The Pulse
UCL Study Finds AI Heart MRI Measures Better Track Survival
A study of 27,382 patients found that AI measurements of heart function from MRI scans were slightly but significantly better than clinician measurements at identifying patients at increased risk of death. The tool was faster, but the resea
AI.info Team · Written with AI · fact-checked ·
AI measurements edged clinician readings
On 3 November, University College London reported that AI measurements of heart function from MRI scans were more strongly linked to how long patients lived after their scans than measurements made by clinicians. The study compared AI-generated measurements of ejection fraction with assessments by experienced clinicians. Ejection fraction describes how effectively the heart pumps blood and is used to help guide treatment decisions.
The researchers analysed scans from 27,382 patients being investigated for known or suspected heart conditions in the UK and United States. Patients were followed for an average of more than five years, during which 3,458 deaths occurred. The AI-derived measurements were slightly but significantly better at predicting which patients were at increased risk of death. The difference remained significant after the researchers accounted for factors including age and sex.
The study evaluated how measurements related to later mortality; it did not test whether using AI to guide care improves survival. Its finding is about prediction, not evidence that the tool itself changes patient outcomes.
Ejection fraction helps guide treatment
Doctors use ejection fraction to help decide whether patients may benefit from specialist heart-failure treatments, implanted heart devices or procedures to improve blood flow to the heart. The measure is also used to gauge surgical risk and monitor heart function during chemotherapy. UCL described it as a “gatekeeper” for many treatments, making how it is measured relevant to decisions at several stages of care.
Senior author Dr Rhodri Davies, based at UCL’s Institute of Cardiovascular Science and the tool’s inventor, said:
“Even modest improvements in accuracy can have a significant impact when applied across thousands of patients.”
Davies also said the tool was not intended to replace clinicians, but could provide faster, more consistent measurements and support clinical decision-making.
Seconds for AI, minutes for experts
The AI analysed each scan in approximately two to three seconds per patient. UCL compared that with earlier work by the same team, which found that expert clinicians typically took around 13 minutes to analyse a scan. The 13-minute figure comes from that previous work, rather than a new side-by-side timing test reported in this study.
Speed does not remove the need for review. The researchers reported an error rate of around 1% at the patient level, affecting 0.08% of images. They stressed that clinicians should verify results and interpret scans in the wider clinical context.
First-line support, not a replacement
The researchers said the findings support using AI as a possible first-line tool for analysing cardiac MRI scans, with expert review where needed. UCL said the technology could help reduce workloads as healthcare systems face increasing demand, while maintaining standards of patient care and helping patients receive lifesaving treatments sooner.
Lead author Dr George Joy, based at City St George’s, University of London, said the improvement in identifying patients at higher risk was “small but meaningful.” He described ejection fraction as an important measurement in cardiology because it helps determine treatment and how closely patients should be monitored.
The team plans to combine AI measurements from heart scans with information from other tests, particularly electrocardiograms, or ECGs, to look for heart-muscle problems earlier. The researchers’ stated aim is to identify problems in time for treatment before lasting damage occurs; that is a planned next step, not a result demonstrated by this study.