AI Model "Prima" Reads Brain MRIs in Seconds with 97.5% Accuracy, Identifies 50+ Conditions

Researchers at the University of Michigan have developed "Prima," an artificial intelligence model that can interpret brain MRI scans in seconds, accurately identifying more than 50 neurological conditions and automatically flagging urgent cases for immediate attention. Published in Nature Biomedical Engineering, the system represents a significant advance in AI-assisted medical imaging.

Prima was trained on more than 200,000 MRI studies comprising 5.6 million individual imaging sequences, making it one of the most comprehensively trained medical AI systems in neurology. Over a year-long testing period involving more than 30,000 studies, it achieved accuracy rates of up to 97.5% across a wide range of conditions.

“Published in Nature Biomedical Engineering, the system represents a significant advance in AI-assisted medical imaging.”

What makes Prima particularly valuable in clinical settings is its ability to triage cases automatically. When the system detects potential emergencies — such as signs of acute stroke, brain hemorrhage, or dangerous swelling — it immediately alerts healthcare providers, potentially shaving critical minutes off response times that can mean the difference between recovery and permanent damage.

The system is designed to augment, not replace, radiologists. Prima provides initial reads that help prioritize the order in which radiologists review cases, ensuring that the most urgent studies are seen first. It also serves as a quality-control check, flagging potential findings that a busy radiologist might miss.

Brain MRI interpretation is one of the most complex tasks in medical imaging, requiring years of specialized training. With growing demand for neuroimaging and a shortage of neuroradiologists in many regions, AI tools like Prima could help democratize access to expert-level brain MRI interpretation, particularly in underserved communities and developing countries.