Accurate Ambulatory Gait Analysis in Walking and Running Using Machine Learning Models.

Area of Science:

• Biomechanics and Wearable Technology
• Machine Learning in Healthcare
• Gait Analysis and Rehabilitation

Background:

• Traditional gait analysis methods are often confined to laboratory settings, limiting real-world applications.
• Existing wearable sensors for gait analysis exhibit moderate accuracy, hindering widespread adoption.
• Instrumented insoles offer a portable solution, but require accurate data processing techniques.

Purpose of the Study:

• To evaluate the efficacy of Support Vector Regression (SVR) models for accurate gait parameter estimation using instrumented insoles.
• To assess the robustness of SVR models against inter-subject variability in gait analysis.
• To demonstrate the potential of machine learning to enhance the accuracy of wearable gait analysis systems.

Main Methods:

• Development of custom instrumented insoles (SportSole) for gait data collection.
• Application of Support Vector Regression (SVR) models to analyze gait parameters from insole data.
• Validation of SVR model accuracy against reference laboratory equipment in 14 healthy subjects during walking and running.

Main Results:

• SVR models achieved excellent intraclass correlation coefficients (ICC) for stride length, velocity, and foot clearance.
• Mean absolute errors (MAE) were notably low: 1.37% for stride length, 1.23% for velocity, and 2.08% for foot clearance during walking.
• MAE% for running were also low: 2.59% for stride length, 2.91% for velocity, and 5.13% for foot clearance, demonstrating high accuracy.

Conclusions:

• Support Vector Regression (SVR) models significantly enhance the accuracy of gait parameter estimation from wearable insoles.
• Machine learning regression offers a robust and accurate approach for real-time gait analysis in unconstrained environments.
• This study validates the potential of SVR for improving the reliability and applicability of wearable gait analysis technologies.