Uma IA que lê o eletrocardiograma e acha a artéria bloqueada: a história em inglês, no seu nível

Reading the subtle signs: InCor puts a heart attack AI to the test

When a coronary artery is completely blocked, the clock starts ticking. This is the most serious kind of heart attack, and spotting it on an electrocardiogram (ECG) is not always straightforward. A new AI model, put to the test at InCor, the Heart Institute of the University of São Paulo, aims to help doctors make the call and to point to the blocked artery.

The study, published in Nature Communications, is a collaboration with Uppsala University in Sweden. The model learned from 540,372 ECGs from Swedish emergency services; InCor then checked whether it held up elsewhere, validating it with 401 patients, all confirmed by catheterization.

The headline figure is 96% accuracy when the ECG shows the classic pattern. More telling is 82% in cases where that sign is not evident, which is where, according to co-author Marco Antonio Gutierrez, the main advance lies.

The stakes are high: doctors in ambulances must decide whether to give clot-dissolving drugs, which can save lives but can also trigger a hemorrhagic stroke. José Eduardo Krieger said every minute matters and described the AI as a high-precision classifier that gives doctors more confidence to act when the benefit outweighs the risk, not a replacement for clinical judgment.

Still experimental, the tool needs approval from Anvisa, and the team plans to test it at InCor in real-world conditions.

There is a lesson for language learners: real progress shows in the subtle cases. Anyone can follow a slow, carefully spoken sentence; fluency is tested when the signals are faint, like a quick reply or a strong accent. Practising with messy, real-world material, not just textbook examples, builds that kind of accuracy.