Automatic Detection of Gait Perturbations With Everyday Wearable Technology

Area of Science:

• Biomedical Engineering
• Gerontology
• Wearable Technology

Background:

• Older adults have a high risk of falls, significantly impacting their health and independence.
• Individual responses to gait disturbances are key indicators of fall risk.
• Existing monitoring methods may not be suitable for continuous, real-world application.

Purpose of the Study:

• To investigate the automatic detection of gait perturbations using data from everyday wearable devices.
• To evaluate the performance of a deep learning algorithm for identifying gait instability.
• To assess the feasibility of using hearing aids and smartphones for fall risk monitoring.

Main Methods:

• Utilized a deep convolutional long short-term memory (DeepConvLSTM) algorithm for gait perturbation detection.
• Collected acceleration and angular velocity data from hearing aids, smartphones, and professional sensors.
• Tested the algorithm on treadmill-generated gait perturbations.

Main Results:

• Achieved an F1 score of at least 0.68 and a recall of 0.86 across all tested devices (hearing aids, smartphones, professional sensors).
• Detection performance remained consistent regardless of sensor placement or data type (acceleration only).
• No significant performance difference was observed between everyday wearables and professional equipment.

Conclusions:

• Everyday wearables like hearing aids and smartphones can effectively monitor gait perturbations.
• These devices show comparable performance to professional sensors in detecting instability.
• Continuous monitoring of gait perturbations using accessible wearables holds significant potential for proactive fall risk assessment in older adults.