Artificial intelligence tool could identify heart valve disease years earlier: study

A novel artificial intelligence system that analyses heart sounds recorded by digital stethoscopes has demonstrated high accuracy in the early detection of severe heart valve disease, outperforming general practitioners and offering potential to transform screening in primary care

Artificial intelligence (AI) has shown potential to help clinicians detect serious heart valve disease years earlier than is currently typical, with implications for earlier treatment and improved survival, according to a study led by the University of Cambridge, UK.

Researchers have analysed heart sound recordings from 1,767 patients using a novel AI algorithm trained to recognise clinically significant valve disease, a condition that frequently remains undiagnosed until it reaches a life-threatening stage. Each participant also underwent echocardiography (ECG) which served as the reference standard against which the system’s performance was assessed.

The algorithm correctly identified 98 per cent of patients with severe aortic stenosis, the most common form of valvular heart disease that requires surgical intervention, and 94 per cent of those with severe mitral regurgitation, a condition in which the heart valve fails to close fully and allows blood to leak backwards across the valve.

When compared directly with clinical assessment by 14 general practitioners who listened to the same heart sound recordings, the AI system outperformed every clinician and did so consistently. Individual clinicians varied substantially in their diagnostic judgements, with some favouring sensitivity and others specificity, whereas the algorithm delivered stable and reliable results, particularly in cases of severe disease.

“Valve disease is a silent epidemic,” said Professor Anurag Agarwal, from the University of Cambridge’s Department of Engineering, who led the research.

“An estimated 300,000 people in the UK have severe aortic stenosis alone, and around a third do not know it. By the time symptoms appear, outcomes can be worse than for many cancers,” he said.

Valvular heart disease affects more than half of people aged over 65, with around one in ten experiencing clinically significant disease. In its early stages, the condition is often asymptomatic which contributes to delayed diagnosis.

“By the time advanced symptoms develop, the risk of death can be as high as 80 per cent within two years if untreated,” said Professor Rick Steeds, a co-author of the study from University Hospitals Birmingham.

“The only current treatment is surgery to repair or replace the valve,” he added.

At present, diagnosis relies primarily on ECG, which is regarded as the gold standard but remains costly and time-intensive. Waiting times within the UK’s National Health Service (NHS) can extend to many months which prevents its use as a population-level screening tool.

Although clinicians may listen to the heart with a stethoscope, this examination is not routinely performed during short general practice appointments and is known to miss a substantial proportion of cases.

“Cardiac auscultation is a difficult skill and it is used less and less in busy GP surgeries. That has contributed significantly to the large number of missed cases of valve disease,” said Agarwal.

The study represented a collaboration between engineers, cardiologists, research nurses and other clinicians across five NHS trusts. Digital stethoscopes were used to capture heart sounds which were then analysed by the algorithm. Rather than training the system to detect heart murmurs, the traditional clinical marker, the researchers trained it directly against echocardiogram findings. This approach allowed the algorithm to identify subtle acoustic patterns that are often imperceptible to the human ear, including disease in patients without an obvious murmur.

The system was deliberately designed to minimise false positive results, with the aim to avoid overwhelming already stretched ECG services. The researchers have emphasised that the technology is not intended to replace clinicians but to support clinical decision-making by identifying patients who would benefit most from further investigation.

Only a few seconds of heart sound recording are required and the test could be administered by healthcare staff with minimal training.

“If you can rule out people who definitely do not have significant disease you can focus resources on those who need them most,” said Agarwal.

The researchers have noted that further trials in real-world general practice settings, involving more diverse patient populations, will be required before the technology could see widespread deployment. They have also acknowledged that moderate forms of valve disease remain more challenging to detect than severe cases.

Despite these limitations, the study has suggested that AI could play a role in addressing increasing pressure on health services associated with an ageing population.

“Valve disease is treatable. We can repair or replace damaged valves and give people many more years of healthy life. But timing is everything.

“Simple, scalable screening tools like this could make a real difference by identifying patients before irreversible damage occurs,” said Steeds.

For further reading please visit: 10.1038/s44325-026-00103-y