An AI ECG tool heart failure and valve disease detection system developed at Imperial College London identified heart failure in up to 81 per cent of cases and heart valve disease in up to 90 per cent of cases in a trial of 67,000 patients, running on a standard electrocardiogram in under two seconds.
The tool was developed with funding from the British Heart Foundation (BHF) and presented at the European Society of Cardiology annual congress in Munich. Its core purpose is straightforward: take a test that has been in clinical use for a century and make it do something it has never done before.
What ECGs can and cannot do
Roughly one billion ECGs are performed worldwide each year. They capture the heart’s electrical activity, rate and rhythm, and have long been the standard tool for spotting heart attacks and abnormal rhythms. What they cannot do, at least without this kind of processing, is detect heart failure or heart valve disease. For those conditions, patients need an echocardiogram, an ultrasound scan, and in most health systems they wait months for one after referral.
The AI model runs on existing ECG data, pulling information from signals that falls outside what the human eye typically picks up. Trained on millions of patient records, it identifies those most likely to have heart failure or valve disease so clinicians can prioritise them for echocardiograms rather than leaving them on a general waiting list.
The BHF has reported that for reduced heart pumping function specifically, the model achieved a score of 0.86 in a subset of 5,442 patients, and 0.9 in the larger group. Those figures matter because they show the model performing consistently across different cohort sizes, not just in optimal conditions.
The tool cannot definitively diagnose or rule out either condition. What it can do is give clinicians a strong enough signal to decide who needs faster follow-up. Professor Fu Siong Ng of Imperial College London pointed to a broader application beyond the waiting-list problem: running the model across all ECGs performed in a hospital, including those ordered for entirely unrelated reasons. ‘The AI model could be run on all ECGs done in a hospital to flag those at highest risk of these diseases, so that they can be diagnosed earlier,’ he said.
AI ECG tool heart failure detection and the road to deployment
Dr Sonya Babu-Narayan, consultant cardiologist and clinical director of the BHF, said the tool ‘could be a solution to help fast-track the patients who are most likely to have a heart abnormality. When it comes to the heart, earlier diagnosis and treatment saves and improves lives.’
Both conditions are treatable. Patients identified before their condition deteriorates have more options than those who receive a diagnosis after months on a waiting list. Dr Ahmed El-Medany, the BHF clinical research fellow who led the Imperial College analysis, said the next step is designing handheld AI-driven ECG readers for healthcare professionals to use in the field. That ambition points toward a future in which this kind of screening moves out of the hospital and into community settings, though that step has not yet been taken.
The commercial path is also taking shape. Imperial College London has reported that a spinout called Cardiovolt.AI has raised £1.4 million in funding. The spinout is built around this technology, and the fundraise suggests a concrete route from trial results to clinical deployment, though the timeline for that process has not been stated.
The scope of the underlying AI ECG tool extends further than heart disease alone. According to Imperial College London, diagnostic accuracy for non-cardiovascular conditions including diabetes and kidney disease reached 70 to 80 per cent from a single ten-second ECG. That finding is preliminary context rather than a clinically validated claim, but it indicates how much physiological information a standard ECG may carry that current practice does not yet use.
The immediate focus remains on heart failure and valve disease, where the case for faster diagnosis is already clear and the waiting-list problem is real. The Cardiovolt.AI funding round is the next concrete step on that path.
