AI Model Reads Brain MRIs in Seconds with Up to Ninety-Seven Percent Accuracy

AI That Reads Your Brain Scan Before the Doctor Can

A new artificial intelligence model developed at the University of Michigan can read a complete brain MRI and deliver a diagnosis in mere seconds, a process that traditionally takes radiologists considerably longer. The technology, called Prima, was published in Nature Biomedical Engineering and represents the first foundation model built specifically for clinical neuroimaging.

How Prima Works

Prima is a vision language model, meaning it can process images, video, and text simultaneously in real time. Unlike previous AI tools that were trained on small, carefully selected datasets for narrow tasks, Prima was trained on the entirety of the University of Michigan Health system's MRI archive, spanning over two hundred and twenty thousand studies and five point six million imaging sequences accumulated since radiology went digital.

The model integrates a patient's clinical history, the physician's reason for ordering the scan, and the full set of imaging data to produce a comprehensive diagnostic assessment. It uses a hierarchical vision architecture with a contrastive learning framework, allowing it to develop general and transferable features for understanding brain scans.

Impressive Accuracy Across Dozens of Conditions

In a year-long clinical test covering approximately thirty thousand MRI studies, Prima achieved a mean diagnostic accuracy of ninety-two percent across fifty-two different neurological diagnoses, including tumours, inflammatory conditions, infections, and developmental abnormalities. For certain conditions, its accuracy reached as high as ninety-seven point five percent, outperforming other leading AI models.

Beyond diagnosis, Prima can prioritise urgent cases such as brain haemorrhages or strokes, automatically alerting the appropriate specialist, whether a stroke neurologist or neurosurgeon, immediately after a patient completes their scan.

Addressing a Growing Crisis

The timing could not be more critical. Millions of MRI studies are performed globally each year, yet the supply of radiologists is not keeping pace with demand. Some facilities take days to deliver results, and workforce shortages contribute to diagnostic errors. Prima offers a potential solution, particularly for rural hospitals with limited specialist access.

The research team plans to expand the technology by integrating electronic medical records and extending the approach to other imaging types, including mammography, chest X-rays, and ultrasound.