Early Identification of ST-Segment Elevation Myocardial Infarction Equivalents on Nondiagnostic ECGs Using an Artificial Intelligence Model

Overview

This retrospective, multicenter study from four US percutaneous coronary intervention (PCI) centers tested whether an AI-ECG model, PMcardio Queen of Hearts (aOMI v1), could recognize STEMI equivalents on initial ECGs that clinicians had interpreted as nondiagnostic, in patients whose STEMI was diagnosed only on a later ECG. Of 2,588 STEMI activations over three years, 440 (17%) had a nondiagnostic initial ECG, and the mean delay to the diagnostic STEMI ECG was 80 minutes. The AI model identified 347 of these 440 patients (79%) as STEMI equivalent on the initial ECG. It also identified the culprit artery in 76% of cases and predicted the leads that later developed diagnostic ST elevation in 95%.

Published in: SAEM26 – Society for Academic Emergency Medicine Annual Meeting 2026 – presented in an abstract session
Published on: 21 May 2026

Background

Rapid recognition of ST-elevation myocardial infarction (STEMI) is essential for optimal patient outcomes. Prior studies have shown the potential of an artificial intelligence (AI) ECG model to reduce false-positive STEMI activations, but its impact in patients with confirmed occlusion myocardial infarction is less clear. This study evaluated the AI model's ability to identify STEMI equivalents on initial ECGs interpreted as nondiagnostic, in patients subsequently diagnosed with STEMI on repeat ECGs.

Methods

This retrospective, multicenter study included patients from four PCI centers (Atrium Health and Albany Medical Center) over three years. All patients whose initial ECG was interpreted as nondiagnostic for STEMI, who had a subsequent ECG diagnostic of STEMI and who had a cardiologist-confirmed, catheterization-proven STEMI diagnosis were included. The initial and subsequent ECGs were analyzed retrospectively with the AI model (PMcardio Queen of Hearts, aOMI v1). Descriptive statistics were used for data analysis.

Results

Over the three-year period there were 2,588 STEMI activations, of which 440 (17%) had an initial ECG interpreted as nondiagnostic. Of these 440 patients, 70% were male and the mean age was 62±13 years; 51% arrived as walk-ins and 49% by ambulance, and 45% presented directly to a PCI center while 55% were transferred. Culprit arteries were the left anterior descending artery (44%), right coronary artery (32%), left circumflex artery (19%) and others (5%). Thirty-three patients died within two months. The mean delay between the initial nondiagnostic ECG and the diagnostic STEMI ECG was 80 minutes. The AI model correctly identified 347 of 440 patients (79%) as STEMI equivalent on the initial "nondiagnostic" ECG; in this group, the mean delay to the diagnostic STEMI ECG was 76 minutes and 31 deaths occurred. The model correctly identified the culprit artery in 76% of cases and predicted the ECG leads that later developed diagnostic ST elevation in 95% of cases.

Conclusion

The AI-ECG model identified STEMI equivalents in 79% of patients whose STEMI was diagnosed only on a subsequent ECG, enabling detection approximately 80 minutes earlier on ECGs that had initially been interpreted as nondiagnostic by clinicians.