University of Michigan AI System Reads and Diagnoses Brain MRIs in Seconds with 97.5% Accuracy

Researchers at the University of Michigan have created an AI system called Prima that can interpret brain MRI scans in just seconds, achieving accuracy as high as 97.5% across more than 50 different neurological diagnoses. The technology, described as first-of-its-kind, has the potential to reshape how brain imaging is handled across health systems.

Prima is classified as a vision language model (VLM), a type of artificial intelligence that can process images, video, and text together in real time. Unlike traditional AI diagnostic tools that focus on a single condition, Prima can simultaneously assess for dozens of different neurological disorders, including strokes, brain hemorrhages, tumors, and degenerative diseases.

“5% across more than 50 different neurological diagnoses.”

The system was trained on an unprecedented dataset: more than 200,000 MRI studies and 5.6 million individual imaging sequences collected since radiology records were digitized at University of Michigan Health. The training data also incorporated patients' clinical histories and the reasons physicians ordered each scan, giving the AI crucial context for its diagnoses.

Across its evaluation, Prima outperformed other advanced AI models, delivering stronger diagnostic performance on a comprehensive range of conditions. The system was specifically designed to identify emergency cases — such as strokes and brain hemorrhages — that demand immediate attention, enabling it to notify the most appropriate subspecialist.

The clinical implications are significant. Neuroimaging departments often face backlogs, with radiologists struggling to keep pace with the volume of scans. Prima could serve as a triage tool, rapidly identifying the most urgent cases and allowing radiologists to focus their expertise on complex interpretations.

The research team emphasizes that Prima is designed to augment, not replace, radiologists. The AI provides rapid preliminary assessments that can be confirmed and refined by human experts, combining the speed of machine intelligence with the nuanced judgment of experienced clinicians.