Performance of Artificial Intelligence-Powered ECG Analysis in Suspected ST-Segment Elevation Myocardial Infarction

Artificial intelligence (AI)-based electrocardiogram (ECG) analysis has emerged as a promising adjunct to human ECG interpretation in suspected ST-segment elevation myocardial infarction (STEMI).

To expand knowledge in this evolving field, the authors retrospectively analyzed the performance of a novel AI-ECG model in patients with cardiac catheterization laboratory activation for suspected STEMI.

Consecutive patients were gathered from a multicenter U.S. STEMI registry (2018-2022) and categorized into 3 clinical cohorts based on the presence or absence of angiographic culprit and troponin elevation: acute myocardial infarction (AMI) with culprit, AMI without culprit, and no-AMI. Cardiac catheterization laboratory-activating ECGs were analyzed using an AI-ECG model trained to identify acute coronary occlusion and classified as occlusion myocardial infarction, OMI(+) or not, OMI(-).

The study included 2,523 patients, 68.3% male, with a median age of 63 years. AMI with culprit was present in 2076 (82.3%), AMI without culprit in 314 (12.4%), and no-AMI in 133 (5.3%). Among AMI with culprit patients, the model correctly identified 93.8% as OMI(+). Sensitivity for TIMI flow 0/1, 2, and 3 was 96.3%, 93.1%, and 86.9% respectively; P < 0.001. The model correctly identified 79.7% of no-AMI patients as OMI(-). The AUCROC was 0.952 (95% CI: 0.924-0.966). The AMI without culprit cohort included takotsubo syndrome OMI(+) = 78%, MI with nonobstructive coronary arteries OMI(+) = 61%, and myopericarditis OMI(+) = 67%.

In suspected STEMI, this AI-ECG model correctly identified nearly all patients with acute coronary obstruction and most of those without AMI. If prospectively validated, this approach could improve management of patients with suspected AMI.