Development and Validation of the Automated Imaging Differentiation in Parkinsonism (AID-P): A Multi-Site Machine...

Area of Science:

• Neuroimaging
• Biomarker Development
• Machine Learning in Neurology

Background:

• Critical need for non-invasive biomarkers for Parkinsonian syndromes.
• Current diagnostic methods can be subjective and time-consuming.
• Distinguishing between Parkinson's disease (PD) and atypical Parkinsonism (multiple system atrophy - MSA, progressive supranuclear palsy - PSP) is clinically challenging.

Purpose of the Study:

• To assess the efficacy of non-invasive diffusion MRI (dMRI) in differentiating between various Parkinsonian syndromes.
• To develop and validate machine learning models using dMRI and clinical scores for disease classification.
• To establish a generalizable imaging approach for objective diagnosis.

Main Methods:

• Utilized dMRI data from 1002 subjects across 17 international sites.
• Developed and validated disease-specific machine learning models comparing PD vs. Atypical Parkinsonism (MSA, PSP) and MSA vs. PSP.
• Models integrated dMRI data with Movement Disorders Society Unified Parkinson's Disease Rating Scale part III (MDS-UPDRS III) scores, evaluated using Area Under the Curve (AUC) of Receiver Operating Characteristic (ROC) curves.

Main Results:

• High AUCs were achieved in the test cohort for combined dMRI + MDS-UPDRS III models (PD vs. Atypical: 0.962; MSA vs. PSP: 0.897) and dMRI Only models (PD vs. Atypical: 0.955; MSA vs. PSP: 0.926).
• MDS-UPDRS III Only models showed significantly lower AUCs (PD vs. Atypical: 0.775; MSA vs. PSP: 0.582).
• Demonstrated superior performance of dMRI-based models over clinical scores alone in differentiating Parkinsonian syndromes.

Conclusions:

• Multi-site dMRI cohorts provide an objective, validated, and generalizable imaging approach to distinguish Parkinsonian syndromes.
• The automated dMRI method is non-invasive, requires minimal scan time (<12 min), and is compatible with global 3T scanners.
• This dMRI approach has the potential to significantly improve clinical care, reduce misdiagnoses, and enhance clinical trial accuracy for Parkinson's disease and Parkinsonism.