Elsevier

Clinical Neurophysiology

Volume 114, Issue 9, September 2003, Pages 1580-1593
Clinical Neurophysiology

EMG contamination of EEG: spectral and topographical characteristics

https://doi.org/10.1016/S1388-2457(03)00093-2Get rights and content

Abstract

Objective: Electromyogram (EMG) contamination is often a problem in electroencephalogram (EEG) recording, particularly, for those applications such as EEG-based brain–computer interfaces that rely on automated measurements of EEG features. As an essential prelude to developing methods for recognizing and eliminating EMG contamination of EEG, this study defines the spectral and topographical characteristics of frontalis and temporalis muscle EMG over the entire scalp. It describes both average data and the range of individual differences.

Methods: In 25 healthy adults, signals from 64 scalp and 4 facial locations were recorded during relaxation and during defined (15, 30, or 70% of maximum) contractions of frontalis or temporalis muscles.

Results: In the average data, EMG had a broad frequency distribution from 0 to >200 Hz. Amplitude was greatest at 20–30 Hz frontally and 40–80 Hz temporally. Temporalis spectra also showed a smaller peak around 20 Hz. These spectral components attenuated and broadened centrally. Even with weak (15%) contraction, EMG was detectable (P<0.001) near the vertex at frequencies >12 Hz in the average data and >8 Hz in some individuals.

Conclusions: Frontalis or temporalis muscle EMG recorded from the scalp has spectral and topographical features that vary substantially across individuals. EMG spectra often have peaks in the beta frequency range that resemble EEG beta peaks.

Significance: While EMG contamination is greatest at the periphery of the scalp near the active muscles, even weak contractions can produce EMG that obscures or mimics EEG alpha, mu, or beta rhythms over the entire scalp. Recognition and elimination of this contamination is likely to require recording from an appropriate set of peripheral scalp locations.

Introduction

Contamination of the electroencephalogram (EEG) signal by muscle activity is an old and well-recognized problem for clinical and experimental electroencephalography (Barlow, 1986). Electromyogram (EMG) artifact is of greatest concern for research studies or clinical applications that depend on automatic online detection and measurement of EEG features. EEG-based brain–computer interface (BCI) systems fall into this group. BCI systems allow their users to communicate with others or control devices (such as a simple word-processing program) by controlling specific features of the EEG (Wolpaw et al., (2002) for review). Thus, they require either artifact-free EEG or the capacity to detect artifacts in real-time and prevent them from affecting communication and control. This requirement is particularly important because BCI systems are intended for individuals with severe motor disorders such as amyotrophic lateral sclerosis (ALS) and cerebral palsy that may be associated with frequent involuntary contractions of cranial or facial muscles.

Several present-day BCI systems rely on mu (8–12 Hz) or beta (18–24 Hz) rhythms recorded from the scalp over sensorimotor cortices (e.g. Pfurtscheller et al., 2000, Wolpaw et al., 2000, Kostov and Polak, 2000). Mu- or beta-rhythm amplitudes are translated into cursor movements on a video screen. The system user learns to control these rhythms so as to move the cursor to select a desired item from among a set of choices. Even though these rhythms are typically recorded from central locations, they might still be obscured by EMG. EMG contamination could be a particular problem for BCI users in whom ALS has led to degeneration of sensorimotor cortex. In such individuals, rhythms from less affected frontal or parietal cortex (Hudson et al., 1993, Sasaki and Iwata, 2000) may prove better for BCI application. Because these rhythms are recorded from scalp locations closer to facial or temporal muscles, their recording is particularly susceptible to EMG contamination. Unrecognized EMG contamination can mimic actual EEG control and thereby mislead and otherwise impede research aimed at improving these BCI systems. Thus, continued productive development and eventual practical application of these systems to the needs of those with severe motor disabilities require the ability to recognize and eliminate EMG contamination in real-time during online operation.

In theory, a variety of methods might be used to reduce or eliminate EMG contamination. These include relatively simple methods such as linear or non-linear low-pass filtering (Barlow, 1986, Ives and Schomer, 1988, Panych et al., 1989, Sadasivan and Dutt, 1995, Klass, 1995) or rejection of EEG segments that exceed a predefined amplitude threshold (Brunner et al., 1996, Anderer et al., 1999, Junghofer et al., 2000) and more sophisticated methods such as factor decomposition using principal component (Lagerlund et al., 1997) or independent component analysis (Jung et al., 2000). Successful application of any of these methods requires detailed knowledge of the spectral and topographical patterns of cranial EMG contamination. This information is not available at present.

It is known that EMG of skeletal muscles recorded from the skin has a broad frequency distribution from 0 to >200 Hz with several more or less distinct spectral components (Farmer et al., 1993, McAuley et al., 1997, Halliday et al., 1998, Hari and Salenius, 1999, Marsden et al., 1999, Mima and Hallett, 1999, Brown, 2000). These include a 0–5 Hz component thought to reflect a common drive to the motor units, a 10 Hz component thought to reflect motor unit firing and physiological tremor, a 20–30 Hz component (the so-called EMG beta rhythm), and a 35–60 Hz component (the Piper rhythm). EMG from facial muscles also shows a broad frequency distribution from 0 Hz up, with two peaks corresponding to the EMG beta and Piper rhythms (Van Boxtel et al., 1983, Van Boxtel et al., 1984, Van Boxtel, 2001).

Current understanding of EMG distribution over the scalp is largely qualitative rather than quantitative. Studies to date have been limited to a few electrode locations, have included data from only a few subjects, have not controlled the strength of muscle contraction, and/or have not included statistical analyses (O'Donnell et al., 1974, Lee and Buchsbaum, 1987, Friedman and Thayer, 1991, Willis et al., 1993). It is known that EMG activity affects alpha, beta, and even delta frequency bands (Barlow, 1986, Klass, 1995, Brunner et al., 1996). However, most automated methods for EMG artifact detection and elimination incorporate the assumption that EMG recorded from the scalp has a broad spectral distribution that begins at 15–20 Hz, and thus can obscure only higher frequency EEG activity (Gotman et al., 1975; Chiappa, 1986; Lee and Buchsbaum, 1987; Sadasivan and Dutt, 1995).

The purpose of the present study was to facilitate the continued development of BCI technology by obtaining from a representative group of adults a detailed quantitative description of the spectral and topographical distributions over the scalp of EMG from frontalis and temporalis muscles. We focused on these two muscles because they are the most common sources of EMG over frontal and central head regions (Barlow, 1986, Klass, 1995). Since the detection and elimination of EMG contamination must ultimately be achieved in each individual BCI user, the goal was to describe both the average results from the entire group of subjects and the range of variation across individuals.

Section snippets

Subjects

Data were collected from 25 healthy adults, 13 women and 12 men (16–53 years, mean 35). None had a history of a neurological or psychiatric disorder and none was on chronic medication. Most had previously participated in BCI studies and were familiar with the experimental environment. The study was approved by the New York State Department of Health Institutional Review Board and informed consent was obtained from all participants. Lateralization of psychomotor functions was estimated from

Spectral distribution as a function of muscle contraction level

Fig. 3 presents average (i.e. all 10 subjects of Experiment 1) amplitude spectra for signals recorded from the 4 facial locations and from the anterior half of the scalp during relaxation and during 4 levels of frontalis muscle contraction. The spectra are calculated from the data as they were recorded: bipolar for EMG, and monopolar referenced to the right earlobe for EEG.

As expected, spectral amplitudes are higher at stronger contraction levels, and this increase is most prominent at anterior

Discussion

This study set out to describe the amplitude and spectral content of EMG from frontalis and temporalis muscles recorded from a comprehensive set of scalp locations. Its central purpose was to provide basic knowledge needed for the evaluation of EMG artifacts during the operation of EEG-based BCI systems and essential for the design of methods for detecting these artifacts online and minimizing their impact.

Acknowledgements

We are grateful to our subjects for their conscientious participation in the study. We thank Mr. William A. Sarnacki for technical assistance and Dr. Ann M. Tennissen for reviewing the manuscript. This work was supported by grants from the National Center for Medical Rehabilitation Research, NICHD, NIH (HD30146); from NIBIB and NINDS, NIH (EB00856); and from the ALS Hope Foundation.

References (56)

  • D.J. McFarland et al.

    Spatial filter selection for EEG-based communication

    Electroenceph clin Neurophysiol

    (1997)
  • R.D. O'Donnell et al.

    Contamination of scalp EEG spectrum during contraction of cranio-facial muscles

    Electroenceph clin Neurophysiol

    (1974)
  • L.P. Panych et al.

    Practical digital filters for reducing EMG artefact in EEG seizure recordings

    Electroenceph clin Neurophysiol

    (1989)
  • S. Sasaki et al.

    Immunocytochemical and ultrastructural study of the motor cortex in patients with lower motor neuron disease

    Neurosci Lett

    (2000)
  • D.A. Winter et al.

    Crosstalk in surface electromyography: theoretical and practical estimates

    J Electromyogr Kinesiol

    (1994)
  • J.R. Wolpaw et al.

    Brain–computer interfaces for communication and control

    Clin Neurophysiol

    (2002)
  • American EEG Society

    Guideline thirteen: guidelines for standard electrode position nomenclature

    J Clin Neurophysiol

    (1994)
  • P. Anderer et al.

    Artifact processing in computerized analysis of sleep EEG – a review

    Neuropsychobiology

    (1999)
  • J.S. Barlow

    Artifact processing (rejection and minimization) in EEG data processing

  • N.N. Bragina et al.

    Funkcionalnye asimmetrii cheloveka

    (1981)
  • D.P. Brunner et al.

    Muscle artifacts in the sleep EEG: automated detection and effect on all-night EEG power spectra

    Sleep Res

    (1996)
  • F. Cecere et al.

    Is quantitative electromyography reliable?

    J Orofac Pain

    (1996)
  • G.T. Clark et al.

    Co-activation of sternocleidomastoid muscles during maximum clenching

    J Dent Res

    (1993)
  • G. Cohen

    Theoretical interpretations of lateral asymmetries

  • B.A. Conway et al.

    On the relation between motor-unit discharge and physiological tremor

  • S.F. Farmer et al.

    The frequency content of common synaptic inputs to motoneurons studies during voluntary isometric contraction in man

    J Physiol

    (1993)
  • J. Gibbs et al.

    Organization of inputs to motoneurone pools in man

    J Physiol

    (1995)
  • R. Hari et al.

    Rhythmical corticomotor communication

    NeuroReport

    (1999)
  • Cited by (524)

    View all citing articles on Scopus
    View full text