Designing a Gait Recognition Algorithm for Older Adults Using Mobility Aids: Prospective Cohort Study.

Area of Science:

• Gerontology
• Biomedical Engineering
• Signal Processing

Background:

• Maintaining mobility is crucial for older adults' independence and fall prevention.
• Wearable technology can track physical activity, but existing gait recognition algorithms lack accuracy for older adults, especially those using mobility aids.

Purpose of the Study:

• Develop a gait recognition algorithm for older adults that is robust to mobility aid use.
• Utilize wrist-worn wearable devices for a widely applicable approach.

Main Methods:

• Collected walking and daily activity data from 17 older adults, including 10 who used mobility aids.
• Developed a computationally lightweight, heuristic-based algorithm using accelerometer data and harmonic patterns.
• Optimized algorithm parameters using hyperparameter tuning with a Parzen tree estimator in a leave-one-subject-out validation.

Main Results:

• Calibrated algorithm detected subtle walking patterns of older adults, widening the frequency range and adjusting the superharmonic ratio.
• Sensitivity increased from 0.11 to 0.73, F1-score from 0.15 to 0.73, and specificity decreased from 0.99 to 0.75.
• Successfully recognized walking patterns in older adults with and without mobility aids.

Conclusions:

• The developed algorithm shows promise for monitoring physical activity in older adults, being computationally lightweight and explainable.
• The calibration approach is adaptable to new populations, with low adoption barriers due to reliance on standard accelerometer data.
• Further validation with larger cohorts using mobility aids is needed to refine parameters for this population.