What happened
University of Michigan researchers unveiled Prima – an AI system that reads brain MRI scans in seconds and determines which patients need immediate care. The model achieved 97.5% diagnostic accuracy across more than 50 neurological conditions, outperforming other advanced AI tools.
Published in Nature Biomedical Engineering on February 10, 2026, the research represents a first-of-its-kind approach to medical imaging AI.
“As the global demand for MRI rises and places significant strain on our physicians and health systems, our AI model has potential to reduce burden by improving diagnosis and treatment with fast, accurate information,” said senior author Todd Hollon, M.D., a neurosurgeon at University of Michigan Health.
Why this matters
The numbers tell the story: millions of MRI scans performed worldwide each year, growing demand, shrinking neuroradiology workforce. Depending on location, patients currently wait days for results. In emergencies – strokes, brain hemorrhages – every minute counts.
Prima changes that equation. The system automatically flags life-threatening conditions and alerts the appropriate specialist – stroke neurologist, neurosurgeon – immediately after imaging completes.
The model was tested over one year using more than 30,000 MRI studies. Unlike previous AI tools designed for narrow tasks (detecting specific lesions, estimating dementia risk), Prima handles the full diagnostic spectrum a radiologist faces daily.
How it works
Prima is a vision language model (VLM) – AI that processes images and text together in real time. Think of it as giving the system both the scan and the patient’s medical history, just like a human radiologist would review.
The training scale matters here. Hollon’s team used every available MRI collected since U-M Health digitized its radiology records: more than 200,000 MRI studies and 5.6 million imaging sequences. The model also incorporated clinical histories and the reasons physicians ordered each scan.
“Prima works like a radiologist by integrating information regarding the patient’s medical history and imaging data to produce a comprehensive understanding of their health,” explained co-first author Samir Harake.
Hollon describes Prima as “ChatGPT for medical imaging” – a general-purpose diagnostic tool rather than a single-task algorithm.
The bigger picture
This isn’t just about faster results. The research addresses a structural problem in healthcare.
“Whether you are receiving a scan at a larger health system that is facing increasing volume or a rural hospital with limited resources, innovative technologies are needed to improve access to radiology services,” said Vikas Gulani, M.D. Ph.D., chair of U-M’s Department of Radiology.
Rural hospitals and smaller systems often lack specialist coverage. An AI that can flag emergencies and route cases appropriately could extend expert-level triage to underserved areas.
What comes next
The researchers emphasize Prima remains in early evaluation. Future work will incorporate more detailed patient information and electronic medical records to improve accuracy further.
The approach could extend beyond brain imaging. Hollon suggested similar technology might eventually handle mammograms, chest X-rays, and ultrasounds – any imaging where speed and accuracy affect patient outcomes.
Funding came from the National Institute of Neurological Disorders and Stroke, Chan Zuckerberg Initiative, and several U-M foundations.
