A new AI-driven approach is revolutionizing how neurologists distinguish between Parkinson’s disease and similar neurodegenerative disorders.

In a landmark multicenter study published in JAMA Neurology on March 17, 2025, researchers demonstrated that machine learning combined with diffusion MRI can accurately differentiate between Parkinson’s disease (PD), multiple system atrophy (MSA), and progressive supranuclear palsy (PSP)—three major forms of parkinsonism that are notoriously difficult to distinguish clinically.

The Challenge of Diagnosing Parkinsonism

Diagnosing parkinsonian disorders has traditionally relied on clinical evaluations and specialist expertise. Tools like DaT SPECT scans and biomarker assays can help, but they’re expensive, time-consuming, and often fall short in differentiating between PD, MSA, and PSP. Even experienced neurologists can misdiagnose, especially in early stages where symptoms overlap.

This diagnostic uncertainty has major implications—not just for patient care, but also for inclusion in clinical trials and targeted treatments.

Enter AIDP: Automated Imaging Differentiation for Parkinsonism

The study introduced a novel tool called Automated Imaging Differentiation for Parkinsonism (AIDP). This AI model leverages 3-Tesla diffusion MRI—an imaging technique already common in hospitals—and applies machine learning to extract meaningful patterns from the scans. Specifically, it uses features like free water (FW) and fractional anisotropy to detect microstructural brain changes that distinguish between neurodegenerative disorders.

The AI is trained on data from over 640 patients across 21 sites in the U.S. and Canada, making it one of the most comprehensive prospective studies in the field to date.

Outstanding Accuracy and Real-World Reliability

AIDP exceeded expectations in diagnostic accuracy:

• PD vs Atypical Parkinsonism: AUROC of 0.96
• MSA vs PSP: AUROC of 0.98
• PD vs MSA: AUROC of 0.98
• PD vs PSP: AUROC of 0.98

Crucially, the model’s predictions were also compared to postmortem brain pathology in a subset of cases. AIDP correctly predicted the final diagnosis in 93.9% of autopsy cases, outperforming the clinical diagnosis, which had an 81.6% confirmation rate.

Even when tested on unseen data from holdout medical centers, the model’s performance held strong—underscoring its generalizability and potential for widespread clinical adoption.

Why This Matters

This breakthrough positions AIDP as a scalable, non-invasive, and accurate tool that can assist neurologists in diagnosing parkinsonian syndromes earlier and more confidently. By integrating into standard MRI workflows—without the need for radioactive tracers or expensive biomarker assays—AIDP could significantly reduce diagnostic delays and errors.

Moreover, combining AIDP with emerging blood- and skin-based biomarkers might further improve diagnostic precision and help establish biological subtypes of Parkinson’s disease, similar to the ATN framework used in Alzheimer’s research.

Looking Ahead

The team behind AIDP plans to extend the model to cover prodromal cases, Lewy body dementia, and other movement disorders. They also aim to scale the cloud-based software solution for seamless integration into hospital PACS systems.

With neurodegenerative diseases on the rise and accurate diagnosis being a persistent challenge, AIDP represents a major leap forward in neurological care powered by artificial intelligence.

Source:
Vaillancourt, D.E., Barmpoutis, A., Wu, S.S., et al. (2025). Automated Imaging Differentiation for Parkinsonism. JAMA Neurology. Published online March 17, 2025. doi:10.1001/jamaneurol.2025.0112