Step Length and Gait Speed Estimation Using a Hearing Aid Integrated Accelerometer: A Comparison of Different...

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Digital Health

Background:

• Gait analysis is vital for health monitoring, with abnormalities linked to falls and cognitive decline.
• Wearable sensors, particularly unobtrusive earables, offer continuous, in-home gait assessment.
• A standardized pipeline for gait analysis using ear-worn accelerometers is currently lacking.

Purpose of the Study:

• To develop and compare algorithms for estimating spatial-temporal gait parameters (step length, gait speed) using ear-worn accelerometers.
• To establish a comprehensive gait analysis pipeline for ear-worn sensors.
• To validate the precision and clinical relevance of these estimations against optical motion capture.

Main Methods:

• Three algorithmic approaches were developed: a biomechanical model, feature-based machine learning (ML), and a convolutional neural network.
• Performance was evaluated on step and walking bout levels, comparing estimations with optical motion capture.
• The models utilized data from ear-worn accelerometers.

Main Results:

• The feature-based ML model demonstrated superior performance, achieving 4.8cm precision for step length on a walking bout level.
• The ML approach estimated gait speed with an absolute percentage error of 5.4% (± 4.0%).
• The ML model showed robustness across age groups and sampling rates, though sensitive to walking speed variations.

Conclusions:

• Machine learning models can accurately estimate step length and gait speed from ear-worn accelerometers with clinically relevant precision.
• This research pioneers gait analysis using ear-worn sensors, enabling continuous, long-term health monitoring.
• The developed framework supports unobtrusive digital health assessments integrated into daily routines.