Muscle force estimation from lower limb EMG signals using novel optimised machine learning techniques

Area of Science:

• Biomedical Engineering
• Rehabilitation Robotics
• Signal Processing

Background:

• Electromyography (EMG) signals are crucial for understanding muscle activity.
• Accurate processing of lower limb EMG is essential for effective robotic rehabilitation.
• Existing methods require optimization for enhanced muscle force estimation.

Purpose of the Study:

• To develop a robust framework for processing and evaluating lower limb EMG signals for a knee rehabilitation robot.
• To enhance the accuracy of muscle force estimation using machine learning.
• To integrate real-time measurements of muscle force and joint angle.

Main Methods:

• Designed and constructed a knee rehabilitation robot utilizing surface EMG (sEMG) signals.
• Employed machine learning techniques including Support Vector Machine (SVM), Support Vector Regression (SVR), and Random Forest (RF) for muscle force estimation.
• Utilized Genetic Algorithm (GA) for parameter optimization and feature extraction.
• Integrated a load cell and an Inertial Measurement Unit (IMU) for force and angle measurements.

Main Results:

• Achieved a high muscle force estimation accuracy of 98.67% for lower limb muscles.
• Demonstrated the effectiveness of GA in improving the accuracy of SVM, SVR, and RF models.
• Validated the system's performance through extensive experiments and comparisons.

Conclusions:

• The proposed framework provides accurate processing of lower limb EMG signals for robotic rehabilitation.
• The integration of machine learning and GA optimization significantly enhances muscle force estimation.
• The system's high accuracy promises improved therapeutic outcomes in knee rehabilitation.