What is it about?
Some heart attacks (myocardial infarctions) are caused by an artery that becomes completely blocked, cutting off blood flow to the heart. These are especially dangerous and need emergency treatment, but they are sometimes missed or delayed because the standard heart tracing (ECG) doesn't always show the classic warning pattern doctors look for, this is true for roughly a third of these blockages. We trained an artificial intelligence model to read standard 12-lead ECGs from over 11 000 patients and learn to recognize these dangerous blockages, including the harder-to-spot cases. The AI correctly identified the large majority of true emergencies, and in many cases flagged them faster than they were recognized in routine hospital care, particularly for the subtler cases that clinicians most often miss. These results suggest AI-assisted ECG reading could help doctors get patients with hidden, high-risk heart attacks to lifesaving treatment sooner.
Featured Image
Photo by Sasun Bughdaryan on Unsplash
Why is it important?
Heart attacks caused by a fully blocked artery are a race against time, the longer the artery stays blocked, the more heart muscle dies. Right now, doctors rely heavily on ECG patterns that only reliably catch about two-thirds of these blockages. The rest slip through because they don't show the "textbook" signs, so those patients wait longer for the emergency procedure that could save their heart muscle (and sometimes their life). This study matters because it shows that an AI reading the same ECG doctors already collect can catch many of these hidden, high-risk blockages, including a large share of the ones that are currently missed or delayed, without requiring any new equipment or tests. If validated further, this kind of tool could help hospitals and ambulance crews flag high-risk patients earlier, shortening the time to treatment for exactly the patients who are most likely to be overlooked today.
Perspectives
This project was, for me, fundamentally a labeling problem. Figuring out how to define the ground truth well enough, by combining registry data, ICD-10 codes, troponin levels, and blinded TIMI reassessment such that a model trained on it could be trusted, with the long-term goal of real clinical deployment rather than just a benchmark exercise. One of the things that struck me most was how consistently the model's confidence tracked with independent markers of disease severity we never trained it on, like troponin and coronary flow, it reassured me that we were capturing something clinically real rather than a statistical artifact. Looking ahead, I would consider this project a success only once it moves beyond a paper and into a prospective trial, a randomized clinical trial and ultimately into everyday clinical practice.
Bjørn-Jostein Singstad
Akershus University Hospital
Read the Original
This page is a summary of: Artificial intelligence-enabled electrocardiographic detection of acute occlusive myocardial infarction, August 2026, Springer Science + Business Media,
DOI: 10.21203/rs.3.rs-10574825/v1.
You can read the full text:
Contributors
The following have contributed to this page
