Information

Removing ocular artifacts from EEG signals

Removing ocular artifacts from EEG signals



We are searching data for your request:

Forums and discussions:
Manuals and reference books:
Data from registers:
Wait the end of the search in all databases.
Upon completion, a link will appear to access the found materials.

I have read the following paper , which is try to remove EOG artifacts from EEG signal , the authors give the following figure .

The authors claim that components 1 and 2 represent electro-oculographic (EOG) artifacts. This is because its scalp map refers to locations very near to eye. My first question is how we map an ICA component to a certain location on scalp map. Secondly, how do we know a component lies on the front of the head, and not on another location, for example the center?


In short: we know that eye-blinks are reflected frontally in the EEG data and we use that knowledge to identify which components reflect for example eye-blinks. It would not make sense to identify a component related to eye-blinks on the back of the head - there would be something wrong with your data.

What ICA does is (data driven) estimate a number of statistically independent components (if you have 64 channels you get 64 components) based on your data. The input data is usually in the form of channels X timepoints X trials.

Eventually the output of ICA gives you, for each component, a weight for each channel which you can use to create the component scalp map. Together, the component scalp maps, the time course of the components, and frequency information of the component provide supplementary information that you can use to identify artifacts such as (but not limited to) eye-blinks, horizontal eye movements and heart beat.

In the above figure component 14 and 15 do not seem to be eye artifacts. (1) they are very focal, which is probably related to noise specific to that channel. (2) They look to be completely uncorrelated with the eye blinks in the original signal.


Removing ocular artifacts from EEG signals - Psychology

Abstract

Abstract—EEG signal is one of the oldest measures of brain activity that has been used vastly for clinical diagnoses and biomedical researches. However, EEG signals are highly contaminated with various artifacts, both from the subject and from equipment interferences. Among these various kinds of artifacts, ocular noise is the most important one. Since many applications such as BCI require online and real-time processing of EEG signal, it is ideal if the removal of artifacts is performed in an online fashion. Recently, some methods for online ocular artifact removing have been proposed. One of these methods is ARMAX modeling of EEG signal. This method assumes that the recorded EEG signal is a combination of EOG artifacts and the background EEG. Then the background EEG is estimated via estimation of ARMAX parameters. The other recently proposed method is based on adaptive filtering. This method uses EOG signal as the reference input and subtracts EOG artifacts from recorded EEG signals. In this paper we investigate the efficiency of each method for removing of EOG artifacts. A comparison is made between these two methods. Our undertaken conclusion from this comparison is that adaptive filtering method has better results compared with the results achieved by ARMAX modeling

To submit an update or takedown request for this paper, please submit an Update/Correction/Removal Request.


Removal of the Ocular Artifact from the EEG: A Comparison of Time and Frequency Domain Methods with Simulated and Real Data

Frequency-dependent transfer from EOG to EEG may be insufficiently accounted for by simple time domain regression methods (Gasser, Sroka, & Möcks, 1986 Woestenburg, Verbaten, & Slangen, 1983). In contrast, a multiple-lag time domain regression analysis, using lagged regression of EEG on EOG, must theoretically account for both frequency dependence and independence.

Two data sets were constructed, in which the transfer from EOG to EEG was either frequency-independent (constant gain) or frequency-dependent. Subsequently, three different correction methods were applied: 1) a simple regression analysis in the time domain 2) a multiple-lag regression analysis in the time domain and 3) a regression analysis in the frequency domain.

The major results were that, for data set 1, the three methods constructed the original EEG equally well. With data set 2, reconstruction of the original EEG was achieved reasonably well with the frequency domain method and the time domain multiple-lag method, but not with simple time domain regression. These three correction procedures were also applied to real data, consisting of concomitantly recorded EEG and high-variance EOG series. No appreciable differences in outcome of the three methods were observed, and estimated transfer parameters suggested that these data were marked by weak frequency dependence only, which can be accounted for by simple time domain regression (and also by the other two methods).


Removing ocular artifacts from mixed EEG signals with FastKICA and DWT

Affiliations: [ a ] College of Electronic Information & Control Engineering, Beijing University of Technology, Beijing, China | [ b ] Beijing Key Laboratory of Computational Intelligence and Intelligent System, Beijing, China

Correspondence: [*] Correspondence to: Li Mingai. Tel.: +86 10 6739 6309 Fax: +86 10 6739 1625 [email protected]

Abstract: Ocular movements are inevitable in electroencephalograme (EEG) collection, and the resulting Ocular Artifact (OA) becomes one of the main interferences of EEG due to its great amplitude. Many methods have been proposed to remove OA from EEG recordings based on Blind Source Separation (BSS) algorith m. Often regression is performed in time or frequency domain by completely deleting the OA components. This can cause the overestimation of OA and the information loss of EEG, because EEG and electrooculogram (EOG) mix or spread bidirectionally. Furthermore, there exists a variety of noises, except for OA, and interference coupling in EEG, this also affects the OA removal performance, such as the robustness and anti-interference ability. Here, we propose a novel and generally applicable method, denoted as FKD, for removing OA from mixed EEG signals with the Fast Kernel Independent Component analysis (FastKICA) and Discrete Wavelet Transform (DWT). In two cases of linear and nonlinear mixed models, many experiments are conducted with Brain Computer Interface (BCI) data set. The experiment results show that FKD has good performance comparing with other BBS-based OA removal methods, and it is more acceptable in actual BCI system.

Keywords: Ocular artifact removal, fast kernel independent component analysis, discrete wavelet transform, overestimation, robustness


Getting Rid of Eye Blink in the EEG

One of the challenges of EEG is that the surface electrodes also pick up the electrical activity of muscle contraction on the face (the electromyograph or EMG). While a subject can be asked to try and not change their facial expression too much, the eye blink or Electro-oculographic (EOG) simply can only be controlled so much. It is therefore one of the biggest contaminating signals and can greatly hinder the interpretation of EEG signals. The eye blink is characterized by a sharp deflection that is usually strongest at the front of the brain and is easy to recognize. For this reason, many ‘BCI’ or brain-computer interface applications actually make use of the eye blink rather than brain activity. However, if you want to make sure you are studying brain activity alone it is necessary to get rid of it and researchers have come up with many methods to do so. The two most commonly used and popular methods are linear regression and independent component analysis (ICA) which we describe here. Yet despite how easy it may seem to visually ‘see’ the EOG in the signal, specifically removing it has its challenges.

Linear regression

Linear regression algorithms have been commonly used to remove ocular artifacts from EEG data due to their simplicity. This approach typically requires an additional set of electrodes on the face to record electrical activity due to eye blinks and movements.

The main assumption in this approach is that each EEG channel can be expressed as the sum of noise-free EEG signal and a fraction of the source artifact available through EOG electrodes. The aim of the regression approach is then to correct the EOG artifacts by subtracting weighted portion of each EOG channel from the contaminated EEG signal.

Let S be the recorded EEG signal which can be expressed as the sum of noise-free EEG signal E and EOG or eye blink signal B multiplied by a weight matrix W.

S = WB + E

The weight matrix W contains what is known as the regression co-efficients that describes the contribution of the EOG artifact in each EEG channel and has to be figured out and calculated somehow. Once W is known, the noise-free EEG activity is easily obtained by simply subtracting the weighted eye blink B from the signal.

E = SWB

The most commonly used to method to perform linear regression is the ordinary least squares (OLS) method – essentially imagine fitting a regression line using the mean squared sum of residuals through a scatter of the recorded signal on each channel vs the recorded EOG and finding its regression coefficient. This amounts to estimation of W by minimizing the residuals over all the channels, i.e. minimizing the term (S – WB) T (S – WB) by setting its gradient with respect to W to zero.

One of the major problems with this approach is the assumption that the EOG electrodes do not record any EEG activity, which is simply not true. Since EOG signals which are recorded near the forehead will also record some EEG activity, the OLS method will tend to overestimate W leading to the removal of some EEG activity as well along with the EOG activity.

Independent component analysis

Independent component analysis (ICA) is a blind source separation (BSS) technique that is widely used in an array of signal processing applications. ICA has become quite popular in denoising biomedical signals and is the most preferred / popular method to clean EEG data. It is available in EEGLAB [1] for example, which also provides a nice visualization for ICA analysis.

The intuition behind ICA is best explained using cocktail party problem

Figure adapted from [5]

Imagine you are at a party where the music is being played and at the same time there is some speech going on (pretty bad party arrangement!). These two sources of sound, let’s call them s1 and s2 are being picked up by microphones m1 and m2. ICA assumes that the sounds s1 and s2 add up or mix linearly and the output of each microphone can be expressed as

where a1, b1, a2, b2 are the linear weights describing the contribution of each original sound to the microphone. The goal of ICA is to retrieve the weights and the original sounds s1 and s2. So if one is just interested in listening to the music and wants to get rid of the speech signal, then one simply has to set the weights b1 and b2 to zero in the above equation!

Similarly, in the context of EEG recording, one can thing of EEG electrodes as microphones that are picking up sound from various sources like neural activity, eye activity, muscle activity etc. ICA can thus be used to disentangle these sources, which are called as independent components. Let S be the contaminated EEG signal. ICA assumes the following linear model

S = M*X

Where M is the mixing matrix, that contains the linear weights describing the contribution of neural and various non-neural sources to the EEG recording and X the underlying sources, also known as independent components. ICA seeks to estimate

Xest = W*S

where W is the un-mixing matrix. There are many algorithms to solve the ICA problem (and are readily available in EEGLAB) and the majority of them use higher-order statistics. We will not go into the math or the details of it, but it is important to know that this is an under-constrained problem, meaning that we observe only S and try to estimate M and X so the number of unknowns are more than the observations. It is hence a challenging problem.

It is important to note that ICA methods makes four main assumptions – 1) Summation of sources at scalp electrodes are linear, 2) Artifacts and neural sources are spatially fixed across time, 3) Component activations are temporally independent and 4) Underlying sources are non-Gaussian. In addition, it is also assumed that the number of independent sources are at most equal to the number of recording channels.

The amount of data given to the algorithm plays an important role in ICA. As ICA is based on statistical features, results will not be reliable if the data length is short or insufficient. Romero, Mañanas and Barbanoj have suggested that the length of the data should be some constant times the square of the number of channels [8]. Jung, Makeig and others have shown that epochs of 10s are enough to obtain reliable results [9]. Also, assumption no. 2 may not always be true in practice (neural activity need not be spatially stationary through time) and hence the choice of data length should be such that assumption no. 2 is reasonably satisfied.

Identification of blinks and eye movement components

Once the components have been identified, to remove the EOG artifacts, one can visually determine which independent component corresponds to eye-blinks or movements based on the following criteria.

To identify blink and horizontal eye-movement related components the following properties are typically used,

  1. Presence of frontal topography (for blinks, shown on left) and bilateral with opposite sign frontal topography (for horizontal eye-movements, shown in right) in scalp map (adapted from here).

2. Flat power spectra with no peaks at physiologically relevant frequencies

Figure adapted from [2]

  1. High correlation with vertical EOG recordings (for blinks) or with horizontal EOG recordings in case of horizontal eye-movements.
  2. Presence of high amplitude in the component.

After identifying the components related to EOG artifacts, the corresponding column in the mixing matrix to M is set to zero, which essentially means zeroing the weights of EOG-related sources to all the EEG electrodes. The noise free EEG can now be obtained as

S = M*Xest

In the figure above (taken from here) one can see that the independent components 1 and 2 are clearly related to EOG artifacts, which is also evident from the scalp maps showing high activity in the frontal sites. Setting these components to zero gets rid of artifacts related to eye-movements as seen in the figure above.

Another point to note is that, as with regression, the components related to EOG activity may also contain some EEG activity and thus in practice there is no guarantee that discarding the EOG related components may not result is throwing away some useful EEG information as well!

Automated selection of components

Since visual identification of artifacts can be very subjective, researches have come up with objective statistical measures to automatically identify EOG-related components. These statistical measures include correlation with EOG electrodes, temporal kurtosis, spatial average and variance difference, maximum epoch variance are available and are implemented in tools like SASICA [2], FASTER [3] and ADJUST [4] which are available as plugins in EEGLAB.

Other methods

Adaptive and Weiner filtering can be used to reduce EOG artifacts. Adaptive filtering is an improvement over simple linear regression where the weights can vary with time unlike in linear regression. Weiner filtering is another filtering approach that aims minimize the mean square error between the desired signal and its estimate using power spectral densities. This approach does not require a reference EOG waveform.

Improvements / extensions to ICA approach have also been suggested. By imposing spatial constraints on the mixing matrix, i.e., prior assumption of the spatial topography of source sensor projections, it has been shown that the performance of ICA in denoising EEG signals can be greatly improved.

Then there are methods such as principal component analysis, signal space projection, cannonical correlation analysis that can be employed for EOG artifact reduction, but ICA is mostly preferred in the EEG community due to its ability to deal with many kinds of artifacts (EOG, muscle artifact, cardiac artifacts, line noise) at once. Although, ICA is widely used, it is important to keep in mind that this approach is also not devoid of pitfalls and makes several assumptions that in practice may not be true. Thus, being aware of these assumptions while choosing the data length and properly classifying artifactual components is critical.

Thus while none of the methods can do a perfect job in removing the eye blink and some EEG signal may be lost in the process, its presence can substantially distort metrics of the signal – particularly metrics like the Hurst exponent or DFA. Thus it is always prudent to do the best job possible.


Removing ocular artifacts from mixed EEG signals with FastKICA and DWT

Affiliations: [ a ] College of Electronic Information & Control Engineering, Beijing University of Technology, Beijing, China | [ b ] Beijing Key Laboratory of Computational Intelligence and Intelligent System, Beijing, China

Correspondence: [*] Correspondence to: Li Mingai. Tel.: +86 10 6739 6309 Fax: +86 10 6739 1625 [email protected]

Abstract: Ocular movements are inevitable in electroencephalograme (EEG) collection, and the resulting Ocular Artifact (OA) becomes one of the main interferences of EEG due to its great amplitude. Many methods have been proposed to remove OA from EEG recordings based on Blind Source Separation (BSS) algorith m. Often regression is performed in time or frequency domain by completely deleting the OA components. This can cause the overestimation of OA and the information loss of EEG, because EEG and electrooculogram (EOG) mix or spread bidirectionally. Furthermore, there exists a variety of noises, except for OA, and interference coupling in EEG, this also affects the OA removal performance, such as the robustness and anti-interference ability. Here, we propose a novel and generally applicable method, denoted as FKD, for removing OA from mixed EEG signals with the Fast Kernel Independent Component analysis (FastKICA) and Discrete Wavelet Transform (DWT). In two cases of linear and nonlinear mixed models, many experiments are conducted with Brain Computer Interface (BCI) data set. The experiment results show that FKD has good performance comparing with other BBS-based OA removal methods, and it is more acceptable in actual BCI system.

Keywords: Ocular artifact removal, fast kernel independent component analysis, discrete wavelet transform, overestimation, robustness


Barlow, J. S. (1979) Computerized clinical electroencephalography in perspective.IEEE Trans.,BME-26, 377–391.

Barlow, J. S. and Rémond, A. (1981) Eye movement artifact nulling in EEGs by multichannel on-line EOG subtraction.Electroenceph. Clin. Neurophysiol.,52, 418–423.

Barlow, J. S., Creutzfeldt, O. D., Michael, D., Houchin, J. and Epelbaum, H. (1981) Automatic adaptive segmentation of clinical EEGs.,51, 512–523.

Bierman, G. J. (1976) Measurement updating using the U-D factorization.Automatica,12, 375–382.

Clarke, D. W. (1981a) Introduction to self-tuning controllers. InSelf-tuning and adaptive control: theory and applications. Harris, C. J. and Billings, S. A. (Eds.), Peter Peregrinus, 36–71.

Clarke, D. W. (1981b) Implementation of self-tuning controllers. InSelf-tuning and adaptive control: theory and applications. Harris, C. J. and Billings, S. A. (Eds.), Peter Peregrinus, 144–165.

Eykhoff, P. (1974)System identification. John Wiley.

Fortgens, C. and de Bruin, M. P. (1983) Removal of eye movement and ECG artifacts from the non-cephalic reference EEG.Electroenceph. Clin. Neurophysiol.,56, 90–96.

Girton, D. G. and Kamiya, J. (1973) A simple on-line technique for removing eye-movement artifacts from the EEG.,34, 212–216.

Gotman, J., Gloor, P. and Ray, W. F. (1975) A quantitative comparison of traditional reading of the EEG and interpretation of computer-extracted features in patients with supratentorial lesions.,38, 623–639.

Harris, M. A. (1983) Brain waves appear on T.V. in real-time.Electronics, 24th Feb., 47–48.

Hillyard, S. A. and Galambos, R. (1970) Eye movement artifact in the CNV.Electroenceph. Clin. Neurophysiol.,28, 173–182.

Jervis, B. W., Nichols, M. J., Allen, E., Hudson, N. R. and Johnson, T. E. (1980) The quantitative assessment of electroencephalograms corrected for eye movement artefacts. 1st European Conf. on Signal Processing, Lausanne, Switzerland.

Jervis, B. W., Nichols, M. J., Allen, E., Johnson, T. E. and Hudson, N. R. (1984) Pattern recognition techniques applicable to the contingent negative variation for the differentiation of subject categories.IEEE Trans.,BME-31, 342–349.

Jervis, B. W., Nichols, M. J., Allen, E., Hudson, N. R. and Johnson, T. E. (1985) The assessment of two methods for removing eye movement artefact from the EEG.Electroenceph. Clin. Neurophysiol.,61, 444–452.

Johnston, J. (1972)Econometric methods. McGraw-Hill.

McCallum, W. C. and Walter, W. G. (1968) The effects of attention and distraction on the contingent negative variation in normal and neurotic subjects.Electroenceph. Clin. Neurophysiol.,25, 319–329.

Morris, E. L. and Abaza, B. A. (1976) Adaptive control of steam turbine.Proc. IEE,123P, 549–553.

Peterka, C. (1975) A square-root filter for real-time multivariate regression.Kybernetika,11, 53–67.

Quilter, P. M., MacGillivray, B. B. and Wadbrook, D. G. (1977) The removal of eye movement artefact from EEG signals using correlation techniques. InRandom signal analysis. IEE Conf. Publ. 159, 93–100.

Thornton, C. L. and Bierman, G. J. (1978) Filtering and error analysis via the UDU covariance factorization.IEEE Trans.,AC-23, 901–907.

Wasman, M., Morehead, S. D., Lee, H. and Roland, V. (1970) Interaction of electro-ocular potentials with the contingent negative variation.Psychophysiology,7, 103–111.

Weerts, T. C. and Lang, P. J. (1973) The effects of eye fixation and stimulus and response location on the contingent negative variation.Biol. Psychol.,1, 1–19.

Whitton, J. L., Lue, F., Moldofsky, H. (1978) A spectral method for removing eye movement artifacts from the EEG.Electroenceph. Clin. Neurophysiol.,44, 735–741.


Abstract

Ocular artifacts (OAs) are one the most important form of interferences in the analysis of electroencephalogram (EEG) research. OAs removal/reduction is a key analysis before the processing of EEG signals. For classic OAs removal methods, either an additional electrooculogram (EOG) recording or multi-channel EEG is required. To address these limitations of existing methods, this paper investigates the use of deep learning network (DLN) to remove OAs in EEG signals. The proposed method consists of offline stage and online stage. In the offline stage, training samples without OAs are intercepted and used to train an DLN to reconstruct the EEG signals. The high-order statistical moments information of EEG is therefore learned. In the online stage, the trained DLN is used as a filter to automatically remove OAs from the contaminated EEG signals. Compared with the exiting methods, the proposed method has the following advantages: (i) nonuse of additional EOG reference signals, (ii) any few number of EEG channels can be analyzed, (iii) time saving, and (iv) the strong generalization ability, etc. In this paper, both public database and lab individual data for EEG analysis are used, we compared the proposed method with the classic independent component analysis (ICA), kurtosis-ICA (K-ICA), Second-order blind identification (SOBI) and a shallow network method. Experimental results show that the proposed method performs better even for very noisy EEG.


Removing ocular artifacts from EEG signals - Psychology

EEG signal is one of the oldest measures of brain activity that has been used vastly for clinical diagnoses and biomedical researches. However, EEG signals are highly contaminated with various artifacts, both from the subject and from equipment interferences. Among these various kinds of artifacts, ocular noise is the most important one. Since many applications such as BCI require online and real-time processing of EEG signal, it is ideal if the removal of artifacts is performed in an online fashion. Recently, some methods for online ocular artifact removing have been proposed. One of these methods is ARMAX modeling of EEG signal. This method assumes that the recorded EEG signal is a combination of EOG artifacts and the background EEG. Then the background EEG is estimated via estimation of ARMAX parameters. The other recently proposed method is based on adaptive filtering. This method uses EOG signal as the reference input and subtracts EOG artifacts from recorded EEG signals. In this paper we investigate the efficiency of each method for removing of EOG artifacts. A comparison is made between these two methods. Our undertaken conclusion from this comparison is that adaptive filtering method has better results compared with the results achieved by ARMAX modeling.


Removal of EOG Artifacts from EEG Recordings Using Stationary Subspace Analysis

An effective approach is proposed in this paper to remove ocular artifacts from the raw EEG recording. The proposed approach first conducts the blind source separation on the raw EEG recording by the stationary subspace analysis (SSA) algorithm. Unlike the classic blind source separation algorithms, SSA is explicitly tailored to the understanding of distribution changes, where both the mean and the covariance matrix are taken into account. In addition, neither independency nor uncorrelation is required among the sources by SSA. Thereby, it can concentrate artifacts in fewer components than the representative blind source separation methods. Next, the components that are determined to be related to the ocular artifacts are projected back to be subtracted from EEG signals, producing the clean EEG data eventually. The experimental results on both the artificially contaminated EEG data and real EEG data have demonstrated the effectiveness of the proposed method, in particular for the cases where limited number of electrodes are used for the recording, as well as when the artifact contaminated signal is highly nonstationary and the underlying sources cannot be assumed to be independent or uncorrelated.

1. Introduction

The electroencephalographic (EEG) provides a noninvasive facility to investigate the intricacy of human brain. It has been applied in numerous applications such as brain-computer interface and clinical diagnosis of neurological disorders [1]. A common problem in EEG applications is that the EEG is susceptible to artifacts in the data acquisition [2]. For example, the electric potentials created during eye movement and blinks can be orders of magnitude larger than the EEG and can propagate across much of the scalp, distorting EEG signals. Consequently, such electrooculographic (EOG) artifacts will hinder the interpretation of EEG, it is thereby important to remove the EOG artifacts before further analysis of EEG.

In the literature, the most common EOG artifacts removal method is based on the blind source separation (BSS), usually by independent component analysis (ICA) [3] and second-order blind identification (SOBI) [4]. Such approaches assume that the recorded EEG signals are represented by a limited number of components (or “sources”) and then discarded by those responsible for artifacts during the reconstruction. It has been shown that they are useful in many EOG artifact removal applications [5–8].

However, the classic BSS techniques such as ICA and SOBI may not be effective on the highly nonstationary EOG artifact-contaminated EEG recordings. On one hand, the ocular artifacts resulting from eye movement and blink demonstrate strongly nonstationary characteristics within a considerably long interval: it is often localized with abruptly large amplitude and low frequency its duration and amplitude appear to differ stochastically and considerably between successive eye movements or blinks. This implies that there are significant distribution changes in the artifact-contaminated EEG observations, such as the changes in the mean and the covariance matrix. However, ICA is not devoted to the understanding of the distribution changes but to find the components that are both statistically independent and non-Gaussian [3]. Although SOBI exploits the temporal changes in the covariance matrix of the observations by the joint diagonalization of several covariance matrices with different time delays, the changes in the mean of the observations have not been taken into account. On the other hand, the underlying sources associated with artifacts may not be assumed to be independent or uncorrelated among each other. It is known that ICA and SOBI can perform well on the eye blink artifacts contaminated EEG signals [5–8]. This is because blink artifacts mainly involve the vertical movement, the sources corresponding to the vertical movement and horizontal movement of eyes thus, can be assumed to be independent or uncorrelated. However, when both the vertical and horizontal eye movements are involved in the contamination of the EEG observations, the sources associated with such EOG artifacts cannot be assumed to be independent nor uncorrelated anymore, because the vertical and horizontal eye movements are often accompanied with each other, in particular for eye ball rolling, forming two dependent components. Nevertheless, ICA instead has imposed the requirement of sourcewise independency, while SOBI is based on the assumption that the sources should be uncorrelated with each other. Thus, on the highly nonstationary EOG artifact-contaminated EEG recordings, they raise the problem of “misallocation of variance” [9, 10] and result in suboptimal separation: the artifacts fail to be concentrated in a small number of components by ICA and SOBI. Since it is usually assumed that the number of sources is no greater than the number of channels in the BSS literature, in cases where there are limited number of electrodes used for EEG recording (e.g., often used in sleep studies [11] and when subjects are neonates or young infants, due to the size of the head [12]), it may lead to the loss of information related to brain activity by rejecting those sources found by ICA or SOBI, as such sources generally also contain neural activity aside from pure artifacts [13, 14].

To addresses the weakness of ICA-based and SOBI-based artifact correction methods on highly nonstationary EEG recordings with limited number of electrodes, this paper proposes a novel EOG artifact removal approach that utilizes the recently proposed effective BSS method, that is, stationary subspace analysis (SSA) [15, 16]. Unlike the classic blind source separation with ICA or SOBI, the adopted BSS algorithm SSA is explicitly tailored to the understanding of distribution changes [16]. The type of distribution changes that SSA detects is changes in both the mean and the covariance matrix. In addition, neither independency nor uncorrelation is required among the sources by SSA. Subsequently, artifacts can be concentrated in only a few components. To the best of our knowledge, SSA has not been applied to EEG signals for removing artifacts thus far, even though it has been shown to have interesting applications in robust motor imagery prediction for Brain-Computer Interface [15, 17], geophysical data analysis [16], WiFi localisation [18], computer vision [19], and change-point detection [20]. Experiments on both simulated data and real EEG recordings are conducted, and the results show that the proposed method can effectively improve the artifact correction on raw EEG recordings.

The remainder of the paper is organized as follows. In Section 2, we present the proposed approach for EOG artifact correction in detail. Experimental results are given in Section 3. Finally, conclusions and discussion are presented in Section 4.

2. Proposed Approach for EOG Artifact Removal

Figure 1 shows the block diagram of the proposed approach. It comprises the following key steps:


Watch the video: O παιδονευρολόγος Μηνάς Καπετανάκης μιλά για το ηλεκτροεγκεφαλογράφημα στο παιδί (August 2022).