AI-based ECG diagnosis of heart attacks
We trained an artificial intelligence (AI) model to diagnose large and small heart attacks on ECGs. The model became at least as good as doctors at distinguishing heart attacks from non-infarctions. Now we are moving on to train it to find different variants of heart attacks, heart failure, and other acute diseases.
Project description
The aim of this research is to use AI to automate the interpretation of ECGs for heart attack diagnosis. Doctors' accuracy in diagnosing heart attacks with ECGs is poor, and in the emergency room a large proportion, estimated at about 5% of all heart attacks, are missed. In addition to the large number of false negatives, false positives are common, with less than half of those hospitalised for a suspected heart attack having one. Many heart attacks are clear on ECG, but up to half lack clear ECG signs, at least those visible to the human eye.
AI-based heart attack diagnostics would have a huge value for emergency physicians' decision-making. Our first AI model performed very well, and we train the AI model to find other heart diseases, and also diseases outside the heart.