AI Advances Parkinson’s Detection Using Standard MRI Scans

A simple brain scan may soon be all that’s needed to accurately diagnose Parkinson’s disease, thanks to a new AI-powered tool. The advancement could help doctors expedite detection and treatment, getting patients the care they need and improving their quality of life.  

Developed by teams from the University of Florida (UF) and top-tier medical centers, the machine learning model analyzes MRI scans to distinguish between Parkinson’s disease, multiple system atrophy (MSA), and progressive supranuclear palsy (PSP). These conditions often look similar on brain scans in the early stages, making diagnosis complex and leading to delayed treatment. 

“This AI-based technology is already moving the needle in a more practical and very exciting direction,” said study senior author Michael S. Okun, medical advisor at the Parkinson’s Foundation and director at the Fixel Institute at UF Health. “Doctors are routinely ordering brain MRI scans as part of a regular workup for the potential of a lurking neurodegenerative disease. AI has the potential to transform the scenario to move beyond the eyes of the doctor or radiologist.” 

Researchers introduced the technology, called the Automated Imaging Differentiation for Parkinsonism (AIDP) platform, in a study published in JAMA Neurology. With AIDP, doctors can make faster, more accurate diagnoses without relying on invasive testing or specialized scans using radioactive tracers. By advancing early detection and treatment, the work reflects goals emphasized during Parkinson’s Awareness Month and World Parkinson’s Day.

To train the AI, the researchers used 645 brain scans: 249 from new patients, 396 from earlier studies, and 49 from donated brains examined postmortem. All had confirmed diagnoses of Parkinson’s, MSA, or PSP. Pairing the scans, which show slight changes in brain tissue, with information such as age, gender, and symptoms, AIDP picked up on markers that distinguish one disease from another.

“AI can be used to detect specific patterns of neurodegeneration that reflect the pathological fingerprint of a specific disease,” said study lead author David Vaillancourt, a professor in the Department of Applied Physiology and Kinesiology at UF.

According to Angelos Barmpoutis, study coauthor and UF Digital Worlds Institute professor, the team ran their model using NVIDIA GPUs, including an NVIDIA Quadro P400 on local machines. They analyzed MRI image volumes using the TensorFlow library with NVIDIA CUDA and four NVIDIA A100 Tensor Core GPUs. The large-scale training took about 36 hours to complete.  

The final version of the model trains in just minutes, and a full brain scan with diagnosis is processed in about two hours. 

The researchers found that the AI tool correctly identified the diagnosis in 95% of cases, outperforming expert neurologist teams in some of the most challenging scenarios. Among the postmortem cases, AIDP matched with the ‌confirmed disease 94% of the time—compared with 82% accuracy for clinical diagnosis alone. 

This high level of accuracy could help reduce misdiagnosis and ease the emotional toll on patients and families searching for answers—while getting them on the correct treatment path sooner. With the potential for widespread adoption—the AI tool works across multiple hospitals and various MRI scanners—the cloud-based software could be integrated into care settings, from big hospitals to small clinics and even remote telehealth services. 

Beyond diagnosis, it also has the potential to improve clinical trials by making sure the correct patients are enrolled, which is an ongoing challenge in Parkinson’s research. 

“AIDP is licensed by Neuropacs and will be used in clinical settings once the regulatory hurdle is reached. It can be used now in clinical trials to enrich a sample and make sure that the study includes the right people,” Vaillancourt said. 

Read the study Automated Imaging Differentiation for Parkinsonism.