Influence of the feature space on the estimation of hand grasping force from intramuscular EMG

https://doi.org/10.1016/j.bspc.2012.05.002Get rights and content

Abstract

The study compares the performance of different combinations of nine features extracted from intramuscular electromyogram (EMG) recordings for the estimation of grasping force within the range 0–100% maximum voluntary contraction (MVC). Single-channel intramuscular EMG was recorded from the flexor digitorum profundus (FDP) muscle from 11 subjects who exerted three force profiles during power grasping. The ability of the features to estimate force with a 1st order polynomial (poly1) and an artificial neural network (ANN) model was assessed using the adjusted coefficient of determination (R2). Willison amplitude (WAMP) and root mean square (RMS) showed the highest R2 (∼0.88) values for poly1. The performance of all the features to predict force significantly increased (P < 0.01) when an ANN was applied. In this case, the Modified Mean Absolute Value (MMAV) demonstrated the best performance (∼0.91). The results showed that a single channel intramuscular EMG recording represents the entire grasping force range (0–100% MVC) measured from the FDP muscle. The association between EMG and force depends on the features extracted and on the model.

Highlights

► We modeled intramuscular EMG–force relationship from 0 to 100% maximum voluntary contraction. ► We examined the influence of the feature space on the estimation of force. ► Five features are enough to obtain accurate force estimation. ► Advanced model through artificial neural network performed better than simple model based on first order polynomial.

Introduction

For most patients with amputations, the only possibility for restoration of movement is through the use of prosthetic devices [1]. The use of prosthetic devices has a positive effect on the quality of life of amputees, although the usability of active prostheses is still limited. For this reason, advanced control schemes have been proposed to allow the delivery of commands in an intuitive way for the users [2]. Surface electromyography (sEMG) is widely used for the control of prosthetic devices, where in most commercial devices the applied force is estimated proportionally to muscle activity. Several studies have investigated the estimation of force from features extracted from sEMG signals. These features range from the traditional ones, such as Mean Absolute Value (MAV), to parametric autoregressive models.

The relationship between sEMG features and force is usually monotonic, but both linear [3], [4] and non-linear [5], [6] associations have been observed. The use of sEMG as control signal for prostheses has some limitations: (1) sEMG can only be measured from superficial muscles, (2) it is sensitive to crosstalk, and (3) surface EMG electrodes can cause irritation of the skin during repeated use [7], [8]. In the attempt to overcome these limitations, the application of intramuscular EMG signals for the control of prosthetic devices has been proposed. Furthermore, intramuscular electrodes may provide possibilities for chronic implants. However, intramuscular EMG is invasive; and because of their high selectivity, intramuscular signals may be less representative of the global muscle activity and thereby contain less information about the level of force produced by the muscle.

Only few studies have investigated the use of intramuscular recordings in relation to estimation of muscle force or hand grasp force, and thereby only a limited number of features in a limited force range have been investigated for intramuscular EMG compared to sEMG. Kamavuako et al. [8] has shown that there is a high correlation (R  0.9) between the global discharge rate feature of intramuscular recordings and hand grasp force. However, in that study the grasping force was limited to 50 N and the EMG signals were recorded from a wrist extensor muscle, which is not a primary muscle for generating grasping force. For a limited range of knee extension forces, Onishi et al. [9] showed a coefficient of determination (R2) above 0.85 between the integrated intramuscular EMG and force. In a previous study [10] we compared four features using three models (linear, exponential and artificial neural network), however in that study, the force level was also limited to 50 N using a wrist extensor. Furthermore performance based on combining different features was not investigated.

In this study, we build upon the previous study [10] by measuring intramuscular EMG signals from the flexor digitorum profundus (FDP) muscle, which is a primary finger flexor, and the association between features (and their combinations) of intramuscular EMG and grasping force for the complete force range from 0 to 100% maximum voluntary contraction (MVC). Although maximal forces are not usually required for controlling prosthetic devices (up to 100 N in i-Limb pulse and Bebionic hands for example [11]), the assessment of the relation between EMG and force in a broad range is needed to verify that one electrode location is sufficient to represent muscle force. The aim of the study was to compare the predictive capabilities of different combinations of features of the EMG and models of association with force over the full force range.

Section snippets

Subjects

The experiments were conducted on 11 able-bodied human subjects (4 w/7 m, age range: 22–26 yr, mean 23.8 yr). The procedures were in accordance with the Declaration of Helsinki and approved by the Danish local ethical committee (approval no.: N-20080045). Subjects provided their written informed consent prior to the experimental procedures. The subjects had no history of upper extremity or other musculoskeletal disorders.

Experimental procedures

The subjects exerted hand grasping (power grasp) force while seated in a

Results

The average MVC force was 481 ± 69 N and the range of R2 across features and subjects was 0.53–0.97 (median = 0.87) and 0.75–0.98 (median = 0.89) for the poly1 and ANN models, respectively. When using poly1, a significant difference (P < 0.001) was found between the nine features (mean R2, range: 0.79–0.89) with WAMP and RMS (0.89 ± 0.01) having the highest R2 values (Fig. 4). Thus, the degree of estimated linearity between intramuscular EMG and force depended on the feature. There was also a significant

Discussion

The results showed that grasping force could be estimated with high accuracy using features extracted from single channel intramuscular EMG for the full force range. The performance depended on the selected feature and model.

Conclusions

This study compared the combination of nine features extracted from intramuscular EMG signals and showed that the degree of performance in terms of force estimation depends on the selected feature and on the model used to describe the association. The study also showed that features extracted from a single channel intramuscular EMG signal measured from FDP muscle may be used to represent the grasping force within the entire force range (0–100% MVC).

Acknowledgements

This study was supported by a grant from the Danish Agency for Science, Technology and Innovation (Council for Independent Research|Technology and Production Sciences, Grant number 10-080813) (ENK) and by the ERC Advanced Research Grant DEMOVE (“Decoding the Neural Code of Human Movements for a New Generation of Man-machine Interfaces”; no.: 267888) (DF).

References (22)

  • J.W. Clark

    Medical Instrumentation—Application and Design

    (1998)
  • Cited by (0)

    View full text