Human, Take this Patient to the Cath Lab – AI and STEMI Detection
September 13, 2024
We have one wish…
Would you write a quick testimonial? A nice word from you is all we need 🥹!
If you have more time this weekend…
Do you have suggestions for improvement? Give us your feedback about JournalFeed!
We need your input on wisely and ethically using AI. Also, see if you can you tell what’s AI vs human-written.
Many thanks,
Clay
Written by Gracey Mcgrory and Ketan Patel
Spoon Feed
These researchers developed and trained a deep ensemble artificial intelligence (AI) model to classify ECGs as STEMI versus non-STEMI. The AI performed well in both accuracy and in improving sensitivity.
Alexa, hold my calipers
Treatment of myocardial infarction with angioplasty requires accurate identification of STEMI on ECG. This study aimed to determine if AI can improve the identification of STEMIs while limiting unnecessary angioplasty and potential patient harm by intervention.
The AI model was trained on a data set from a single center in Korea with 15,641 ECGs verified as STEMI vs. non-STEMI via two cardiologists with access to corresponding angiogram results. After training on this data set, the model was set against three resident physicians and out-performed them (accuracy 92.1% vs. 79.6%; sensitivity 95.4% vs. 81.1%; specificity 91.8% vs. 77.4%). The AI was then internally validated and externally validated. Comparisons to current ECG computer algorithms revealed decreased specificity of the AI model (89.1% vs. 98.3%) but improved sensitivity (95% vs. 60%).
There are several glaring limitations to application of this study to the real world. ECGs were from a relatively homogenous population at a single center. STEMI-equivalents were also not included in analysis. When the AI was challenged against physicians, it was pitted against three second year internal medicine residents rather than senior emergency medicine physicians.
How will this change my practice?
Even if AI performed perfectly, I would still be hesitant to lean on it for ECG interpretation. For one, it does not provide reasoning for its interpretation – it’s simply trained on ECGs and then spits out “STEMI” or “non-STEMI.” More elucidation of AI reasoning in the future would make me more amenable to incorporating AI into my practice.
Despite these limitations, I think AI could soon be a helpful adjunct to help facilitate getting patients to the catheterization lab faster and with greater accuracy.
Source
Development of Clinically Validated Artificial Intelligence Model for Detecting ST-segment Elevation Myocardial Infarction. Ann Emerg Med. 2024 Jul 25:S0196-0644(24)00327-5. doi: 10.1016/j.annemergmed.2024.06.004. Epub ahead of print. PMID: 39066765.