An artificial intelligence tool that measures how well the heart pumps predicted patients’ risk of death slightly better than measurements made by clinicians, in a peer-reviewed study of 27,382 people’s cardiac MRI scans. The AI heart MRI work, led by researchers at UCL and City St George’s, University of London, was announced by UCL on 3 November and published in The Lancet Digital Health.
The finding is less about a dramatic leap than about consistency. The study did not test the tool on new patients in a trial; it went back over existing scans and followed what happened to the patients. The AI’s edge was small, but it came at a fraction of the time and cost of expert analysis, and that is where the researchers see the value.
In brief
- In scans from 27,382 UK and US patients, AI-measured ejection fraction predicted death risk “slightly but significantly better” than clinician-measured values.
- The AI took about 2-3 seconds per patient, against about 13 minutes for expert analysis in the team’s earlier work.
- This is an analysis of existing scans, not a trial. The senior author invented the tool, and the researchers say human oversight remains essential.
How the AI heart MRI study was done
The researchers used cardiac MRI scans from 27,382 patients in the UK and US who were being investigated for known or suspected heart conditions. Patients were followed for more than five years on average, during which 3,458 died.
The study focused on ejection fraction, the share of blood the heart’s main pumping chamber pushes out with each beat. According to UCL, the AI-derived ejection fraction was “slightly but significantly better” at predicting death risk than clinician-derived values, and the difference held after accounting for age and sex. UCL calls it the largest study of its kind.
The team reports that the AI made errors in about 1% of patients, affecting 0.08% of images. In its earlier work, the AI took about 2-3 seconds per patient, compared with about 13 minutes for expert analysis.
Dr George Joy of City St George’s, the lead author, said: “Our findings suggest AI can measure this important marker with remarkable consistency and precision, resulting in a small but meaningful improvement in identifying patients at higher risk.”
The senior author built the tool
Readers should know that Dr Rhodri Davies of UCL, the study’s senior author, invented the AI tool being tested. That does not undermine a peer-reviewed result, but it is an interest worth stating, and independent replication would strengthen the case. The release does not name the tool.
Davies said: “Even modest improvements in accuracy can have a significant impact when applied across thousands of patients. AI is not replacing clinicians, but it can provide faster, more consistent measurements at low-cost and support more streamlined clinical decision-making that ultimately benefits patients.”
The work was funded by an NIHR Clinical Lectureship, the NIHR UCLH Biomedical Research Centre and a British Heart Foundation Accelerator Award. The next step, the team says, is to combine the measurements with ECG data.
Consistency, not brilliance, is the gain
The honest reading of this study is that the AI is a steadier measurer, not a better cardiologist. The lead author’s own emphasis is on consistency and precision. A tool that gives the same answer every time removes some of that noise, and less noise means a slightly sharper signal about who is at risk. That fits the small size of the improvement.
The time saving may matter more in practice than the accuracy gain. Thirteen minutes of specialist time per scan against a few seconds is the kind of difference that changes how many scans a department can report, which links to TSN’s coverage of the NHS adding scanners and AI MRI software to raise capacity. It also echoes the FDA clearance of TumorSight, another tool that turns MRI images into risk scores.
The limits are clear, though. An analysis of past scans shows the measurements are associated with outcomes; it does not show that using them would change treatment or save lives. That needs a prospective trial. With a 1% patient-level error rate, the researchers’ insistence on human oversight is not a formality: someone has to catch the cases the tool gets wrong.
What comes next for AI heart MRI
For UK patients, nothing changes yet; this is research, not a service. The questions to watch are whether the tool is tested prospectively, whether adding ECG data improves prediction further and whether NHS centres adopt it with expert checking in place. More stories like this one appear in TSN’s Tech for Good weekly round-up.
Sources
- UCL News: AI heart scan measurements better predict survival (3 November 2026)
- The Lancet Digital Health: the study paper (2026)
Related stories
- NHS Credits New Scanners and AI MRI Software for a Million More Scans
- FDA Clears MRI-Based AI That Scores 10-Year Breast Cancer Return Risk
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X
