← All research storiesHealth

Peer-reviewedHealth

An AI measured hearts on MRI scans in seconds, and its readings predicted survival slightly better than experts’

In more than 27,000 patients, an AI’s heart-pumping measurements from MRI scans predicted who would die over about five years a little better than doctors’ own measurements.

How far alongPeer-reviewedReviewed by independent experts for a journal before publication. Peer review is a quality check, not a grade of evidence strength: confidence still depends on study design, limitations, replication and the wider evidence.
How soon could it reach you?A few years outPromising, but needs more testing or engineering first. Our rough editorial guess is several years, with a lot of uncertainty.
Where it comes fromThe Lancet Digital HealthCity St George’s, University of London; University College London · Nov 2, 2026Read the original →

What the scientists did

Doctors often judge how well a heart is working with one number called the ejection fraction. To get it from a heart MRI scan, an expert traces the outline of the heart’s main pumping chamber, frame by frame. That can take around 13 minutes per patient, and two experts don’t always draw the same outline.

The research team compared an AI tool that does this tracing automatically with measurements made by experienced clinicians. They used scans from 27,382 patients in the UK and US who were being checked for known or suspected heart problems, then followed them for more than five years on average, during which 3,458 patients died. The question was simple: whose measurements did a better job of predicting who was at higher risk?

Words to know

Ejection fraction

The share of blood the heart’s main pumping chamber pushes out with each beat.

  • A healthy heart often pumps out a bit more than half of the blood in that chamber each beat
  • A much lower number can be a sign of heart failure

Segmentation

Outlining exactly where an object is in an image, pixel by pixel, so it can be measured.

  • Tracing the inside edge of the heart’s pumping chamber on an MRI
  • Outlining a tumor on a scan
  • Marking where a road is in a satellite photo

Retrospective study

A study that looks back at records and scans that were already collected, instead of testing something on new patients going forward.

  • Re-analysing old hospital scans
  • Checking past medical records to see what happened to patients later

What they found

The AI’s ejection-fraction measurements predicted the risk of death modestly but significantly better than the clinicians’ measurements, and the advantage held up after taking patients’ age and sex into account. The AI analysed each patient’s scan in about 2 to 3 seconds.

It was not perfect: the AI made mistakes for about 1% of patients, affecting 0.08% of all images, which is why the team says clinicians still need to check results.

27,382patients whose heart MRI scans were analysed
2–3 secondsAI time per patient, versus about 13 minutes for an expert
~1%of patients had an AI measurement error

Why it matters to you

If you or someone in your family ever needs a heart MRI, tools like this could mean faster results and more consistent measurements, no matter which hospital or which specialist reads the scan. It won’t replace the doctor, who still checks the work and decides what it means. If you have questions about your heart health, talk to a doctor.

Where this could lead

Faster heart scan reports

Hospitals could get key heart measurements in seconds, freeing specialists to focus on the hard cases.

3–10 years for wide use

Earlier warning of heart muscle problems

The team plans to combine AI MRI measurements with other tests such as ECGs to spot heart muscle disease sooner.

3–10 years

These are possibilities the research points toward, not promises. Most early findings take years of testing before they reach everyday life, and some never do.

Why scientists care

Medical AI is often judged by whether it matches human experts. This study asked a tougher question: do the AI’s numbers actually tell us more about what will happen to patients? Linking measurements to real-life outcomes like survival is a stronger test of whether a tool is worth using.

Keep in mind

This was a retrospective study of past scans, not a trial in which doctors used the AI in real time. The improvement was described as modest, and the news report did not give exact effect sizes. The AI still made occasional errors, so human checking remains important. We read a news report on the paper, not the full paper.

How far along is the science?

  • Peer-reviewedReviewed by independent experts for a journal before publication. Peer review is a quality check, not a grade of evidence strength: confidence still depends on study design, limitations, replication and the wider evidence.
  • PreprintA manuscript shared publicly, often before or alongside journal review. Check its separate journal and review status (each story says what we know). Results may change.
  • Conference talkPresented to other scientists at a meeting. Usually short and early; a full paper may come later.
  • Agency reportPublished by a science agency such as NASA, NOAA or NIH. Reviewed internally, but not by an outside journal.
  • Press releaseAn announcement from a university or company. Useful, but always check the research it describes.

How soon could it reach you?

  • Could reach you soonCould show up in products, services or advice soon. Our rough editorial guess is about 1 to 3 years, not a forecast from the research.
  • A few years outPromising, but needs more testing or engineering first. Our rough editorial guess is several years, with a lot of uncertainty.
  • Long-term scienceBasic research that builds knowledge. Real-world uses, if any, are far off.
  • Small stepA small update to earlier work. Useful to researchers, but not a change you would notice yet.

Every story links to its original source: the paper itself, or the conference program when no paper is out yet. Explanations are written with AI, usually from the paper’s abstract (each story says what it was written from), then checked against the source. We only use journals, recognized preprint servers, science agencies, scientific conferences and established science news outlets.