AI spotted stroke blockages on ordinary scans that doctors missed half the time
What happened
Researchers have tested a computer program that looks at ordinary brain scans and flags up one of the most time-critical emergencies in medicine. The condition is called a large vessel occlusion, or LVO. It is a stroke caused by a clot blocking one of the big arteries supplying the brain. These are the most severe strokes, and they are the ones where a procedure called a thrombectomy, physically pulling the clot out, can make an enormous difference. But only if it is spotted in time. The usual way to find an LVO is a special scan called a CT angiogram, which involves injecting a dye. Not every hospital can do one quickly, and not every patient can safely have the dye. The study, published in 2026 in the Journal of NeuroInterventional Surgery, tested whether artificial intelligence could find these blockages on a plain CT scan of the head, with no dye at all. That is the scan almost every suspected stroke patient gets straight away. The work was led by Dr Chi Kyung Kim of the Department of Neurology at Korea University Guro Hospital, with collaborators in the United States. It was reported in the medical press on 14 September 2026. The team tested the software on two separate groups of patients: 723 consecutive patients in Korea, of whom 127 had an LVO, and 240 patients in the United States, of whom 120 had one.
Why this matters
In stroke, the phrase doctors use is 'time is brain'. Roughly two million nerve cells die every minute that a major artery stays blocked. Every delay costs a patient some amount of movement, speech or independence they will not get back. The UK has a real problem here. Thrombectomy is one of the most effective treatments in modern medicine, but it is only available at a limited number of specialist centres, and patients have to be identified and transferred quickly enough for it to be worth doing. Audits have repeatedly found that only a minority of people who could benefit actually receive it. One of the bottlenecks is spotting who needs it. A patient arrives at a district general hospital at three in the morning. They get a plain CT scan, mainly to check they are not bleeding into the brain. Someone has to look at that scan and decide whether this is the kind of stroke that needs an urgent blue-light transfer to a specialist centre. That is exactly the decision this software is designed to help with, at exactly the moment when the most experienced person may not be in the building.
What the evidence actually says
The numbers in this study are worth looking at carefully, because they say something interesting about both the technology and the humans. The software performed well on its own. Researchers measure this with a score called the area under the curve, where 1.0 is perfect and 0.5 is no better than guessing. The program scored 0.963 on the Korean group and 0.899 on the American group. Testing on two different countries, with different scanners and different patients, is a meaningful check: it suggests the program had learned something real rather than memorising quirks of one hospital's equipment. The more striking figures come from the doctors. Working from plain CT scans without help, the clinicians scored 0.718. Their sensitivity, meaning the proportion of real blockages they correctly picked up, was 46.6 per cent. In other words, looking at a plain scan, they missed more than half. Their specificity, meaning how well they avoided false alarms, was 91.9 per cent. With the AI assisting, their overall score rose to 0.852. Sensitivity rose to 63.7 per cent, and specificity to 94.9 per cent. So they caught considerably more of the real blockages while also raising slightly fewer false alarms. Two honest caveats. First, 63.7 per cent is better, but it is still not most cases, and it is nowhere near what a proper CT angiogram achieves. This is a triage aid, not a replacement for the dye scan. Second, this was a validation and reader study, not a trial in a live emergency department. It shows doctors reading scans more accurately with help. It does not yet show patients getting to thrombectomy faster or doing better, which is the outcome that ultimately matters.
Practical advice
Nothing in this study changes what you should do if you suspect a stroke. But the basics are worth repeating, because they save lives now, without any technology at all. The UK guidance is the FAST test. Face: has their face fallen on one side? Can they smile? Arms: can they raise both arms and keep them there? Speech: is their speech slurred or muddled? Can they understand you? Time: if you see any single one of these signs, it is time to call 999 immediately. You do not need all three signs. One is enough. Do not wait to see if it passes. Do not drive them to hospital yourself if you can call an ambulance, because the ambulance service can alert the stroke team before you arrive and can take them to the right hospital rather than the nearest one. Other signs worth knowing: sudden weakness or numbness down one side, sudden loss or blurring of vision, sudden severe dizziness or trouble walking, and a sudden very severe headache unlike any before. If symptoms disappear within minutes, still get seen urgently. That may have been a transient ischaemic attack, sometimes called a mini-stroke, and it is a serious warning that a full stroke may follow within days. On prevention: high blood pressure is the single biggest risk factor for stroke, and it usually has no symptoms at all. If you have been prescribed tablets for blood pressure, atrial fibrillation or cholesterol, taking them consistently is one of the most powerful things you can do to reduce your risk.
What to know
A study published in the Journal of NeuroInterventional Surgery, reported on 14 September 2026, tested AI software that detects major stroke-causing blockages on plain CT scans, without the dye normally required. It was validated on 723 patients in Korea and 240 in the United States, scoring 0.963 and 0.899 respectively on a standard accuracy measure. Most strikingly, doctors reading plain scans alone caught only 46.6 per cent of real blockages. With AI support, that rose to 63.7 per cent, with slightly fewer false alarms as well. This is a triage aid, not a replacement for a proper CT angiogram, and it has not yet been shown to improve patient outcomes in a live emergency setting. What matters today is unchanged: learn the FAST test, and call 999 at the first sign. Sources: Journal of NeuroInterventional Surgery, 'AI assisted detection of large vessel occlusion on non-contrast CT: multinational validation and reader study', 2026, https://pubmed.ncbi.nlm.nih.gov/42173676/. Medical Xpress, 'Stroke diagnosis improved by AI support, multinational study finds', 14 September 2026, https://medicalxpress.com/news/2026-09-diagnosis-ai-multinational.html. NHS, 'Stroke symptoms', https://www.nhs.uk/conditions/stroke/symptoms/. This article is for general information and does not replace advice from a doctor, pharmacist or other qualified healthcare professional. If you think someone is having a stroke, call 999 immediately.
Related articles
Left alone together, AI systems invented their own dialect within days
When AI systems were put together in groups and left to get on with it, they started inventing their own way of talking. The findings come from Emergence, an A…
Scientists have worked out the weather on a failed star 20 light-years away
Astronomers at Trinity College Dublin have found a way to read the weather on distant worlds, and they have used it on one 20 light-years from Earth. The resea…
AI searched 7,000 existing medicines for a pneumonia fighter. Nine of them worked
Researchers at Imperial College London have used artificial intelligence to hunt through thousands of medicines that already exist, looking for ones that could…