TL;DR
An AI tool reading heart MRI scans predicted patients’ risk of death slightly but significantly better than experienced clinicians, in a UK-led study of 27,382 patients published in The Lancet Digital Health. The tool took 2-3 seconds a patient. Its authors, from UCL and City St George’s in London, say it could be a first-line tool, with clinicians still checking its results.
The study looked at ejection fraction, a measure of how well the heart pumps blood that guides decisions on heart failure treatment, implanted devices and surgical risk. Researchers compared AI and clinician readings against what then happened to patients in the UK and US, who were followed for over five years on average. There were 3,458 deaths in that period.
The AI’s advantage held after adjustment for age, sex and similar factors. According to the institution’s announcement, earlier work by the team put expert analysis at around 13 minutes a scan.
How much better is “better”
The size of the gain deserves attention, because the language around it is modest for good reason. Lead author Dr George Joy calls it “a small but meaningful improvement in identifying patients at higher risk”. Senior author Dr Rhodri Davies, who also invented the tool, argues that “even modest improvements in accuracy can have a significant impact when applied across thousands of patients.”
That interest should be kept in mind. The announcement we read does not quantify the improvement, and we have not reviewed the paper’s full results, which are in The Lancet Digital Health. The paper’s title describes it as retrospective and multicentre. In our view, that means it shows association with outcomes in past patients, not a benefit from using the tool in care.
The researchers put the AI error rate at around 1% of patients, affecting 0.08% of images, and they say people must still verify results and interpret scans in context.
Why UK health leaders should read it
Speed is the practical draw for a health system under pressure, and the authors say the tool could reduce workloads. The work was funded through the National Institute for Health and Care Research and the British Heart Foundation, not by a commercial vendor, according to the announcement.
Looking Forward
The team next plans to combine scan measurements with electrocardiograms to catch heart muscle problems earlier. In our view, the step that matters for NHS adoption comes first: a prospective study showing whether AI-first reading changes treatment decisions, workload and patient outcomes in practice. Until then, this is good evidence on measurement quality, and it stops short of evidence that patients do better as a result.