Continuous Estimation of Knee Joint Angle Based on Surface Electromyography Using a Long Short-Term Memory Neural...

Area of Science:

• Biomedical Engineering
• Robotics
• Signal Processing

Background:

• Surface electromyography (sEMG) signals are crucial for estimating joint angles in human-machine interfaces like exoskeletons.
• Accurate joint angle estimation is vital for improving man-machine coordination and the overall performance of exoskeleton systems.
• Existing methods using root mean square (RMS) features of sEMG have limitations in capturing complex joint dynamics.

Purpose of the Study:

• To propose and evaluate a novel time-advanced feature (RMSTAF) combined with RMS for sEMG signals to estimate knee joint angles.
• To compare the performance of a Long Short-Term Memory (LSTM) network utilizing the proposed RRTAF feature against other methods.
• To enhance the accuracy and reliability of knee joint angle estimation for improved exoskeleton control.

Main Methods:

• Utilized Long Short-Term Memory (LSTM) network with a combined root mean square (RMS) and time-advanced feature (RMSTAF) of sEMG signals, termed RRTAF.
• Collected sEMG data from eight leg muscles (rectus femoris, biceps femoris, semitendinosus, gracilis, semimembranosus, sartorius, medial gastrocnemius, tibialis anterior) of five healthy subjects.
• Compared the proposed LSTM-RRTAF method with LSTM-RMS, BPNN-RRTAF, and BPNN-RMS using root mean square error (RMSE) and cross-correlation coefficient (ρ).

Main Results:

• The LSTM-RRTAF method demonstrated superior performance, achieving average RMSE reductions of 8.57%, 46.62%, and 68.69% compared to LSTM-RMS, BPNN-RRTAF, and BPNN-RMS, respectively.
• The average cross-correlation coefficient (ρ) values were increased by 0.31%, 4.15%, and 18.35% with the LSTM-RRTAF method, indicating better agreement between estimated and actual knee joint angles.
• The inclusion of the time-advanced feature in the sEMG signal processing significantly improved the accuracy of knee joint angle estimation.

Conclusions:

• The proposed time-advanced feature (RMSTAF) integrated into the RRTAF feature set significantly enhances knee joint angle estimation accuracy.
• LSTM networks combined with RRTAF offer superior performance for real-time knee joint motion estimation compared to traditional methods.
• This advancement holds promise for developing more intuitive and effective human-exoskeleton interactions and control systems.