Grasp force estimation from the transient EMG using high-density surface recordings

Abstract

Objective. Understanding the neurophysiological signals underlying voluntary motor control and decoding them for prosthesis control are among the major challenges in applied neuroscience and bioengineering. Usually, information from the electrical activity of residual forearm muscles (i.e. the electromyogram, EMG) is used to control different functions of a prosthesis. Noteworthy, forearm EMG patterns at the onset of a contraction (transient phase) have shown to contain predictive information about upcoming grasps. However, decoding this information for the estimation of grasp force (GF) was so far overlooked. Approach. High density-EMG signals (192 channels) were recorded from twelve participants performing a pick-and-lift task. The final GF was estimated offline using linear regressors, with four subsets of channels and ten features obtained using three channels-features selection methods. Two different evaluation metrics (absolute error and R 2), complemented with statistical analysis, were used to select the optimal configuration of the parameters. Different windows of data starting at the GF onset were compared to determine the time at which the GF can be ascertained from the EMG signals. Main results. The prediction accuracy improved by increasing the window length from the moment of the onset and kept improving until the steady state at which a plateau of performances was reached. With our methodology, estimations of the GF through 16 EMG channels reached an absolute error of 2.52% the maximum voluntary force using only transient information and 1.99% with the first 500 ms of data following the onset. Significance. The final GF estimation from transient EMG was comparable to the one obtained using steady state data, confirming our hypothesis that the transient phase contains information about the final GF. This result paves the way to fast online myoelectric controllers capable of decoding grasp strength from the very early portion of the EMG signal.

Characterization of Forearm Muscle Activation in Duchenne Muscular Dystrophy via High-Density Electromyography: A Case Study on the Implications for Myoelectric Control

“…Every 3-s contraction was further segmented in 1-s segments by keeping only the middle second of the contraction (steadystate phase) and discarding the transient phase (31). For every electrode, the average of this 1-s contraction was calculated and used to construct 10 heatmaps per gesture (Figure 2).…”

Automated Channel Selection in High-Density sEMG for Improved Force Estimation

“…Channel selection and dimensionality reduction have been widely used in classification studies of bio-signals to provide faster processing and improved model performances [ 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 ]. Geng et al [ 27 ] proposed a novel channel selection approach called multi-class common spatial pattern (MCCSP) for channel selection in EMG pattern-recognition-based movement classification.…”

“…A feature selection algorithm with backward elimination was used for channel selection, while the mean detection accuracy only experienced a slight decrease. Several studies have focused on channel selection in EMG applications [ 31 , 32 , 33 ]. Martinez et al [ 31 ] recorded HD-EMG signals to estimate grasp force.…”

“…Several studies have focused on channel selection in EMG applications [ 31 , 32 , 33 ]. Martinez et al [ 31 ] recorded HD-EMG signals to estimate grasp force. They examined subsets of 4, 8, and 16 channels, in comparison to the full set of 168 channels, to determine the minimum amount of information needed for grasp force prediction.…”

“…They examined subsets of 4, 8, and 16 channels, in comparison to the full set of 168 channels, to determine the minimum amount of information needed for grasp force prediction. Three channel reduction methods were used: selecting channels in the center of the grid only (Fix-Ridge); selecting the Fix-Ridge channels with feature selection using elastic nets analysis (Fix-EN); and selecting channels using lasso with EN feature selection (LassoG-EN) in [ 31 ]. The Fix-EN was selected as the best method, since there was no statistical difference between methods, and 16 channels proved to have the best trade-off between complexity and performance [ 31 ].…”