
Sensitivity was 90.0% and specificity was 93.2%, indicating high diagnostic reliability.
Out of NSTEMIs, the model correctly identified 115 out of 142 confirmed occlusions.

Structured PICO

Does an AI-powered ECG model improve the detection of angiographically confirmed occlusion myocardial infarction in patients with suspected OMI compared to standard STEMI criteria?

Population

Patients with suspected occlusion myocardial infarction (OMI)

Intervention

AI-powered ECG model using a single 12-lead ECG

Comparator

Standard STEMI criteria

Outcome

Identification of total thrombotic coronary artery occlusion myocardial infarctions (OMIs) confirmed by angiographysurrogate

An AI-powered ECG model accurately detects angiographically confirmed occlusion myocardial infarction, outperforming standard STEMI criteria, particularly in patients traditionally classified as NSTEMI.

Main Result

Absolute Event Rate: 0% vs 0%

Abstract

Background: Patients with NSTEMI who are found with a totally occluded culprit vessel on coronary angiography are at higher risk of mortality and major adverse cardiac events. Artificial intelligence (AI) models can help identify this subgroup of NSTEMIs. Objectives: The purpose of this study was to examine the performance of an AI model in identifying patients with total thrombotic coronary artery occlusion myocardial infarctions (OMIs) using a single 12-lead ECG as input. Methods: In this retrospective cohort study, 12-lead ECGs corresponding to patients with suspected OMI were analyzed by an AI model. Confirmation of OMI was based on angiographic evidence of acute culprit coronary artery stenosis. Results: Over a one-year period, emergency physicians at our hospital identified 474 patients with suspected OMI, of whom 88 met STEMI criteria. Out of the 142 angiographically confirmed OMIs, the AI model correctly identified 115 (81%) with high confidence, corresponding to an accuracy of 89.4%, sensitivity of 90.0%, specificity of 93.2%, positive predictive value (PPV) of 84.6%, and negative predictive value (NPV) of 91.4%. Out of the 74 angiographically confirmed OMIs that did not meet STEMI criteria, the AI model correctly identified 49 (66%) with high confidence, corresponding to an accuracy of 87.9%, sensitivity of 66.2%, specificity of 93.4%, PPV of 72.1%, and NPV of 91.5%. Out of the 68 angiographically confirmed OMIs that met STEMI criteria, the AI model correctly identified 66 (97%) with high confidence, corresponding to an accuracy of 95.5%, sensitivity of 97.1%, specificity of 90.0%, PPV of 97.1%, and NPV of 90.0%. Conclusions: The AI model examined in this study outperformed the STEMI criteria for the identification of OMI with respect to accuracy, sensitivity, specificity, PPV, and NPV and accurately identified a significant portion of NSTEMIs found to have total thrombotic coronary artery occlusion.