A new AI system from the University of Michigan can scan a brain MRI in seconds, spot serious conditions and flag which patients need urgent care. Researchers say the tool could help overburdened hospitals deliver faster, more accurate diagnoses.
An artificial intelligence system developed at the University of Michigan can read a brain MRI and help diagnose a patient in seconds, potentially reshaping how hospitals handle one of the most critical and time-sensitive imaging tests in medicine.
The model, nicknamed Prima, analyzed brain scans with up to 97.5% accuracy in a new study and could also predict how urgently a patient needed treatment, according to researchers. The work, published in the journal Nature Biomedical Engineering, points to a future in which AI helps radiologists manage soaring demand for MRI while reducing delays and diagnostic errors.
The technology is designed to address a growing bottleneck in care, according to senior author Todd Hollon, a neurosurgeon at University of Michigan Health and assistant professor of neurosurgery at the U-M Medical School.
“As the global demand for MRI rises and places significant strain our physicians and health systems, our AI model has potential to reduce burden by improving diagnosis and treatment with fast, accurate information,” Hollon said in a news release.
Each year, millions of MRI studies are performed worldwide, and a large share focus on the brain. But there are not enough neuroradiologists to read all those scans quickly, especially at smaller or rural hospitals. In many places, patients can wait days or longer for results, even when they have conditions that require rapid treatment.
Prima is built to help close that gap.
What makes Prima different
Prima is a type of artificial intelligence called a vision language model, or VLM. Unlike earlier tools that were trained to do just one narrow job — such as spotting a specific type of lesion or estimating dementia risk — Prima was trained on nearly everything the health system had.
Hollon’s team fed the model every MRI performed at University of Michigan Health since radiology went digital decades ago: more than 200,000 MRI studies and 5.6 million individual imaging sequences. They also included patients’ clinical histories and the reasons doctors ordered the scans.
Co-first author Samir Harake, a data scientist in Hollon’s Machine Learning in Neurosurgery Lab, noted that combination of imaging and clinical data is key to how the system works.
“Prima works like a radiologist by integrating information regarding the patient’s medical history and imaging data to produce a comprehensive understanding of their health,” Harake said in the news release. “This enables better performance across a broad range of prediction tasks.”
In testing, the researchers ran Prima on more than 30,000 MRI studies over the course of a year. Across more than 50 radiologic diagnoses covering major neurological disorders, the model outperformed other state-of-the-art AI systems on diagnostic performance, the team reported.
Crucially, Prima is not just grading images in the background. It is designed to plug directly into clinical workflows.
Speed and triage for urgent cases
Some brain conditions, such as hemorrhages and strokes, are medical emergencies where minutes can mean the difference between recovery and permanent disability. In those cases, the ability to flag a scan quickly and get it in front of the right specialist can be lifesaving.
Prima can automatically identify scans that suggest an urgent problem and alert the appropriate subspecialist — for example, a stroke neurologist or neurosurgeon — as soon as the patient finishes imaging.
Co-first author Yiwei Lyu, a postdoctoral fellow in computer science and engineering at U-M, emphasized that the system is designed to be both fast and precise.
“Accuracy is paramount when reading a brain MRI, but quick turnaround times are critical for timely diagnosis and improved outcomes,” Lyu added. “At key steps in the process, our results show how Prima can improve workflows and streamline clinical care without abandoning accuracy.”
That kind of automated triage could be especially valuable in busy urban hospitals facing high imaging volumes and in rural facilities with limited access to neuroradiology expertise.
“Whether you are receiving a scan at a larger health system that is facing increasing volume or a rural hospital with limited resources, innovative technologies are needed to improve access to radiology services,” added co-author Vikas Gulani, the chair of the Department of Radiology at U-M Health. “Our teams at University of Michigan have collaborated to develop a cutting-edge solution to this problem with tremendous, scalable potential.”
A co-pilot, not a replacement
The researchers stress that Prima is not meant to replace radiologists. Instead, they describe it as a kind of “ChatGPT for medical imaging” that can act as a co-pilot — rapidly reviewing scans, suggesting likely diagnoses and prioritizing which cases need immediate human attention.
Like popular AI tools that can draft emails or summarize documents, Prima is built to assist, not to make final decisions.
“Like the way AI tools can help draft an email or provide recommendations, Prima aims to be a co-pilot for interpreting medical imaging studies,” Hollon added. “We believe that Prima exemplifies the transformative potential of integrating health systems and AI-driven models to improve health care through innovation.”
Because the model was trained on health system–scale data, it may also be more adaptable than earlier, narrowly focused tools. The team believes similar approaches could eventually be extended beyond brain MRI to other imaging types, such as mammograms, chest X-rays and ultrasounds.
What comes next
Despite the promising results, the researchers describe this work as an early stage of evaluation. Before a system like Prima could be widely adopted, it would need to be tested in more hospitals, on more diverse patient populations and under real-world conditions.
The team’s next steps include integrating more detailed patient information and electronic health record data into the model, further mirroring how radiologists and clinicians interpret imaging in practice. They also note that health systems and policymakers are still working out how best to regulate and deploy AI safely in clinical care.
For now, the study offers a glimpse of how large-scale data and advanced AI models might help health systems keep pace with growing imaging demands — and get critical information to patients and doctors when it matters most.
Source: Michigan Medicine