TL;DR

Researchers have built an AI tool that detects heart failure and heart valve disease from a routine ECG, a test that has never been able to show either. Across 67,000 US patients it caught up to 81% of heart failure cases and up to 90% of valve disease. The work was funded by the British Heart Foundation and analysed at Imperial College London.

Why the choice of test is the point

The clever part is not the model, it is where it has been pointed. An electrocardiogram is cheap, fast and utterly routine — around a billion are done worldwide every year. Diagnosing heart disease currently needs an echocardiogram, an ultrasound scan that patients often wait months for.

So the tool does not replace a diagnosis. It reranks a queue. Someone flagged as high risk can be pushed to the front for a scan instead of waiting their turn, and Professor Fu Siong Ng of Imperial notes a second use: running it across every ECG a hospital performs would surface people whose heart condition nobody suspected, because they came in for something else entirely.

The caveats are the interesting part

Dr Sonya Babu-Narayan of the BHF is careful about what the tool does not do. It will miss people. It cannot confirm or exclude either condition on its own. What it offers is a fast-track for those most likely to have an abnormality — a triage instrument, not a diagnostic one.

That distinction determines how it would land in the NHS. A tool that sorts an existing waiting list needs a much lighter approval path than one that makes the call itself, and it does not create demand for scanning capacity that does not exist. Our reporting last week on AI scribes producing wrong diagnoses is the cautionary half of the same picture: the safe deployments are the ones where a clinician still decides.

Looking forward

The findings were presented at the European Society of Cardiology congress in Munich, which means conference-stage evidence rather than a published trial — the point at which enthusiasm usually outruns the data. The next step described is handheld AI-enabled ECG readers. Delegates also heard that five-second facial videos can be analysed to flag undiagnosed hypertension and type 2 diabetes, which suggests opportunistic screening, rather than better diagnostics, is where this field is heading.