In a retrospective study of 33 diagnostically challenging admission ECGs, an AI app (PMCardio) outperformed 3 interventional cardiologists for detecting occlusion myocardial infarction requiring immediate angiography. Using angiography with ≥90% stenosis as the reference, AI reached 100% sensitivity and 100% negative predictive value, with a 24% misclassification rate vs 33%–42% for cardiologists.

Why It Matters To Your Practice

• OMI does not always present as classic STEMI and may appear as NSTEMI, ST depression, de Winter T waves, or hyperacute T waves.

• Missed OMI on the admission ECG can delay cath lab activation and reperfusion.

• In this small, high-difficulty set, AI was especially strong as a rule-out aid, with no false negatives against the angiographic standard.

Clinical Implications

• AI may help clinicians identify subtle coronary occlusion patterns on ECGs that are debated or nonpathognomonic.

• A high NPV could support escalation decisions when clinical concern remains high despite uncertain human interpretation.

• Specificity was 60%, so AI should augment rather than replace clinician judgment, symptoms, troponin data, and serial ECG review.

Insights

• Among 33 patients, 29 (88%) had a cardiac cause of chest pain; 23 had acute coronary syndromes, and 13 (57%) were OMI.

• Cardiologist sensitivity ranged from 62% to 85%; specificity from 50% to 70%; PPV from 47% to 57%; and NPV from 69% to 83%.

• The study focused on nonpathognomonic ECGs that had already generated diagnostic debate, making it a test of performance in difficult real-world edge cases.

The Bottom Line

• For challenging admission ECGs, AI improved detection of coronary occlusion and reduced misclassification compared with experienced cardiologists.

• The promise is faster recognition of patients who may need urgent invasive evaluation — but the evidence comes from a retrospective study with just 33 ECGs, so broader validation is still needed.