A new study from Hospital Professor Doutor Fernando Fonseca in Portugal evaluates ChatGPT version 5.0's ability to detect ST-elevation myocardial infarction (STEMI) from ECG images.
The retrospective analysis examined 176 ECG images from actual STEMI activations. Researchers prompted ChatGPT with a standardized question asking whether an occluded coronary artery was present. The AI's predictions were then compared to results from invasive coronary angiography, the definitive diagnostic tool.
Overall, ChatGPT demonstrated 83% diagnostic accuracy. Its sensitivity was 82.9% and specificity 83.3%. The key finding was a high positive predictive value of 96.9%, meaning when ChatGPT indicated a blockage, it was almost always correct. This strong rule-in capability could provide practical reassurance for clinicians deciding whether to activate the catheterization lab.
However, the model's negative predictive value was low at 43.5%. This means a negative result from ChatGPT should not be used to rule out a coronary occlusion. The AI missed 26 confirmed occlusions.
The study concludes that while ChatGPT v5.0 shows promise as a supportive tool for confirming high-risk findings, it is not a substitute for clinical judgment or established diagnostic pathways. Its accessibility, compared to integrated AI platforms, warrants further investigation in similar clinical settings.