Prediction of grasping force based on features of surface and intramuscular EMG

Myoelectric prosthetic devices can be controlled based on surface electromyography (sEMG). However, intramuscular EMG (iEMG) has been proposed as an alternative control signal, since it may provide more stable and selective recordings with several advantages such as electrode implantation. An earlier study quantified the relationship between iEMG and grasping force, however, this was solely based on one feature and force ranging from 0-50 N. The present study quantified the linear relationship between grasping force and 14 different EMG features using the entire force range from 0 to 100 % maximum voluntary contraction (MVC) and assessed their predictive capabilities. Single-channel iEMG and sEMG were recorded concurrently from the muscle flexor digitorum profundus (FDP) from 11 subjects who exerted four force profiles during power grasping. The predictive capability of all the features was assessed using R2 with a 1st order polynomial and an Artificial Neural Network (ANN). Wilson Amplitude (WAMP) showed the best results for both sEMG and iEMG in linear relationship and linear prediction (R̄2 > 0.9), with no significant difference between the two signals for linear prediction. For the ANN, constraint sample entropy (CSE) and WAMP showed the best results for iEMG and sEMG, respectively (R̄2 > 0.9). No significant difference was found for the ANN between the two signals. Furthermore, there was no significant difference between linear prediction and ANN. Thus, the choice of prediction model (linear prediction or ANN) did not play a significant role. These results indicate that iEMG can be used for force prediction with a 1st order polynomial and thus in proportional control (0-100 % MVC) with same accuracy as for sEMG.