Classification of Parkinson's disease and its stages using machine learning

Area of Science:

• Digital Health
• Machine Learning
• Neuroscience

Background:

• Digital health technology generates vast data, necessitating advanced analysis methods like machine learning (ML).
• Parkinson's disease (PD) diagnosis and staging traditionally rely on clinical assessments and questionnaires, which may not capture the full spectrum of neurocognitive changes.
• Understanding neurocognitive function in PD is crucial for disease management and technological development.

Purpose of the Study:

• To employ ML classification for assessing the utility of tablet-based neurocognitive features and self-reported metrics in relation to PD and its Hoehn and Yahr (H&Y) stages.
• To compare perceived versus sensor-based neurocognitive abilities in individuals with PD.
• To identify significant neurocognitive features for PD detection and staging.

Main Methods:

• 75 participants (PD and controls) completed 14 tablet-based neurocognitive tests, functional movement assessments, and health questionnaires.
• Decision tree classification was used to analyze sensor-based neurocognitive features and self-reported metrics.
• Feature significance was determined for discriminating PD from controls and early from advanced PD stages.

Main Results:

• ML classification accurately distinguished PD from healthy controls ([Formula: see text]) and differentiated early from advanced PD stages ([Formula: see text]).
• Sensor-based features, particularly device acceleration magnitude, were highly significant across multiple tests.
• Perceived neurocognitive function differed from sensor-based measures, with early-stage PD patients underestimating and advanced-stage patients overestimating certain abilities.

Conclusions:

• ML applied to digital health assessments provides a robust method for PD detection and staging, outperforming traditional tools.
• Sensor-based neurocognitive features offer valuable, objective data for understanding PD progression.
• Discrepancies between perceived and sensor-based abilities highlight the need for tailored digital health technology configurations in PD management.