Machine Learning Recognizes Stages of Parkinson's Disease Using Magnetic Resonance Imaging

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Neurodegenerative diseases (NDs) like Parkinson's disease (PD) pose a growing global health challenge.
• Early detection of PD is critical for implementing effective intervention strategies.
• Current diagnostic methods may not identify PD in its earliest stages.

Purpose of the Study:

• To investigate the potential of structural analysis of brain regions from T1-weighted MRI scans for modeling PD stages.
• To apply standard machine learning (ML) techniques to differentiate between healthy controls (HC), prodromal (PR), and PD groups.
• To assess the efficacy of using subcortical brain structure volumes and spatial relationships for PD staging.

Main Methods:

• Utilized T1-weighted MRI scans from the PPMI database (N=168).
• Employed machine learning models including Logistic Regression, Random Forest, Support Vector Classifier, and Rough Sets.
• Features included volumes and spatial distances (Euclidean, Cosine) of subcortical structures relative to the thalamus.

Main Results:

• Logistic Regression demonstrated optimal performance with high accuracy (85%), precision (88%), and recall (85%) in PD-stage recognition.
• The models successfully differentiated between HC, PR, and PD groups.
• Interpretable metrics like volume and centroid-based spatial distances contributed to high diagnostic accuracy.

Conclusions:

• Structural analysis of brain regions using MRI and ML offers a promising framework for early PD identification.
• The developed models show significant potential for non-invasive, accurate PD staging.
• This approach could facilitate timely interventions and improve patient outcomes.