Detecting Sensitive Mobility Features for Parkinson's Disease Stages Via Machine Learning.

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Parkinson's disease (PD) severity assessment often relies on subjective clinical scales.
• Objective gait and mobility measures are needed to track PD progression across its spectrum.
• Current understanding of how specific gait parameters reflect PD stages is limited.

Purpose of the Study:

• To identify sensitive gait and mobility measures for PD motor stages.
• To determine optimal wearable sensor locations for different PD stages.
• To utilize machine learning for objective PD assessment.

Main Methods:

• Collected wearable sensor data from 332 PD patients (Hoehn and Yahr I-III) and 100 controls.
• Utilized sensors on the lower back, ankles, and wrists during walking and dual-task conditions.
• Applied machine learning algorithms for feature selection and classification.

Main Results:

• Achieved high discrimination between PD motor stages (sensitivity 72%-83%, specificity 69%-80%, AUC 0.76-0.90).
• Upper-limb sensors distinguished early PD from controls.
• Trunk sensors identified turning measures in mid-stage PD.
• Stride timing and regularity were key in advanced PD stages.

Conclusions:

• Machine learning applied to wearable sensor data reveals distinct gait and mobility features for different PD stages.
• These objective measures can enhance PD monitoring and clinical trial design.
• Wearable technology offers a promising avenue for objective Parkinson's disease assessment.