When a stroke hits, it’s often a race against time. Blood clots block oxygen from reaching the brain, and every delay can cost brain cells or even be life-threatening. But for millions of people, the real danger doesn’t end after the first stroke. It’s what caused it that often goes unnoticed — especially if the culprit is a stealthy heart condition that leaves no trace on a standard ECG.

This condition is atrial fibrillation, or AF. It’s the most common type of irregular heartbeat, and it can lurk undetected for years. Many people with AF only find out they have it after suffering a stroke. And because AF-related strokes require a very different treatment strategy than other kinds, a missed diagnosis can mean a missed chance to prevent the next, potentially fatal, stroke.

Now, researchers in Melbourne may have found a surprising way to uncover this hidden heart risk: by looking not at the heart, but at the brain.

A silent problem

AF increases the risk of stroke by up to five times. But its symptoms (if there even are any) can be fleeting.

“Early detection of atrial fibrillation (AF) is important to offer patients the best chance of preventing a serious cardioembolic stroke. However, many patients first present with an acute ischemic stroke for which the underlying cause of AF is silent because it is asymptomatic and intermittent,” says Craig Anderson, Editor-in-Chief of the journal Cerebrovascular Diseases, where the study was published.

When AF goes unnoticed, it can be deadly. Doctors may wrongly assume that the stroke was caused by a blocked artery, as we see in a different type of stroke known as large artery atherosclerosis. And the distinction matters quite a lot. Patients with AF need blood thinners to prevent new clots. Those with artery disease may need surgery or different medications. An incorrect diagnosis means the wrong treatment which can be life threatening.

But how do you find the right diagnosis?

Researchers suspected that every stroke leaves a distinctive footprint: tiny scars that appear in the brain. Their shape and distribution can offer clues about the stroke’s origin. Cardioembolic strokes — the type caused by AF — often leave behind scattered damage across multiple brain regions. In contrast, artery-block strokes tend to follow a more localized pattern. Neurologists have used these patterns for years to inform diagnoses. But there are limits to what the human eye can see on a brain scan. There are also limits to how much time trained neurologists have to look for these.

This is where AI comes in.

The team at the Melbourne Brain Center and the University of Melbourne turned to 3D convolutional neural networks — an advanced form of machine learning that excels at analyzing complex visual data. They fed the algorithm, called ConvNeXt, a set of MRI brain scans from more than 230 patients who had already suffered strokes. Some had AF. Others had strokes caused by large artery disease.

The AI didn’t know anything about the patients’ heart histories. It just looked at the brain.

The algorithm is pretty good

Their AI model was able to differentiate between AF-related strokes and non-AF strokes with strong accuracy. The researchers used a measure called AUC (area under the curve) to evaluate how accurately their AI could tell whether a stroke was caused by atrial fibrillation (AF) or another issue like a blocked artery. AUC scores range from 0.5 (no better than guessing) to 1.0 (perfect accuracy). In its best tests, the AI scored 0.88, and overall it maintained a strong average of 0.81 — indicating it could reliably spot subtle patterns in brain scans linked to AF, even when those patterns might be too faint or complex for human doctors to detect by eye.

This isn’t perfect, but it’s a potentially game-changing advance.

It also builds on a growing body of work exploring how artificial intelligence might support diagnosis and stroke care.

“Machine learning is gaining greater traction for clinical decision-making and may help facilitate the detection of undiagnosed AF when applied to magnetic resonance imaging,” the study notes. MRIs are already already a routine part of stroke care, and this method doesn’t require extra scans or procedures for patients. This makes it a low-cost, non-invasive way to support more targeted care.

The study is still early-stage. The researchers emphasize that their model is a proof-of-concept, not a finished diagnostic tool. More work needs to be done on larger and more diverse patient populations. And while the current model looks only at MRI data, future versions may include age, blood markers, and genetic factors to boost accuracy even further.

Still, the promise is clear. With more refinement and validation, tools like this could offer a new level of personalized care. They could help doctors see past what the ECG misses. They could give patients answers — and options — sooner.

Journal Reference: Angelos Sharobeam et al, Detecting atrial fibrillation by artificial intelligence enabled neuroimaging examination, Cerebrovascular Diseases (2025). DOI: 10.1159/000543042