SEMG-based prediction of masticatory kinematics in rhythmic clenching movements

https://doi.org/10.1016/j.bspc.2015.04.003Get rights and content

Highlights

  • EMG signals of two masseter and temporalis muscles are used to predict clenching movements.

  • GA is employed to find optimal number of neurons in the hidden layer and total duration of delays.

  • Validity of the proposed models is experimentally demonstrated.

  • The performance of AR-TDANN is better than that of TDANN.

  • The TDANN would be sufficiently efficient for controlling masticatory robots.

Abstract

This paper investigated the ability of a hybrid time-delayed artificial neural network (TDANN)/autoregressive TDANN (AR-TDANN) to predict clenching movements during mastication from surface electromyography (SEMG) signals. Actual jaw motions and SEMG signals from the masticatory muscles were recorded and used as output and input, respectively. Three separate TDANNs/AR-TDANNs were used to predict displacement (in terms of position/orientation), velocity, and acceleration. The optimal number of neurons in the hidden layer and total duration of delays were obtained for each TDANN/AR-TDANN and each subject through a genetic algorithm (GA). The kinematic modeling of a human-like masticatory robot, based on a 6-universal-prismatic-spherical parallel robot, is described. The structure and motion variables of the robot were determined. The closed-form solution of the inverse kinematic problem (IKP) of the robot was found by vector analysis. Thereafter, the framework for an EMG-based human mastication robot interface is explained. Predictions by AR-TDANN were superior to those by TDANN. SEMG signals from mastication muscles contained important information about the mandibular kinematic parameters. This information can be employed to develop control systems for rehabilitation robots. Thus, by predicting the subject's movement and solving the IKP, we provide applicable tools for EMG-based masticatory robot control.

Introduction

The significance of the chewing process on digestion and health necessitates studies of the mastication system. Mastication in humans consists of two basic movements, clenching and grinding. For clenching, the mandible moves in the sagittal plane; for grinding, it traces a circular path in the frontal plane. More than 20 muscles are involved in the process of human mastication, with six of them playing the major role in mandible control during coordinated masticatory movements [40], [45]. These muscles are the temporalis muscles, attached from the side of the skull to the top of the mandible; the masseter muscles, attached between the cheek and the lower rear section of the mandible; the medial pterygoid muscles, attached to the inside of the skull and the mandible; the lateral pterygoid muscles, horizontally attached between the skull and the mandible; and the digastric muscle, attached between the skull and the chin. The masseter and temporalis muscles are primarily employed during clenching, whereas the pterygoid muscles have their main role during grinding. During jaw closing, the mandible is elevated by the temporalis and masseter muscles, while it is protruded by the masseter muscles. Pterygoid muscles protrude the mandible, produce its side-to-side movements, and generate the grinding motion.

Researchers have utilized various methods to study the chewing process, including gnathosonics for measuring the sounds of mastication, ultra-high-speed cinematography for measuring mandibular movement and velocity, and laryngophone for monitoring swallowing [15]. Small markers or magnets have been used to record the chewing trajectory [43], [17]. Electromyography (EMG) has been employed to study changes in the electrical activity of the muscles during mastication [15], [36], [26], [20]. Changes in EMG parameters are better able to assess the sensory characteristics than mechanical measurements [15]. Experimentally obtained signals, together with the physiological cross-sectional area of the muscles, have been employed to estimate instantaneous muscle forces [4], [31] or to differentiate food-texture characteristics [15]. Additionally, EMG has been used to identify differences in chewing patterns between individuals and to classify individuals into groups according to their chewing efficiency [21], [5], [6], [7].

Recently, EMG signals recorded from the muscles have been applied to control and classify the motion of prosthetic limbs [14], wheelchairs, and teleoperated robots [8], [30], [16], [46], [13], [38], [19]. Indeed, there have been various attempts to classify different movements for the purpose of controlling prostheses. Methods that have been applied to classify motions and categorize the mastication process include autoregressive (AR) models [16], Bayesian classifiers [22], artificial neural networks [25], [18], [41], [42], fuzzy neural networks [3], dynamic recurrent neural networks [10], probabilistic neural networks [13], [9], and Bayesian networks [8]. For instance, Graupe et al. [16] used an AR model to extract features from EMG signals and determine motions [47]. Au et al. [2] found that a time-delayed artificial neural network (TDANN) was capable of predicting shoulder and elbow motions from only EMG signals.

EMG signals are nonstationary. The feature patterns vary significantly depending on the tasks and conditions of the users. In addition, EMG signals are very likely to be affected by artifacts and noise. In practical applications, it is difficult to achieve sufficient accuracy and stable performance of motion classification when using only EMG signals. This achievement requires the careful operation of a prosthetic device or human-assisting manipulator. However, predicting the trajectory can be very helpful for guiding prosthetic devices in space. Toward this endeavor, a dentist-guided masticatory robot was recently developed for training patients with jaw disorders [39].

This study aimed to predict the kinematic parameters of motion during jaw opening and closing using surface EMG (SEMG) signals. Because surface electrodes are only able to record the electrical activities of the bilateral masseter and temporalis muscles, this paper focused on clenching movement. For motion prediction, a hybrid TDANN/auto-regressive TDANN (AR-TDANN) was developed from the experimental results. Predicted trajectory parameters were applied in a case study, to solve the inverse kinematic problem (IKP) and to estimate variations in the actuators (muscles) during jaw opening/closing. These findings will provide tools to control masticatory robots using SEMG signals.

The rest of the paper is organized as follows. The experimental setup, protocol, and hybrid algorithm are explained in Section 2. Kinematics of the masticatory robot and the hybrid motion framework for the SEMG-based human mastication robot interface are discussed in Section 2.4. The performance of the TDAAN/AR-TDANN in predicting the kinematic parameters of motion is presented in Section 3. This section also provides the results from solving the IKP problem, based on the predicted time-varying moments for one subject. A discussion and some remarks on future works are presented in Sections 4 Discussion, 5 Conclusion.

Section snippets

Experimental setup

An 8-channel EMG system, with a sampling rate of 1 kHz, was used for recording the electrical activity of muscles (SEMG). For each subject, SEMG signals were recorded from four muscles: namely, the bilateral masseter and temporalis muscles (Fig. 1a). Surface electrodes were placed ∼2 cm apart, oriented parallel to the muscle fibers, between the belly of each muscle and its end. Recorded raw SEMG signals were passed through a bandpass (15–400 Hz) 3rd order Butterworth filter [12]. A notch filter (50

Results

Sample SEMG (before preprocessing) and kinematic data obtained from one subject are shown in Fig. 8. During each trial, the subject opened and closed his jaw in the pre-specified time interval. Due to the symmetric role of the aforementioned muscles in the defined task, the data obtained from one temporalis and one masseter muscle were used for processing steps.

Discussion

The aim of this study was to develop an SEMG-based model by means of which jaw opening and closing during the mastication process can be reproduced. The results presented above indicate that if a gross prediction of the jaw motion is desired a TDANN can predict the trajectory with reasonable accuracy using only SEMG signals. The predictions are less accurate during maximum jaw opening. Accuracy compensation can be obtained by using the AR-TDANN structure (Fig. 9).

Other nonlinear system

Conclusion

EMG has been widely used as a control command for prostheses and powered exoskeleton robots. This paper considered whether recorded SEMG signals from voluntary muscles are sufficient for predicting the kinematics (position/orientation, velocity, and acceleration) of mandible motion in clenching movements. Two different methods, TDANN and AR-TDANN, were proposed. In general, the results showed that: (1) the TDANN and AR-TDANN methods are capable of providing reasonably accurate estimations of

Acknowledgement

The work described in this paper was supported by a grant from the research office of Ferdowsi University of Mashhad, Mashhad, Iran.

References (47)

  • H.J. Smit et al.

    Does prolonged chewing reduce food intake? Fletcherism revisited

    Appetite

    (2011)
  • C.W. Spoor

    Explanation, verification and application of the helical axis error propagation formulas

    Hum. Mov. Sci.

    (1984)
  • D. Staudenmann et al.

    Methodological aspects of SEMG recordings for force estimation – a tutorial and review

    J. Electromyogr. Kinesiol.

    (2010)
  • E.M. Wilson et al.

    The development of jaw motion for mastication

    Early Hum. Dev.

    (2009)
  • U. Posselt

    Movement areas of the mandible

    J. Pros. Dent.

    (1957)
  • R.L. Airoldi et al.

    Precision of the jaw tracking system Jaws 3D

    J. Orofac. Pain

    (1994)
  • A.T.C. Au et al.

    EMG-based prediction of shoulder and elbow kinematics in able-bodied and spinal cord injured individuals

    IEEE Trans. Rehabil. Eng.

    (2000)
  • A. Balbinot et al.

    A neuro-fuzzy system for characterization of arm movements

    Sensors

    (2013)
  • J. Barbenel

    The mechanics of the temporomandibular joint – a theoretical and electromyographical study

    J. Oral Rehabil.

    (1974)
  • N. Bu et al.

    hybrid motion classification approach for EMG-based human–robot interfaces using Bayesian and neural networks

    IEEE Trans. Robot.

    (2009)
  • N. Bu et al.

    EMG-based motion discrimination using a novel recurrent neural network

    Int. J. Intell. Syst.

    (2003)
  • G. Cheron et al.

    A dynamic neural network identification of electromyography and arm trajectory relationship during complex movements

    IEEE Trans. Biomed. Eng.

    (1995)
  • O. Fukuda et al.

    A human-assisting manipulator teleoperated by EMG signals and arm motions

    IEEE Trans. Robot. Autom.

    (2003)
  • Cited by (20)

    • Toward a bio-inspired rehabilitation aid: sEMG-CPG approach for online generation of jaw trajectories for a chewing robot

      2019, Biomedical Signal Processing and Control
      Citation Excerpt :

      Recorded data were preprocessed using Simi Motion software (www.simi.com). The experimental details for trajectory recording are provided in our previous publications [10]. The recorded trajectories were later used to define trajectories of the moving platform of the robot.

    View all citing articles on Scopus
    View full text