Lit Matters #3: AI vs Docs: Who Wins at Reading STEMIs?

Occlusive myocardial infarction is often missed when the ECG shows a STEMI equivalent or a convincing mimic rather than classic ST elevation. On challenging ECGs such as de Winter, hyperacute T waves, and LBBB, a deep-learning tool outperformed both emergency physicians and cardiologists for cath lab activation decisions.

AI and STEMI Equivalent ECGs

• Overall accuracy gap: Physician accuracy on difficult cath lab activation ECGs was about 66%, while the Queen of Hearts AI reached 89%, a large separation that matters when OMI hides outside classic STEMI patterns.
• Patterns physicians miss: De Winter, hyperacute T-wave OMI, transient STEMI, and LBBB produced the lowest physician accuracy, with several of these patterns falling below 50% correct recognition.
• Emergency versus cardiology performance: Emergency physicians and cardiologists performed similarly overall, with emergency physicians trending more sensitive for STEMI equivalents while cardiologists were more specific for mimics.
• High-agreement ECG phenotypes: Classic STEMI, posterior wall OMI, pulmonary embolism, and Wellens were recognized much more reliably, with posterior OMI classified correctly by more than 80% of both groups.
• AI failure points: The algorithm was not flawless: LBBB with OMI and left ventricular aneurysm were its main misses, and those same tracings also caused substantial physician error. We get into why those two remain hard in the episode.
• Cath lab activation implications: Most clinicians treated Wellens as a cath-lab-worthy STEMI equivalent, while hyperacute T waves and RBBB generated the most specialty disagreement about immediate activation.