AI Heart MRI Readings Edge Out Clinicians at Predicting Survival

Every heart MRI scan holds a number that shapes a patient’s care: how well the heart pumps blood. A new peer-reviewed study in The Lancet Digital Health found that AI heart MRI measurements of this number were slightly but significantly better than clinicians’ measurements at predicting which patients were at higher risk of death. The AI was also dramatically faster, taking just seconds per scan.

It’s a modest gain, but one that could matter across thousands of patients. Here’s what the study found and what it doesn’t mean.

The number that guides heart care

The study focused on ejection fraction, which researchers call the most important measure of heart function. It describes how effectively the heart pumps blood with each beat.

Doctors lean on it heavily. According to the research team, ejection fraction helps decide:

• whether patients may benefit from specialist heart failure treatments
• who might need implanted heart devices
• who could benefit from procedures to improve blood flow to the heart
• how risky surgery may be
• how the heart is holding up during chemotherapy

Because it acts as a “gatekeeper” for so many treatments, measuring it accurately and consistently really counts.

How the AI heart MRI study worked

The research was led by City St George’s, University of London, and published in The Lancet Digital Health.

A very large group of patients

The team analysed heart MRI scans from 27,382 patients in the UK and United States who were being investigated for known or suspected heart conditions. The researchers describe it as the largest study of its kind.

Patients were followed for an average of more than five years. During that time, 3,458 deaths occurred.

AI versus experienced clinicians

For each patient, the researchers compared ejection fraction measured by an AI tool with the measurement made by experienced clinicians. Then they checked which set of numbers better predicted who went on to die during follow-up.

What they found

The AI’s measurements came out ahead, and the difference held up after accounting for factors such as age and sex.

In short: the AI was a little better at flagging higher-risk patients, and far faster at doing it.

Speed

The AI analysed scans in roughly 2 to 3 seconds per patient. By comparison, the team’s earlier work found that expert analysis typically takes around 13 minutes.

Consistency

The lead author described the AI’s measurements as remarkably consistent and precise. That matters because two people measuring the same scan can get slightly different answers, and small differences can push a patient over or under a treatment threshold.

Errors were rare, but not zero

AI errors were uncommon. The team reported an error rate of about 1% at the patient level, affecting just 0.08% of images.

What this does and doesn’t mean

It’s tempting to read this as “AI beats doctors,” but that’s not the takeaway.

Humans stay in the loop

The researchers stress that human oversight remains essential. Clinicians should still verify results and interpret scans in the context of each patient’s wider health. The senior author put it plainly: AI is not replacing clinicians.

The improvement is small

The authors themselves call the gain “slight” and “small but meaningful.” The value comes from scale: a modest improvement applied across many thousands of patients can add up.

Limitations to keep in mind

• It predicts risk; it doesn’t prove better outcomes. The study showed AI measurements were more strongly linked to survival. It did not test whether using the AI actually helps patients live longer.
• Senior author link. The senior author is also the inventor of the AI tool, which is worth knowing when weighing the results.
• Not a reason to change your own care. If you have a heart condition, decisions about your scans and treatment belong with your care team.

Why it’s still exciting

Health systems face growing demand, and heart MRI analysis takes skilled time. The researchers say their findings support using AI as a first-line tool for analysing cardiac MRI scans, with expert review where needed. That could ease workloads while keeping standards high and help patients get treatment sooner.

Next, the team plans to combine AI heart scan measurements with results from other tests, particularly electrocardiograms (ECGs), to spot heart muscle problems earlier, before lasting damage occurs.

The bottom line on AI heart MRI

This large, peer-reviewed study suggests AI heart MRI analysis can measure the heart’s pumping strength in seconds, with consistency that slightly improves how well doctors can identify high-risk patients, as long as clinicians stay in charge. It’s a promising, grounded step toward AI helping cardiologists read heart MRI scans.

Read the full announcement from City St George’s, University of London: AI measurements of heart scans more strongly linked to patient survival than clinician measurements.