AI reads brain MRIs in seconds and flags emergencies

Researchers at the University of Michigan have developed an artificial intelligence system named Prima that can analyze brain MRI scans within seconds and accurately identify a wide range of neurological conditions. The AI model demonstrated diagnostic accuracy of up to 97.5% across more than 50 different radiologic diagnoses, outperforming existing advanced AI tools. Importantly, Prima can also assess the urgency of cases, automatically flagging life-threatening conditions such as strokes and brain hemorrhages to ensure rapid medical intervention. The Prima system was tested over a year using more than 30,000 MRI studies, incorporating extensive real-world data including patient histories. Unlike previous AI models that focused on narrow diagnostic tasks, Prima employs a vision language model (VLM) approach, enabling it to process images, videos, and text simultaneously. This broad training allows the system to deliver comprehensive diagnostic insights and prioritize cases that require immediate attention, notifying the appropriate specialists such as stroke neurologists or neurosurgeons. The development of Prima addresses growing demands on healthcare systems as MRI usage increases globally, often straining physician workloads and delaying diagnoses. By providing rapid and accurate readings, Prima has the potential to streamline clinical workflows, reduce physician burden, and improve patient outcomes through faster treatment decisions. Researchers highlight that the system’s immediate feedback after imaging can transform how brain scans are handled across hospitals and health networks. This advancement marks a significant step forward in medical imaging technology, combining high accuracy with speed and clinical prioritization. As AI continues to evolve in healthcare, tools like Prima may become integral in emergency settings, helping to triage critical neurological cases more effectively and potentially saving lives through timely diagnosis and intervention.