Estimation of Electrically-Evoked Knee Torque from Mechanomyography Using Support Vector Regression.

Area of Science:

• Biomechanics
• Computational Intelligence
• Rehabilitation Engineering

Background:

• Estimating joint torque during neuromuscular electrical stimulation (NMES) is challenging for physical therapy and exercise.
• Recent research explores torque estimation from muscle characteristics like mechanomyography (MMG).

Purpose of the Study:

• To investigate the accuracy of a computational intelligence technique for estimating NMES-evoked knee extension torque.
• To assess the feasibility of using Mechanomyography (MMG) signals for real-time torque estimation.

Main Methods:

• Support Vector Regression (SVR) was used to model knee torque estimation.
• Inputs included MMG amplitude, electrical stimulation intensity, and knee angle.
• A Gaussian kernel function with optimized parameters was applied for SVR modeling.

Main Results:

• The SVR model achieved high accuracy, with R² values up to 94% (training) and 89% (testing).
• Root mean square errors (RMSE) were 9.48 (training) and 12.95 (testing).
• Estimated torque values closely matched experimental data from an isokinetic dynamometer.

Conclusions:

• Support Vector Regression (SVR) provides an accurate method for estimating knee joint torque using Mechanomyography (MMG) signals during NMES.
• This approach supports the development of closed-loop NMES systems for functional rehabilitation and exercise.
• MMG serves as a viable feedback signal source for real-time joint torque estimation in NMES applications.