AI measurements of heart function from MRI scans are more strongly linked to how long patients live after their scan than measurements made by clinicians.
These are new findings that suggest AI could provide a useful first-line approach to measuring heart function.
The research was led by City, University of London and University College London, and was published today in The Lancet Digital Health.
The research focused on ejection fraction, which helps surgeons decide whether patients may benefit from specialist heart failure treatments, implanted heart devices, and procedures to improve blood flow to the heart.
It is used to gauge surgical risk and to monitor heart function during chemotherapy.
In the largest study of its kind, researchers analysed heart MRI scans from 27,382 patients undergoing investigation for known or suspected heart conditions in the UK and US. Patients were followed for an average of more than five years, during which 3,458 deaths occurred.
The team compared ejection fraction measurements generated by an AI tool with those produced by experienced clinicians assessing cardiac MRI scans.
They found that AI-derived ejection fraction measurements were slightly but significantly better at predicting which patients were at increased risk of death than clinician-derived measurements. The findings remained significant after accounting for factors such as age and sex.
The AI tool was also substantially faster, analysing scans in approximately 2-3 seconds per patient. This compares with previous work by the team, which found expert clinician analysis typically takes around 13 minutes.
Dr George Joy, NIHR Clinical Lecturer from the Cardiovascular Clinical Academic Group at City St George’s, University of London and lead author, said: ‘Ejection fraction is one of the most important measurements we make in cardiology because it helps determine how patients are treated and how closely they should be monitored. 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 researchers stress that human oversight remains essential. While AI errors were uncommon, with an error rate of around 1% at the patient level affecting just 0.08% of images, they say clinicians should still play a critical role in verifying results and interpreting scans within the broader clinical context.
Dr Rhodri Davies, associate professor at University College London (UCL), senior author and inventor of the AI tool, 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 findings support using AI as a first-line tool for analysing cardiac MRI scans, with expert review where needed.
The team now plans to combine AI measurements from heart scans with those from other tests, such as ECGs, to detect problems in the heart muscle earlier, so treatment can begin before lasting damage occurs.
The research was funded by an NIHR Clinical Lectureship, the NIHR UCLH Biomedical Research Centre and a British Heart Foundation Accelerator Award.
