Researchers at the University of Michigan have developed a new artificial intelligence system that can interpret brain MRI scans in mere seconds, offering rapid and accurate identification of neurological conditions and highlighting cases needing urgent care. This system, known as Prima, was trained on hundreds of thousands of real-world MRI scans along with associated patient histories, achieving diagnostic accuracy of up to 97.5 % — outperforming many existing AI tools. The findings were recently published in Nature Biomedical Engineering, and scientists believe this technology could significantly transform how brain imaging is handled in clinical settings.
Prima’s performance was evaluated over a year using more than 30,000 MRI studies, where it demonstrated strong capabilities across more than 50 different radiologic diagnoses involving major neurological disorders. Importantly, the system doesn’t just identify disease; it also assesses how urgent a patient’s condition may be, such as strokes or brain hemorrhages, and can automatically alert the appropriate medical specialists. This feature could dramatically shorten the time between imaging and medical intervention, potentially improving patient outcomes in life-threatening situations.
Unlike earlier AI tools that were often trained for narrow tasks or limited datasets, Prima functions as a vision language model (VLM) — able to process imaging data together with textual clinical information in real time. The research team included virtually all MRI data available since digitization at the University of Michigan Health system, amounting to over 200,000 MRI studies and 5.6 million imaging sequences. By integrating a patient’s imaging with their medical history and reasons for imaging, Prima mimics aspects of how human radiologists interpret scans, leading to more comprehensive and reliable results.
Researchers acknowledge that while Prima shows great promise, it’s still in the early stages of evaluation. Future work will focus on incorporating even more detailed electronic medical record data and further refining the model’s abilities. The ultimate vision is for AI systems like Prima to act as co-pilots for medical imaging, supporting clinicians across a range of imaging types — including mammograms, chest X-rays, and ultrasounds — and thereby improving workflow efficiency and patient care worldwide.
