Knee Joint Torque Estimation in Femoral Prosthesis Users Using MMG and Low-Frequency Acceleration

Abstract

In prosthetic manipulation and rehabilitation, accurately correlating biological signals with joint torque is essential. The mechanomyogram (MMG), a readily measurable biological signal, has been extensively studied for this purpose. Torque estimation is possible with relatively high accuracy during an open kinematic chain (OKC) in which the foot moves freely, but the problem is that estimation accuracy deteriorates when the toe comes into contact with the ground in a closed kinematic chain (CKC).This study introduces a method to estimate knee joint torque during dynamic movements, including CKC motion, by utilizing acceleration signals below 6 Hz and MMG signals. The method was validated in a femoral prosthesis user. Knee joint torque was predicted using six-dimensional features combining four-channel MMG (20–100 Hz) and sub-6 Hz acceleration signals along the x- and z-axes. Support Vector Regression (SVR) was used for regression analysis. Tasks included walking, transitioning from walking to standing, and rising from a chair. Results showed that incorporating low-frequency acceleration signals, in addition to MMG, enhanced the mean coefficient of determination in healthy subjects and reduced the root mean square error (RMSE) in CKC motion. The method proved beneficial for femoral prosthesis users, significantly advancing knee joint torque estimation and suggesting that femoral amputees may effectively utilize their residual functionality. The inclusion of low-frequency acceleration signals significantly contributed to the robustness of torque estimation.

Keywords

• Human-machine interface (HMI)
• joint torque estimation
• knee torque
• mechanomyogram
• motor intention prediction (MIP)