A new artificial intelligence system developed by researchers at Imperial College London can analyse a routine ECG in less than two seconds and identify patients who may have hidden heart failure or valve disease.
The technology analyses standard electrocardiograms (ECGs) and searches for subtle electrical patterns associated with conditions including heart failure and heart valve disease.
In a US study involving around 67,000 patients, the AI identified 81% of people who had heart failure and 90 per cent of those with valve disease.
Across the wider research programme, the AI produced diagnostic accuracy ranging from 83 per cent to 93 per cent for heart disease in testing datasets.
AdvertisementAn ECG cannot by itself establish whether someone has heart failure or valve disease.
A new artificial intelligence system developed by researchers at Imperial College London can analyse a routine ECG in less than two seconds and identify patients who may have hidden heart failure or valve disease. Early results suggest it could help hospitals prioritise echocardiograms, although further testing is needed before wider clinical use.
A routine heart test that takes only seconds could reveal more about a patient's cardiovascular health than doctors can see on the tracing itself, according to researchers developing an artificial intelligence system at Imperial College London.
The technology analyses standard electrocardiograms (ECGs) and searches for subtle electrical patterns associated with conditions including heart failure and heart valve disease. The system can process a recording in less than two seconds, potentially giving clinicians a rapid way to decide which patients should undergo more detailed cardiac investigations.
The findings were presented at the European Society of Cardiology annual congress in Munich. In a US study involving around 67,000 patients, the AI identified 81% of people who had heart failure and 90 per cent of those with valve disease.
Advertisement
Millions of ECGs helped train the system
An ECG measures the heart's electrical activity, normally recording around 10 seconds of information. Doctors routinely use the test to assess heart rate and rhythm and to identify signs associated with problems such as heart attacks.
Imperial researchers believe the recordings contain additional information that is difficult to recognise through conventional visual assessment. Their models were trained using more than 1.6 million ECGs from Brazil, each linked with medical-history data, as well as several million recordings from the US.
More from Tech More from Tech
By comparing electrical patterns with subsequent diagnoses, the researchers trained the system to associate features in an ECG with diseases that would normally require other investigations to confirm.
Dr Ahmed El-Medany, who led the Imperial analysis, described the system as “superhuman”, referring to its ability to detect patterns that clinicians cannot consistently identify from an ECG alone.
Across the wider research programme, the AI produced diagnostic accuracy ranging from 83 per cent to 93 per cent for heart disease in testing datasets.
AI could help prioritise echocardiograms
The researchers see the technology primarily as a screening and prioritisation tool, rather than a replacement for conventional diagnosis.
Advertisement
An ECG cannot by itself establish whether someone has heart failure or valve disease. Those conditions generally require further clinical assessment and tests such as an echocardiogram, which uses ultrasound to examine the heart's structure and function.
That distinction could become important in health systems where demand for echocardiography exceeds capacity. Professor Fu Siong Ng of Imperial College London said some patients currently wait months for an echocardiogram after being referred.
An AI-generated risk assessment could allow hospitals to identify patients more likely to have significant disease and move them towards further testing sooner.
British Heart Foundation clinical director Dr Sonya Babu-Narayan said earlier identification could help patients reach treatment more quickly, while stressing that the technology would not detect every form of heart disease.
The system could also be used on ECGs already being performed for unrelated reasons. A patient having their heart rhythm checked, for example, could have the recording analysed automatically for signs of conditions that were not initially suspected.
From research system to clinical tool
Researchers are now exploring how the technology could be incorporated into everyday medical equipment, including a possible handheld ECG reader for healthcare professionals.
Advertisement
Imperial's research group has also established a spinout, Cardiovolt.ai, to support development towards clinical applications. The immediate goal is not to allow AI to make independent diagnoses, but to flag patients who may benefit from prompt specialist assessment.
The researchers acknowledge that substantial work remains before the system can become a routine clinical tool. It will need further validation across different populations and healthcare settings, as well as testing within real-world clinical workflows.
For now, the promise lies in extracting additional information from one of medicine's most familiar tests. If the results hold up in broader studies, a standard ECG could become an earlier warning system for heart disease, helping doctors decide who needs a closer look without requiring every patient to undergo the same pathway.
