Gait analysis with a custom Instrumented Insole with Embedded Machine Learning: Feasibility Study.

Area of Science:

• Biomechanics
• Wearable Technology
• Machine Learning

Background:

• Accurate human gait analysis is crucial for diagnosing and managing various neurological and musculoskeletal conditions.
• Current gait analysis methods often require specialized equipment and controlled laboratory settings, limiting their accessibility.
• There is a need for portable, real-time gait monitoring solutions suitable for both clinical and everyday environments.

Purpose of the Study:

• To develop and evaluate a wireless instrumented insole for real-time human gait analysis.
• To embed machine learning models for estimating Center of Pressure (CoP) and identifying eight gait phases (8GP) directly onto the insole's processor.
• To assess the system's performance in both clinical and non-clinical settings.

Main Methods:

• Development of a wireless instrumented insole with an embedded processor (ESP32 microcontroller).
• Implementation of machine learning models (decision trees for 8GP, ANNs for CoP and vertical ground reaction force - vGRF) trained using subject-specific and cross-subject strategies.
• Validation with three volunteers performing walking tasks, with data compared against a force platform.
• Real-time data transmission via Wi-Fi to a host PC for analysis.

Main Results:

• The system achieved real-time gait parameter estimation at a 96 Hz sampling rate with fully embedded processing.
• Subject-specific models yielded an average F1-score of 0.74 for gait phase classification, while general models achieved 0.61.
• Artificial neural networks demonstrated high accuracy in estimating CoP (R²=0.94 for xCoP, R²=0.93 for yCoP) and vGRF (R²=0.78).

Conclusions:

• The developed wireless instrumented insole provides an accessible and accurate solution for real-time human gait analysis.
• The system's low-cost, open-source hardware design facilitates integration with rehabilitation technologies and use in diverse environments, including daily living.
• This technology holds significant potential for remote patient monitoring, clinical diagnostics, and personalized rehabilitation strategies.