WO2012049362A1

WO2012049362A1 - A projection method and system for removing muscle artifacts from signals based on their frequency bands and topographies

Info

Publication number: WO2012049362A1
Application number: PCT/FI2011/050867
Authority: WO
Inventors: Hanna MÄKI; Risto Ilmoniemi
Original assignee: Aalto University Foundation
Priority date: 2010-10-13
Filing date: 2011-10-07
Publication date: 2012-04-19

Abstract

The present invention is a method for separating the contribution of muscle activity from signals of interest recorded from the brain or other parts of the nervous system. The muscle activity is often present in a frequency band which differs from the frequency band of the signal of interest, but these frequency bands may overlap. Because of this overlap, frequency filtering alone is not adequate in removing the muscle artifacts from the signals of interest. However, it is sometimes possible to filter the signal in such a way that only contribution from the muscle artifact remains. When the signals are measured with multiple channels, the topographies of the muscle-artifact-containing filtered signal can be determined and projected out of the original data. If the signals in the frequency band present in the filtered signal originate in the same muscles as the rest of the muscle signals, they tend to have similar topographies so that the projection can remove all or most of the muscle contribution.

Description

A projection method and system for removing muscle artifacts from signals based on their frequency bands and topographies

The present invention relates to a method for removing muscle artifacts from signals based on their frequency bands and topographies, according to the preamble of Claim 1.

The invention also relates to a system for removing muscle artifacts from signals based on their frequency bands and topographies.

Prior art

The brain activity measured with current methods and devices typically consist not only of signals of interest but also of signals not of interest, which are sometimes considered artifacts and sometimes otherwise disturbing. When the signals are recorded with several channels that are distributed spatially or as a function of some other experimental parameter such as angle, polarization, color, stimulation strength or other stimulation parameters, etc., the different signal sources produce a characteristic signal pattern called a topography in the dimensions of the parameters, i.e., a set of signals the relative amplitudes of which measured by each channel are time-invariant. Topographies are vectors in the signal space spanned by the signals recorded with the measurement channels; when a signal is measured with d channels, the topographies are (/-dimensional vectors and linear combinations of the measured signals. In case the topographies of the signals of interest and the artifacts differ, it is possible to project all or some of the artifact topography vectors out of the measured signals, with the effect of reducing the artifact but preserving a significant amount of the signal of interest.

For example, electroencephalography (EEG) and magnetoencephalography (MEG) can be used to record brain activity, but sometimes the recorded signals contain artifacts produced by the activity of muscles around the head. These muscle artifacts are especially problematic when combining EEG with transcranial magnetic stimulation (TMS), which, in some cases, activates the muscles directly or through nerves innervating them and, consequently, produces very large muscle artifacts. Muscle artifacts may also appear when sensory stimulation (e.g., visual, auditory, somatosensoty) or electrical stimulation such as transcranial electrical stimulation (TES; Merton and Morton, 1980), transcranial direct current stimulation (tDCS; see Priori, 2003, for a review), or transcranial alternating current stimulation (tACS; Kanai et al., 2008) is used to evoke brain activity. With TMS, it is possible to activate the brain directly and in a controlled manner with fast- changing magnetic fields (Barker et al., 1985). Combined with simultaneous EEG, which records the reactions of the brain to the stimulation, TMS provides information about cortical reactivity and connectivity (Ilmoniemi et al., 1997). Although TMS-EEG has been successfully applied in the study of several cortical regions, stimulating areas near cranial or facial muscles and recording the evoked EEG responses is challenging because of the muscle artifacts arising in the EEG signal as a result of muscle stimulation (Maki and Ilmoniemi, 2011). These artifacts are especially pronounced when stimulating lateral parts of the brain, where, for example, large part of language processing takes place, but also stimulation of areas near the neck and forehead is problematic because of the muscle artifacts. The invention is intended to eliminate at least some of the defects of the prior art disclosed above and for this purpose create a new type of method and system for enhancing measurement accuracy in connection with measurement of brain activity.

The present invention is a method for separating the contribution of muscle activity from signals of interest recorded from the brain or other parts of the nervous system. The muscle activity is often present in a frequency band which differs from the frequency band of the signal of interest, but these frequency bands may overlap. Because of this overlap, frequency filtering alone is not adequate in removing the muscle artifacts from the signals of interest. However, it is sometimes possible to filter the signal in such a way that only contribution from the muscle artifact remains. When the signals are measured with multiple channels, the topographies of the muscle-artifact-containing filtered signal can be determined and projected out of the original data. If the signals in the frequency band present in the filtered signal originate in the same muscles as the rest of the muscle signals, they tend to have similar topographies so that the projection can remove all or most of the muscle contribution. The invented method is based on projecting the topographies of the muscle activity out of the measured signals. In case of an EEG or MEG signal, a topography describes the relative signal amplitude measured by each EEG or MEG channel as a result of the activation of a certain current source.

More specifically, the measuring method according to the invention is characterized by what is stated in the characterizing portion of Claim 1.

The system according to the invention is, in turn, characterized by what is stated in the characterizing portion of Claim 14.

Considerable advantages are gained with the aid of the invention.

Measurement accuracy may be enhanced in connection with measurement of brain activity.

The following examples referring to figures l-2b explain the use of the invention in removing muscle artifacts from EEG or MEG signals, while the invention is not limited to this application.

In the figures

Figure 1 shows graphically TMS-EEG measurement results of Broca's area stimulation, which include large muscle artifacts, 1-3 orders of magnitude larger than the brain signals. These data show the need for removing the artifacts, as they mask the brain signal components that are of interest.

Figure 2a shows TMS-EEG measurement results before and after applying the method according to the invention. After applying the method according to some embodiments of the invention the artifact is largely reduced, while a significant amount of brain signal is preserved. The figure shows the global mean field amplitudes (GMFA; Lehmann and Skrandies, 1980) normalized by the value of peak Al before (grey) and after (black) the artifact removal method was applied. The relative amplitudes of the brain signal peaks B1-B5 compared to the artifact peak Al increase as a result of the artifact removal. Figure 2b: (B) Data presented in Fig. 1 after applying the artifact correction.

The brain signal consists of relatively low frequencies compared to the muscle signal; while usually no significant brain signal is present at frequencies above 100 Hz, the muscle component is easily observable up to 400-500 Hz (Clancy et al. 2002; Merletti, 1996). The spectra overlap at frequencies below 100 Hz, so lowpass filtering is not adequate to separate the contributions. On the contrary, highpass filtering the data with a suitable cutoff frequency (around 100 Hz) results in signal with predominantly high-frequency muscle activity and noise. Because the high- and low-frequency muscle components are produced in the same muscles, they are composed of similar current distributions and thus the potential differences detected on the scalp are spatially similar. The spatial distribution of the potential differences is called a topography and a topography measured with several channels (d EEG electrodes + a reference electrode or d MEG channels) can be described as a (/-dimensional vector in the signal space (Uusitalo ja Ilmoniemi, 1997). Projecting the topographies of the highpass- filtered data out of the measured signal removes not only the high-frequency component, but also the low-frequency component of the muscle activity.

The method can be expressed mathematically as follows: the (/-dimensional signal m(t) is a weighted sum of the brain and muscle activity and noise, which can be further divided into low- and high-frequency components using a frequency threshold fi . Here, Xi and yi describe the (/-dimensional time-independent topographies of the muscle and brain sources and _;(ί) and bi(t) are their time-dependent amplitudes. The measured signal is a sum of low- frequency and N¹¹ high-frequency muscle components as well as low-frequency and A/¹¹ high-frequency brain components and noise n(t):

m (t) = b« (t) yf + n (t) . ( 1 )

By choosing so that the high-frequency brain signal is negligible ^^.^ b (f) yf « 0J, highpass filtering the signal with cutoff frequency f_t produces a signal which is mainly composed of high-frequency muscle activity and noise:

H(m(t)) ~∑a? (t) x? +H(n(t)) , (2) i=\

where H is the highpass-filter operator. If the low-frequency muscle components belong to the signal-subspace spanned by the high-frequency muscle components

}e span( ,..., JC" )), projecting out the topographies of the highpass-filtered data also removes the low-frequency muscle components.

The topographies of the highpass-filtered data can be determined in different ways, for example:

1) The topographies can be determined as the relative signal amplitudes of the highpass- filtered data at one or several time points when the muscle artifact is known to be present.

2) The topographies can be determined with methods dividing the highpass-filtered signal into a sum of topographies μί weighted appropriately, such as principal component analysis (PCA; Pearson, 1901), singular value decomposition (SVD), or independent component analysis (ICA).

For example, PCA and SVD transform the highpass-filtered data to a new coordinate system of orthonormal eigenvectors = [μ ,ι μ ,₂... μ ]^Τ, each being a linear combination of the original variables (the EEG/MEG signals); these vectors are the topographies of the highpass- filtered data. Each eigenvector is associated with an eigenvalue The relative values of the eigenvalues describe how large part of the variance in the data each component explains. The highpass-filtered data can be expressed in the eigenvector basis:

Η(^ηι{()) = γ α_] (ί)μ_] , (3)

7=1 where α/t) is the time -varying amplitude of component j. Since noise is present in all signal- space directions, the highpass-filtered data consist of d orthogonal components. Therefore, projecting out all the dimensions of the highpass-filtered data from the original signal would remove all data. Classification of the components as those that reflect the muscle artifact and those that do not can be done in several ways, for example:

A) The components with the largest eigenvalues reflect the largest amount of variance in the data and thus the muscle artifacts in cases where they are larger than the noise. Accordingly, projecting out the N components (N < d), i.e., high-frequency topographies, with the largest eigenvalues using the signal-space projection method (SSP; Uusitalo ja Ilmoniemi, 1997) reduces the muscle artifact:

N

m_cm (t) = m(t) -∑ _{J J} ^T m(t) , (4) j=i

where the components have been sorted according to the eigenvalues in decreasing order (s₁≥ s₂≥...≥ s_N ) , m_∞n (t) is the corrected signal, and T stands for transpose.

B) The components can be classified according to their spatial patterns, such as relative amplitudes in the channels near the assumed muscle activity source.

C) A combination of A) and B) can be used, taking into account both the eigenvalues and the spatial patterns of the components.

A variation of the method: modeling the topographies

The topographies μ_ί that are projected out of the signals with the SSP method (Eq. 4) can also be determined by modeling the artifact source as current multipoles, i.e., dipoles (first order multipole; n = 1), quadrupoles (n = 2), octupoles (n = 3), etc. and by determining the potential differences or magnetic fields detected by each measurement channel as a result of each current multipole. In the model, the current multipoles are positioned in the assumed locations of the muscle artifact sources; for example, in a TMS-EEG measurement it could be assumed that the artifact source is located in those points of the muscle where the electric field induced by the stimulation is the largest. To minimize the number of topographies projected out, also the orientation of the current multipoles can be assumed, for example, based on the knowledge about the structure of the muscle. On the other hand, in each location r there can be only a certain number M of independent multipoles of order n. Thus, the signals generated by any «:th order multipole in location r can be projected out by projecting out the topographies generated by the M independent «:th order multipoles in location r. M depends on the order of the multipole and also on whether the signal in question is EEG or MEG.

It is also possible to use the muscle-artifact-containing frequency-filtered signals as a priori knowledge in modeling the current multipoles in variation 2. For example, the location of the current multipoles and the prevailing multipole orders can be estimated from the filtered data.

Possible modifications of the method

The number of the components projected out can be made time-dependent so that at those time points when the proportion of the muscle artifact in the signal is large, the number of components projected out is larger than at those time points when the proportion of the muscle artifact is smaller. For example, in case of a TMS-evoked EEG signal, the number of components projected out can depend on the distance in time from the time of the stimulus.

The data containing muscle artifact, from which the muscle artifact topographies are determined, can also be measured at a different time or with a different paradigm or with different stimulus parameters than the composite signal containing the signals of interest. For example, the paradigm eliciting the signals of interest may elicit very small muscle artifacts, whose removal is still desired. The muscle artifacts may then be elicited with different stimulus parameters so that they appear large in the signals and it is easier to determine their topographies.

In figure 1 , TMS of Broca's area produced large muscle artifacts in the evoked EEG signals, 1-3 orders of magnitude larger than the brain signals. These data show the need for removing the artifacts, as they mask the brain signal components that are of interest. In figures 2a and 2b, after applying the method according to preliminary claims 5, 9, 13, 14, and 15, the artifact is largely reduced, while a significant amount of brain signal is preserved, (a) The GMFAs normalized by the value of peak Al before (grey) and after (black) the artifact removal method was applied. The relative amplitudes of the brain signal peaks B1-B5 compared to the artifact peak Al increased as a result of the artifact removal, (b) Data presented in Fig. 1 after applying the artifact correction.

References

Barker AT, Jalinous R, Freeston IL (1985): Non-invasive magnetic stimulation of the human motor cortex. Lancet 1 :1 106-1107 Clancy EA, Morin EL, Merletti R (2002): Sampling, noise-reduction and amplitude estimation issues in surface electromyography. J Electromyogr Kinesiol 12: 1-16

Ilmoniemi RJ, Virtanen J, Ruohonen J, Karhu J, Aronen HJ, Naatanen R, Katila T (1997): Neuronal responses to magnetic stimulation reveal cortical reactivity and connectivity. NeuroReport 8:3537-3540 Kanai R, Chaieb L, Antal A, Walsh V, Paulus W (2008): Frequency-dependent electrical stimulation of the visual cortex. Current Biology 18: 1839-1843

Lehmann D, Skrandies W (1980): Reference-free identification of components of checkerboard-evoked multichannel potential fields. Electroencephalogr Clin Neurophysiol 48:609-621

Merletti R (1996): Standards for reporting EMG data. J Electromyogr Kinesiol 6:III-IV

Merton PA, Morton HB (1980): Electrical stimulation of human motor and visual cortex through the scalp. Journal of Physiology 305:9-10P

Maki H, Ilmoniemi RJ (2011): Projecting out muscle artifacts from TMS-evoked EEG. Neuro Image 54:2706-2710 Pearson K (1901): On lines and planes of closest fit to systems of points in space. Phil Mag 2:559-572

Priori A (2003): Brain polarization in humans: a reappraisal of an old tool for prolonged noninvasive modulation of brain excitability. Clinical Neurophysiology 114:589-595

Uusitalo MA, Ilmoniemi RJ (1997): Signal-space projection method for separating MEG or EEG into components. Med Biol Eng Comput 35: 135-140

Claims

1. A method for processing multichannel MEG, EEG, or other neurophysiological signals measured in order to obtain signals from the brain or other parts of the nervous system, composed of signal components of interest and muscle artifacts, characterized in:

a) filtering the signal so that the remaining part of the signal predominantly contains muscle artifact and a minimal amount of the signal of interest

b) using the filtered signal for modeling the topography or topographies of the muscle artifacts, and

c) projecting the modeled topography and topographies of the muscle artifact out of the original signal or its differently filtered version.

2. A method in accordance with claim 1 , characterized in that in phase b) the topography or topographies of the muscle artifact are determined from the topography or topographies of the filtered signals.

3. A method in accordance with claim 1 , characterized in that in phase b) the filtered signal is used as a priori knowledge in modeling the topography or topographies of the muscle artifacts.

4. A method according to claims 1 or 2 or 3, characterized in that either the signal of interest or the muscle artifacts or both are evoked as a result of electromagnetic stimulation such as TMS, TES, tDCS, tACS, or other type of electrical or sensory stimulation.

5. A method according to one or more of claims 1 to 4, characterized in that the muscle artifact is measured at a different time than the composite signal that contains the signal of interest.

6. A method according to one or more of claims 1 to 5 , characterized in that the muscle artifact is elicited using a different paradigm or different stimulus parameters than those used when the signal of interest is measured.

7. A method according to one or more of claims 1 to 6, characterized in that the topographies of the muscle artifact are determined at a selected time point or time interval.

8. A method according to one or more of claims 1 to 7, characterized in that the topographies of the muscle artifact are represented as a sum of linearly independent components such as those obtained by principal component analysis or singular value decomposition.

9. A method according to one or more of claims 1 to 7, characterized in that the topographies of the muscle artifact are represented as a sum of components that need not be linearly independent such as those obtained by independent component analysis.

10. A method according to one or more of claims 1 to 9, characterized in that prior knowledge from previous experiments is used to estimate which frequency components of the original composite signal would contain predominantly muscle artifacts and a minimal amount of the signal of interest.

1 1. A method according to one or more of claims 1 to 10, characterized in that the number of muscle artifact topographies projected out of the composite signal is time- dependent.

12. A method according to one or more of claims 1 to 11 , characterized in that independent component analysis is used to further analyze the projected signals.

13. A method according to one or more of claims 1 to 12, characterized in that the topographies of the muscle artifacts are modeled using current multipoles.

14. A system for processing multichannel MEG, EEG, or other neurophysiological signals measured in order to obtain signals from the brain or other parts of the nervous system, composed of signal components of interest and muscle artifacts, characterized in that it comprises means for:

a) filtering the signal so that the remaining part of the signal predominantly contains muscle artifact and a minimal amount of the signal of interest b) using the filtered signal for modeling the topography or topographies of the muscle artifacts, and

15. A system in accordance with claim 14, characterized in that element b) includes means for determining the topography or topographies of the muscle artifact from the topography or topographies of the filtered signals.

16. A system in accordance with claim 14, characterized in that element b) includes means for using the filtered signal as a priori knowledge in modeling the topography or topographies of the muscle artifacts.