Deep learning on routine ECG spots hidden occlusion MI missed by STEMI criteria

A deep learning model analyzed 12-lead ECGs and identified occlusion myocardial infarction (OMI) with a recall (sensitivity) of 75% and specificity of 53% in patients classified as NSTEMI by conventional criteria, potentially flagging cases that need urgent angiography.

• Design: Retrospective study using 5 years of clinical and ECG data from acute MI patients, with coronary angiography as the reference standard for OMI vs. NOMI.
• Key secondary: The ResNet-1D model was explicitly tested on the NSTEMI subgroup where ST-elevation criteria fail, providing diagnostic support where clinical uncertainty is highest.
• Limitation: Single-center, retrospective study with modest specificity (53%) — external prospective validation is required before clinical use.

RadPigeon summaries are original and for information only. They are not clinical advice.