Could AI reading brain MRIs at 97.5% accuracy change the future of neurologists?

This AI reads brain MRIs in seconds and doctors are paying attention

A new artificial intelligence system from the University of Michigan can analyze a brain MRI and deliver a diagnosis in seconds. Researchers say it achieved an accuracy of up to 97.5% across a wide range of neurological conditions, a level that could reshape how brain scans are handled in busy hospitals.

The study, published in Nature Biomedical Engineering, reports that Prima can prioritize scans by urgency in a prospective, health system trial, though additional work is needed to integrate such triage into live hospital alerting workflows.

Why brain MRI delays are a growing problem?

MRI demand is rising worldwide, especially for neurological conditions. But the number of trained neuroradiologists has not kept pace, creating pressure on health systems and long wait times for patients who need answers fast.

In some places, it can take days or longer to get MRI results. Researchers say that the gap can lead to delays in diagnosis and treatment, and in certain cases, missed warning signs that require immediate medical attention.

Meet Prima, the AI built to think like a radiologist

The system is called Prima. It is a vision language model, meaning it can process images, video, and text together in real time. That allows it to go beyond just spotting patterns in pictures.

Researchers say Prima works more like a radiologist by combining imaging data with patient history and the reason a scan was ordered. That broader view helps it generate more complete predictions about a patient’s condition.

How accurate is this AI really?

During testing, Prima analyzed more than 30,000 MRI studies over a year. Across more than 50 different radiologic diagnoses tied to major neurological disorders, it outperformed other leading AI systems, according to the research team.

The team reports task-level top performance of up to 97.5% for certain diagnoses, while the model’s mean diagnostic area under the ROC curve across the 52 tasks was reported at nearly 90%.

In a field where small differences can mean life or death, that level of performance has drawn attention from doctors and researchers watching AI’s role in medicine.

It does more than just label a scan

Prima is not only designed to identify conditions. It can also help determine which cases should be prioritized, flagging scans that may point to emergencies like strokes or brain hemorrhages.

The system can prioritize and flag cases that appear urgent, which could be connected to future alerting systems. If validated and properly integrated into clinical workflows, such triage support could shorten time to specialist review when minutes matter.

Helping route patients to the right specialist

The model can suggest subspecialty escalation, for example, recommending a stroke neurologist or neurosurgeon when findings point to those specialties.

Researchers say this feedback can be available immediately after imaging is complete. Instead of waiting for manual review, hospitals could have earlier guidance on where to send urgent cases.

Trained on an enormous amount of real world data

Unlike many earlier AI systems that relied on smaller, handpicked datasets, Prima was trained on every MRI collected since radiology digitization began at the University of Michigan Health. That added up to more than 200,000 studies.

The training also included 5.6 million imaging sequences. By learning from such a large and varied pool, the model was exposed to a broad range of conditions, scanner types, and clinical scenarios.

Patient history plays a key role

Researchers did not stop at images. They also fed the model information from patient medical histories and the reasons doctors ordered each imaging study. That mirrors how human radiologists interpret scans in context.

By combining these details, Prima can form a more complete picture of a person’s health. The team says this approach improves performance across many different prediction tasks instead of focusing on just one disease.

Designed to ease pressure on overworked systems

Millions of MRI studies are performed globally every year, many tied to neurological diseases. Researchers say demand often outpaces the availability of neuroradiology services, contributing to workforce shortages and diagnostic challenges.

Tools like Prima are being explored as a way to reduce that burden. Faster preliminary analysis could help physicians manage growing caseloads while still aiming for accurate and timely decisions.

Still early, but future upgrades are planned

The researchers stress that Prima is still in the early stages of evaluation. More studies and real-world testing will be needed before systems like this are widely integrated into everyday clinical practice.

Next steps include adding even more detailed patient information and electronic medical record data. The goal is to further mirror how physicians combine multiple data sources when making complex diagnostic decisions.

Could this AI expand beyond brain scans?

While Prima focuses on neuroimaging, its creators see broader potential. The same type of system could one day be adapted to help interpret other imaging tests such as mammograms, chest X-rays, and ultrasounds.

Researchers compare the idea to having a co-pilot for medical imaging. Instead of replacing doctors, the aim is to support them with fast, data-driven insights during busy clinical workflows.

For a closer look at the safeguards being debated as AI moves deeper into healthcare, read Europe’s AI healthcare push faces scrutiny over patient protections, says WHO.

What this could mean for neurologists?

If tools like Prima prove reliable in real-world settings, neurologists and radiologists could spend less time on routine scan reviews and more time on complex cases and direct patient care. AI could act as an early screening partner.

At the same time, experts and policymakers are still working out how best to integrate AI into health care. Questions about oversight, safety, and workflow design will shape how systems like this are used.

Will AI take the place of neurologists? See JP Morgan warns AI could trigger violent job churn while boosting productivity.

What do you think about AI reading brain MRIs with high accuracy? Share your thoughts.

This slideshow was made with AI assistance and human editing.

Don’t forget to follow us for more exclusive content on MSN.

Read More From This Brand:

• OpenAI brings on a psychiatrist after users struggle with AI-linked mental health
• Could AI automate your job by 2026? Investors expect layoffs
• China issues a warning on AI addiction while unveiling new draft rules