University of Michigan's Prima AI model reads brain MRI scans in seconds with 97.5% accuracy

Prima AI Model Transforms Brain MRI Diagnosis Speed

Researchers at the University of Michigan have developed Prima, an AI model that can analyze brain MRI scans and deliver diagnoses in seconds, achieving up to 97.5% accuracy in detecting neurological conditions1

1

. Published in Nature Biomedical Engineering, this breakthrough technology addresses a critical gap in healthcare as global demand for MRI studies continues to strain health systems and extend turnaround times2

2

. Led by Todd Hollon, a neurosurgeon at University of Michigan Health, the research team tested Prima on 29,431 MRI studies over a one-year period, demonstrating its potential to reshape how brain imaging is handled across the United States3

3

.

Trained on MRI Studies at Unprecedented Scale

Unlike previous AI systems trained on carefully curated subsets of MRI data for narrow tasks, Prima was trained on over 220,000 MRI studies and 5.6 million imaging sequences collected since radiology digitization began at University of Michigan Health decades ago1

1

. The system incorporates patient clinical histories and physicians' reasons for ordering each imaging study, enabling it to function like a radiologist by integrating multiple data sources3

3

. Samir Harake, co-first author and data scientist in Hollon's Machine Learning in Neurosurgery Lab, explains that this comprehensive approach produces a better understanding of patient health and enables superior diagnostic performance across a broad range of prediction tasks2

2

.

High Diagnostic Accuracy Across Neurological Conditions

Across 52 radiologic diagnoses from major neurological disorders, Prima achieved a mean diagnostic area under the curve of 92.0%, outperforming other state-of-the-art general and medical AI models1

1

. The system demonstrated algorithmic fairness across sensitive groups and offers explainable differential diagnoses to clinicians1

1

. Yiwei Lyu, co-first author and postdoctoral fellow of Computer Science and Engineering at U-M, emphasizes that while accuracy is paramount when reading a brain MRI, quick turnaround times are critical for timely diagnosis and improved outcomes4

4

.

Vision Language Model Architecture Enables Real-Time Analysis

Prima functions as a vision language model (VLM), an AI system that can simultaneously process video, images, and text in real time3

3

. This architecture allows Prima to integrate diverse data types and deliver comprehensive diagnostic insights immediately after a patient completes imaging2

2

. The foundation model uses a hierarchical vision architecture that provides general and transferable MRI features, enabling it to handle real-world clinical MRI studies as input1

1

.

Prioritize Emergency Cases and Reduce MRI Wait Times

Prima can automatically identify conditions requiring immediate medical attention, such as strokes and brain hemorrhages, and alert the appropriate subspecialist—whether a stroke neurologist or neurosurgeon2

2

. The system provides worklist prioritization for radiologists and clinical referral recommendations, with feedback available immediately after imaging completion1

1

. This capability addresses a pressing need as millions of MRI scans are performed globally each year, with demand outpacing the availability of neuroradiology services and contributing to diagnostic delays that can extend for days or longer depending on location3

3

.

Address Radiology Shortages Across Health Systems

Vikas Gulani, co-author and chair of the Department of Radiology at U-M Health, notes that innovative technologies are essential to improve access to radiology services whether patients receive scans at larger health systems facing increasing volume or rural hospitals with limited resources2

2

. The imbalance between growing MRI demand and available neuroradiology services has contributed to staffing shortages, diagnostic delays, and errors—challenges that disproportionately impact patients in low-resource and rural settings1

1

. Hollon describes Prima as "ChatGPT for medical imaging," suggesting the technology could eventually be adapted for other imaging modalities including mammograms, chest X-rays, and ultrasounds4

4

.

Future Development to Diagnose Neurological Conditions More Accurately

While Prima performed strongly in initial testing, researchers emphasize the work remains in an early evaluation phase2

2

. Future research will focus on integrating more detailed patient information and electronic medical record data to further improve diagnostic accuracy, mirroring how radiologists and physicians interpret MRIs in real clinical settings3

3

. Hollon believes Prima exemplifies the transformative potential of integrating health systems and AI-driven models to improve healthcare through innovation, positioning the system as a co-pilot for interpreting medical imaging studies rather than a replacement for human expertise3

3

.