Mobile Spatiotemporal Gait Segmentation Using an Ear-Worn Motion Sensor and Deep Learning.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Gait Analysis

Background:

• Mobile health (mHealth) enables continuous mobility and gait assessment in real-world settings.
• Traditional gait analysis relies on body-fixed sensors, limiting practical applications.
• Early diagnosis and monitoring of gait disorders are crucial for preventing adverse events like falls.

Purpose of the Study:

• To investigate the potential of an ear-worn motion sensor for gait pattern analysis.
• To develop and validate an algorithm for spatiotemporal gait segmentation using ear-worn sensor data.
• To explore the feasibility of integrating ear-worn gait monitoring with in-ear vital-sign monitoring.

Main Methods:

• Collected 3D acceleration data from ear-worn sensors in 53 healthy adults during varied walking speeds.
• Trained temporal convolutional networks to detect stepping sequences and predict spatial gait relations.
• Validated the mEar algorithm's accuracy in detecting ground contacts and determining gait cycle characteristics.

Main Results:

• The mEar algorithm achieved high accuracy in detecting initial (F1 score: 99%) and final (F1 score: 91%) ground contacts.
• Demonstrated good to excellent validity in determining temporal and spatial gait parameters like stride time and length.
• Showed precision sufficient for monitoring clinically relevant changes in walking speed, variability, and asymmetry.

Conclusions:

• The ear is a viable anatomical site for unobtrusive gait monitoring using motion sensors.
• The mEar algorithm provides accurate and valid gait analysis, supporting early diagnosis and disease progression monitoring.
• Integrating ear-worn gait sensors with vital-sign monitoring offers a practical approach for comprehensive telemedical health applications.