Accurate Gait Assessment and Reduced Patient Burden from a Chest-Mounted Accelerometer.

Area of Science:

• Biomechanics
• Wearable technology
• Machine learning

Background:

• Gait analysis is crucial for assessing mobility impairments in neurological disorders.
• Current methods often involve cumbersome equipment, limiting real-world application.
• There is a need for accurate, non-invasive gait assessment in clinical trials and daily monitoring.

Purpose of the Study:

• To investigate the efficacy of chest acceleration data for estimating gait parameters.
• To leverage wavelet analysis and machine learning for gait analysis using chest-mounted sensors.
• To assess the feasibility of a minimally obtrusive method for continuous gait monitoring.

Main Methods:

• Utilized chest acceleration data and wavelet analysis.
• Applied machine learning models, including deep learning and a generalized inverted pendulum model.
• Validated stride event segmentation against motion capture for accuracy.

Main Results:

• Achieved low error rates for stride time (RMSE 0.043s) and stance time (RMSE 0.067s) compared to motion capture.
• Estimated stride length with an RMSE of 0.117m (deep learning) and 0.132m (inverted pendulum model).
• Successfully estimated other gait parameters like duty factor and double support duration, with varied performance.

Conclusions:

• Chest acceleration data provides an efficient and non-invasive approach to gait analysis.
• This method can enhance the utility of wearable devices in clinical settings and research.
• Validated algorithms support quantifying gait and mobility while reducing patient burden and enabling cardiac signal monitoring.