Parkinson's disease severity clustering based on gait activity from mobile device.

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Parkinson's disease (PD) significantly impacts mobility through gait impairments.
• Objective assessment of PD severity is vital for effective clinical management.
• Current methods may lack precision in monitoring disease progression.

Purpose of the Study:

• To investigate the utility of smartphone-derived gait data for objective Parkinson's disease severity clustering.
• To apply unsupervised machine learning techniques for precise disease monitoring.
• To identify gait patterns associated with varying levels of PD severity.

Main Methods:

• Analysis of accelerometer data from the mPower dataset (n=1957).
• Stride cycle segmentation, sequence padding, k-means clustering with Dynamic Time Warping (DTW), and t-SNE visualization.
• Correlation of identified gait clusters with MDS-UPDRS scores (Parts I & II) for severity assessment.

Main Results:

• Four distinct gait clusters were identified, correlating significantly with PD severity.
• The most severe cluster showed substantially higher scores for balance/walking problems and freezing episodes.
• t-SNE visualization confirmed clear separation of clusters, with higher severity concentrated in one group.

Conclusions:

• Smartphone gait data analyzed with unsupervised learning effectively stratifies Parkinson's disease severity.
• Gait features related to balance and freezing are critical biomarkers for PD progression.
• This approach offers a potential for objective, scalable tools for PD assessment using wearable technology.