Stable Sparse Classifiers predict cognitive impairment from gait patterns.

Area of Science:

• Neuroscience
• Gerontology
• Biomedical Engineering

Background:

• Gait disturbances are linked to cognitive decline, but prediction remains challenging.
• Optimal gait features, protocols, and algorithms are needed for accurate prediction.
• This study investigates gait patterns for cognitive status discrimination.

Purpose of the Study:

• Assess Stable Sparse Classifiers (SSC) for differentiating young, healthy older, and cognitively impaired elderly adults based on gait.
• Identify optimal walking tasks and spatio-temporal gait features (STGF) for prediction using SSC.
• Compare SSC performance against other classification methods.

Main Methods:

• 125 participants (40 young, 85 older adults) with varying cognitive statuses.
• Neuropsychological tests and a cognitive index (MDCog) for group classification.
• Four walking tasks (normal, fast, easy/hard dual-task) measured with an IMU.
• Calculated 16 STGF and dual-task costs; applied SSC for classification.
• Assessed performance using Area Under the Curve (AUC) and identified stable biomarkers.

Main Results:

• SSC effectively discriminated between healthy and cognitively impaired elderly adults (AUC = 0.86).
• The combination of easy dual-task and fast walking tasks yielded the best prediction.
• Gait variability and vertical acceleration amplitude were key predictive features.
• SSC outperformed Linear Discriminant Analysis and Support Vector Machines.

Conclusions:

• Gait pattern changes effectively discriminate between young, healthy older, and cognitively impaired adults.
• Specific gait tasks and STGF, analyzed with SSC, are optimal for cognitive impairment detection.
• SSC demonstrates superior performance compared to other classification techniques for gait-based cognitive assessment.