TL;DR
A team at the University of Oxford has built an AI tool that predicts a patient’s risk of developing heart failure up to five years in advance, reading routine cardiac CT scans for signs of inflamed fat around the heart that are invisible to the human eye. Validated across 72,000 patients at nine NHS trusts over a decade of follow-up, the model reached 86% accuracy and is now seeking regulatory approval for NHS deployment.
How it works
Led by Professor Charalambos Antoniades, the Oxford team trained the model to spot subtle imaging biomarkers in pericoronary adipose tissue — the fat immediately surrounding the coronary arteries — that correlate with future cardiac decline. Patients scored in the highest risk group were 20 times more likely to develop heart failure than those in the lowest, with roughly a one-in-four chance of onset within five years. The results were published in the Journal of the American College of Cardiology.
Crucially, the tool produces an absolute risk score from existing scan data without additional tests or human input, and Antoniades says the team is working to extend it to any chest CT regardless of the reason for the scan.
Why this matters for the NHS
Heart failure affects around 60 million people worldwide and is routinely diagnosed late — often only once a patient presents acutely at hospital, by which point irreversible muscle damage has already occurred. Dr Sonya Babu-Narayan, clinical director at the British Heart Foundation, which funded the work, said late diagnosis means patients “already have severe damage to their heart muscle which might have been avoided”. Earlier stratification would let GPs monitor high-risk patients more closely and start cardioprotective treatment sooner.
The research also lands the same week that NIHR committed £1.5 million to AI diagnostics work at Leeds Teaching Hospitals and the MHRA expanded its AI Airlock sandbox with £3.6 million of new funding — a pattern suggesting UK medical AI is moving from pilot to pathway funding.
Looking forward
Regulatory approval is the gating item. If the tool is added to routine cardiac CT analysis in NHS radiology departments, it would be one of the first truly prospective AI diagnostic aids at national scale in the UK.