Hospitals in England that used AI to read stroke scans doubled the share of patients who had a brain clot removed

In the 14 months to February 2020, 2.3 percent of the stroke patients admitted to 26 hospitals in England had a blood clot pulled out of their brain. In the same 14 months three years later, the figure was 4.6 percent. Most strokes are ischaemic strokes, in which a clot blocks the blood supply to part of the brain, and when that clot sits in one of the brain's large arteries the standard treatment is a thrombectomy: a specialist threads a thin tube up from an artery in the thigh and draws the clot out. What changed at these hospitals in between was a piece of software. Each began running Brainomix 360 Stroke, an artificial intelligence tool that reads brain scans as they are taken and flags the signs of a major stroke within minutes. Across the rest of England's stroke hospitals, the rate rose over the same years, but by less, from 1.6 percent to 2.6 percent.

How we know

The figures come from a study published in The Lancet Digital Health in December 2025, by Kiruba Nagaratnam, Ain Neuhaus and colleagues. It used England's national stroke audit, a register of every patient aged 16 or over admitted to an NHS hospital with a stroke, and covered all 107 hospitals in England that admit acute stroke patients from 2019 to 2023: 452,952 patients in all. The 26 hospitals that adopted the software, six specialist centres and 20 local ones in four regional networks, were compared before and after it arrived, against the 81 that did not take part. It is an observational study, which means it followed what happened as hospitals adopted the tool rather than assigning patients to it by chance. Within the 26 hospitals, the researchers also compared patients whose scans were read with AI support against those whose scans were not: 5.9 percent of the first group had a thrombectomy, against 3.4 percent of the second. For the 747 patients who arrived at a local hospital and had to be moved to a specialist centre for the procedure, the median time between arriving and leaving was 128 minutes with AI support and 192 minutes without. Patients read with AI support were also about twice as likely to be given a clot-dissolving drug, 17.7 percent against 9.0 percent. NHS England, which funded the evaluation, published the same figures alongside the study.

Why it matters

Nearly three in four stroke patients in England first reach their nearest hospital that can give a clot-dissolving drug, most often a primary stroke centre: a local unit that cannot perform a thrombectomy and usually has no specialist in reading brain scans on hand. The patients who need the procedure must be recognised and sent on to a comprehensive stroke centre, the larger units that can do it, and every minute of that journey counts. That handover is where the study saw the strongest link between the software and treatment, and it is where the software also helps in a simpler way, by letting the local team and the specialist centre look at the same images at the same moment. The gap it works on is wide. In 2024 the national thrombectomy rate in England was 3.9 percent, against an estimated 15 percent of stroke patients who could be eligible. The NHS rolled AI scan reading out to every stroke centre in England in the summer of 2024. Whether the gains seen in 26 hospitals hold across all of them, and whether they show up as more people living independently months after a stroke, is the question the coming years of national data can answer.