Journal of Physics: Conference Series
PAPER • OPEN ACCESS
Methods for removal of artifacts from EEG signal: A review
To cite this article: Shailaja Kotte and J R K Kumar Dabbakuti 2020 J. Phys.: Conf. Ser. 1706 012093
View the article online for updates and enhancements.
This content was downloaded from IP address 91.132.36.210 on 30/12/2020 at 22:19
First International Conference on Advances in Physical Sciences and Materials IOP Publishing
Journal of Physics: Conference Series 1706 (2020) 012093 doi:10.1088/1742-6596/1706/1/012093
Methods for removal of artifacts from EEG signal: A review
1,2
ShailajaKotte and 3J R K Kumar Dabbakuti
1,3
Department of ECM, Koneru Lakshmaiah Education Foundation, Vaddeswaram,
Guntur, Andhra Pradesh, 522502, India.
2
Department of EIE, Kakatiya Institute of Technology and Science, Warangal
506015.
Abstract. Electroencephalogram (EEG)is the record of cerebral activity, the electric potential
of cerebral activity is of low amplitude, and less frequency ranges between 4 to 60 Hz, which
can easily mix up different non-cerebral signals and other environmental noise signals. The
extraction of actual cerebral signals from the contaminated EEG signal is the major challenge
in medical analysis. Somehow, during the recording of the EEG signal, contamination of other
signals takes place, which increases complexity in analyzing the accurate EEG signal. This
leads to inaccurate information signals in the analysis. Accordingly, the process to eliminate
the unwanted signals in the pre-processing level is mandatory in brain signal analysis. The
unwanted signals from various sources are together termed as artifacts; the researchers have
implemented various techniques to reduce the undesired signals. However, still, there is no
standard technique in detecting and eliminating the artifacts, and hence, the research became
most challenging.
Keywords: Discrete wavelet transform (DWT), Empirical mode decomposition (EMD),
Independent component analysis (ICA), Canonical correlation analysis (CCA), Singular value
decomposition (SVD),
1. Introduction
The Electroencephalogram (EEG) represents states of the brain of the human being’s mental
condition. EEG signals are the electrical potentials of the cerebral, which is contaminated with other
bio-potentials like Electroculogram (EOG), Electrocardiogram (ECG), and Electromyogram (EMG).
Hence the segregation of EEG signals from different bio-potentials is a challenging task for analyzing
EEG signals. Source separation methods focused much attention on EEG signal analysis; primarily,
the source separation techniques are designed based on two types of artifacts; they are extrinsic and
intrinsic artifacts [1-3]. Extrinsic artifacts have occurred electric and electronic components, namely
high electrode impedance and meddling, line noise, earthling problems, electrode failure, ventilation,
and power supply. Intrinsic means various artifacts that occurred in EEG signals like movements of
the eye, eye blinks, bioelectric potentials from muscle, heart etc. Artifacts are unwanted components
that arise from other sources; they mislead the actual cerebral activity of present recorded EEG data,
and hence this leads to more complex in analyzing the EEG data. Due to artifacts, the accurate EEG
signal is misguided. Hence the extraction of original EEG signals from contaminated EEG signals
occupied the highest priority in the field of EEG signal processing.
In the last thirty years, the EEG machines have gained great significance due to low power
consumption. Several techniques were implemented by the researchers for eliminating artifacts. This
review paper will provide a record of a detailed overview of distinguished methods to minimize and
eliminate the artifacts in EEG signals. The first remedy to deduct artifacts in the EEG signal is to
avoid movements that can incur them. Several methods have been implemented to reduce the
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
First International Conference on Advances in Physical Sciences and Materials IOP Publishing
Journal of Physics: Conference Series 1706 (2020) 012093 doi:10.1088/1742-6596/1706/1/012093
artifacts; however, the investigation on reducing the several undesired components remained to be a
prevalent problem.
2. Background
In 1875, Richard Caton exposed the brains of rabbits and monkeys. Later in 1875, Adrian and
Matthews verified perception on “Human brainwaves” recognized standard electrical bio-potential
ranges between 10-12 Hz and named as “Alpha Waves.” EEG is the average sum of ionic current, this
current which flows during the synaptic excitations of the dendrites in a neural ensemble of the
cerebral cortex. EEG measures the action of many neurons. Hans Berger traced the signals by
recording through the EEG machine in 1929 for the first time. The human brain is divided into five
different lobes named as the frontal lobe, parietal lobe, occipital lobe, temporal lobe, and cerebellum.
These lobes are the origins of the various electrical potential that are generated based on human
activities. As the electrical field produced by the neural activation travels, its voltage decreases and
with every barrier (Dura mater, skull, skin) between the cortex and scalp, which are recorded as
electroencephalogram and shortly known as EEG (Figure 1).
Figure1. Elements of Human Brain.
The EEG recording process is non-invasive, trouble-free, and does not create pain for the
patient. EEG recordings resemble the global behavior of the brain. These waveforms display the
rhythmic pattern at characteristic frequencies. EEG signals associated with the behavior of human
activities such as excited, relaxed, drowsy, asleep, deep sleep, coma etc. These activities fall into
distinct rhythms of frequency ranges, as shown in Table 1.
Table 1. Rhythms of distinct frequency ranges.
Type Frequency Activity Location
Gamma 20 – 60Hz Visual attention Occipital
Beta 14 - 20 Hz Mental activity Parietal and frontal
Alpha 8 - 13 Hz Sensory stimulation Occipital and parietal
Theta 4- 8 Hz Emotional Stress Temporal and parietal
Delta Less than 4 Hz Occur during sleep, coma Everywhere
2
First International Conference on Advances in Physical Sciences and Materials IOP Publishing
Journal of Physics: Conference Series 1706 (2020) 012093 doi:10.1088/1742-6596/1706/1/012093
2.1 Types of artifacts
In pre-processing of EEG data identifying the artifacts has the highest priority. The artifacts corrupt
the originality of EEG data. In this regard, the artifacts are requisite to get rid of actively. The cardiac
Artifacts, muscle Artifacts, ocular Artifacts, and external device Artifacts are the typical artifacts that
intermingle with EEG signals (Figure 2).
Figure 2. Various Bio-Potential Artifacts.
2.1.1. Cardiac Artifacts
The pulsating human heart generates the electric potentials known electrocardiogram. These electric
potentials of the heart (ECG signals) are conducted to the scalp, which is intermingled with the EEG
signals, thereby creates the potential change in the measured signal of EEG Cardiac artifacts are of
two types Mechanical and Electrical. Every contraction and irregular interval of cardiac arrhythmia
can be considered as mechanical artifacts. The electrical artifacts may be due to the heart electrodes.
ECG artifacts may disrupt the EEG background activity that which is the replica of epileptiform
discharges. Moreover, it usually is diphonic or triphonic with an active component that has duration
within the spike range. Cardiac artifacts are of two types Mechanical and Electrical. Every contraction
and irregular interval of cardiac arrhythmia can be considered as mechanical artifacts. The electrical
artifacts may be due to the heart electrodes.
2.1.2. Muscle Artifacts
EMG has a broad classification from 0Hz to >200Hz [1, 4]. EMG is the muscle activity potential that
is generated at various muscle groups; the EEG signal that is measured consists of EMG potential that
contaminates the original EEG signal [5, 6]. Moreover, EMG has an effect on both temporally and
spatially.
3
First International Conference on Advances in Physical Sciences and Materials IOP Publishing
Journal of Physics: Conference Series 1706 (2020) 012093 doi:10.1088/1742-6596/1706/1/012093
2 .1.3. Electrode Artifacts
EEG measurement consists of apparatus, equipment, and connections. Most probably electrode
artifact, and may occur due to Amplifier artifact, Capacitive artifact, Inductive artifact, Electrostatic
artifact, aliasing artifact. Poor Contact of the electrode may affect the electrode impedance, which
results in low-frequency artifacts. Temperature variations and biasing error in instrumentation
amplifiers may cause due to the baseline drift. More problems with electrode artifact results in
prolonged monitoring and scalp integrity.
3. Traditional Methods in Artifacts Removal
3.1 Regression Method.
The fundamental process for eliminating unwanted signals in EEG is the regression method [7]. This
method described the amplitude relation of reference and estimated artifacts from EEG; Thereby, this
algorithm requires external reference signals such as ECG, EOG, and EMG to segregate the various
undesired signals. From figure 3, the first stage output consists of EEG, ECG, EOG signals, where
line frequency is omitted with reference channel as a linear filter; finally, the corrected EEG data can
be obtained (Figure 3).
Figure 3. Regression method approach.
e3(n) = e2(n) − e1(n) (1)
Hillyard and Gallambos, [8] done their research in removing EOG based on the time-domain
regression approach. Whitton et al. [9] approached based on frequency domain regression, and in
conjunction with the software tool, the EEG signal had been analyzed. Where ocular potentials can
contaminate the EEG potentials and similarly. The simplified model and reduced computational
method are the pros, and it needs the reference channels for eliminating the EOG, and ECG remained
as cons of the regression methods [1]. The researchers found the counterclaim to a regression
algorithm named Blind source separation techniques, but still, the regression methods remained as a
golden foundation to appraise the novel approaches.
4
First International Conference on Advances in Physical Sciences and Materials IOP Publishing
Journal of Physics: Conference Series 1706 (2020) 012093 doi:10.1088/1742-6596/1706/1/012093
3.2 Wavelet Transform
All waves may be considered forms of time-frequency representation. The spectral analysis of various
signals is performed using the wavelet transform. Choosing the appropriate wavelet and the number
of decomposition levels is very significant in the study of signals in WT. Wavelet decomposing of
levels is based on the frequency domain of the signal [10-12]. The levels are selected such that those
parts of the signal that associate well with the frequencies required for classification of the signal is
retained in the wavelet coefficients. Finally, the transformation is achieved by the selection of subsets
‘j’, and time shifts ‘k’ of the mother wavelet (t).where ‘j’ and ‘k’ are integers. Therefore,
mathematically expressed as:
Y j ,k ( t ) = 21/ 2 Y ( 2 j t − k ) (2)
Now, the wavelet can be performed by:
WY = f , Yj , k (3)
Figure 4. Block Diagram of Wavelet Transform System.
From the figure 4, the input signal is raw EEG signal, in the next stage it is decomposed into
various levels, Discrete Wavelet Transform is the result of continuous wavelet can be applied when
the input signal and the decomposition can be expressed as:
Xa, L[ n] = (Nk =1) Xa − 1, L[2n − k ] g[ k ] (4)
Xa, H [n] = (Nk =1) Xa − 1, L[2n − k ]h[k ] (5)
g[n] –low frequency component of low pass filter.
h[n]- high frequency component of high pass filter.
After decomposing wavelet transformation on EEG data, then the threshold is chosen to reject
that component, which consists of artifacts. Now with the leftover, the signal is reconstructed without
artifact. In the process of artifact deduction, the DWT was declined to categorize artifacts. However,
still, the DWT approach must move forward on artifact attenuation that has common characteristics
with the spectral properties; for this reason, new work carried out by the mixture of DWT with new
methods similar to ICA.
3.3 Blind Source Separation
5
First International Conference on Advances in Physical Sciences and Materials IOP Publishing
Journal of Physics: Conference Series 1706 (2020) 012093 doi:10.1088/1742-6596/1706/1/012093
BSS technique had the superior flexibility with a multiplicity of learning algorithms such that no extra
reference channels and any preliminary information. Assume that X be an electric potential signal
generated by EEG electrodes. Moreover, S is the source signal (original and artifact signal). The
source signal is mixed with an unknown matrix A then,
X = AS (6)
To obtain the observed signals BSS algorithm is an extended version:
U = WX (7)
Where U is the assessment of sources and the W is the reverse mixing of X. Subsequently, the
components as artifacts are removed, and then with the remaining components, the EEG data is
reconstructed. Some representative works that have adopted from BSS are discussed below.
3.3.1 Principal Component Analysis
PCA is the most widely used tool for investigative data analysis and for making predictive models. It
is a static process to convert a set of observations of possibly correlated variables into a set of values
of linearly uncorrelated variables called principal components [13-16]. Berg and Scherg, [17] work
proved that the ocular artifacts could be removed. In this effort, the extraction of EEG data was
obtained by eliminating the blinks and eye movements. Therefore, the accurate EEG signal is
obtained. PCA algorithm failed in segregation, where the potentials of drifts and EEG potentials are
similar. As a result, later research prefers further flexible methods.
3.3.2 Independent Component Analysis
ICA is a part of BSS, in its unique way. A typical example application is the ‘cocktails party’ problem
of listening in one person’s speech in a noisy room. In EEG signal analysis, signal sources are direct
mixtures of cerebral and other different sources, that can be decomposed into independent
components ICs [18]. The Independent components are extracted from the raw original signals. Now
the reconstruction is implemented by rejecting the ICs components that contain artifacts (Figure 5).
To analyze the EEG and EPR signals, the extended version of ICA is proposed, Jung et al. [19]
succeeded in removing the EEG artifacts and judging the results against regression methods [20].
Figure 5. Block Diagram of Independent components Analysis Method.
The expansion of PCA confined to transform orthogonal directions; ICA provided the
evidence in source signal separation in a more effective way. ICA approaches introduced by several
researchers in the extraction of signals from various mixtures and to decrease the artifacts of different
sleep stages. ICA can calculate approximately the original signals which are non-Gaussian.
3.3.3 Canonical Correlation Analysis
6
First International Conference on Advances in Physical Sciences and Materials IOP Publishing
Journal of Physics: Conference Series 1706 (2020) 012093 doi:10.1088/1742-6596/1706/1/012093
Most prevalently and effectively used signal processing technique as that is evolved from BSS is
Canonical correlation analysis (CCA). CCA has variation from ICA in its concept in separating the
components and has less computation time, which is the advantage of ICA. By comparing the CCA
and ICA, CCA, with the correlation of signals, declared that both have similar qualitative results, but
differ in computational complexity. In contrast, ICA considers the statistical distribution of the same
samples into account. Correctly, the CCA was implemented to remove muscle artifacts from EEG
[21]. The author explored the dissimilarity over brain and muscle artifacts, and finally, the
components with the least autocorrelation were choosing to reject.
3.4. Empirical Mode Decomposition
EMD method is a practical and data-driven technique, whereas other methods depend on the selections of essential
functions, such as wavelet analysis [18]. Let the signal x(n) it can be decomposed as a linear combination of finite
number (N)bk(n) and residual r(n). A shifting process is done to calculate the IMF of the given signals,
at the end the original signal is constructed as,
x[n] = (mk =1) bk (n) + r ( n) (8)
EMD algorithm has one negative aspect of that the sensitivity of noise, which incurring mode
mixing complications, that the sensitivity of noise [22]. Another tailored EMD, to decompose
multivariate signals into multivariate IMFs. Therefore, the simultaneous analysis of intrinsic modes
across multiple channels, i.e., Multiple Empirical Mode decompositions, can more efficiently and
accurately remove artifacts related to broadband muscle artifacts.
3.5 Filtering Methods
In eliminating the artifacts, several filtering methods play a vital role in EEG analysis. Most basic
filtering methods are adaptive and wiener filtering methods. This paper, in brief, illustrates the
commonly used filtering approaches.
3.5.1 Adaptive Filtering
An illustration of the adaptive filter is as followed in figure 6. At the input stage, the primary input.
The primary input is a combination of both real EEG data and an artifact source, and expressed as:
EEG pri ( n ) = EEG pure ( n ) + N ( n ) (9)
Figure 6. Adaptive Filter System.
Where, EEGpri– represent the primary signal, EEGpure --represent the pure signal. ‘N’ is the
noise signal, which is a mixture of various artifacts. Another input to the filter is given to the
reference channel. To examine the pure signal, EEGpure, the least mean square, is used. As an extension to
LSM., the recursive least mean square (RLMS) is implemented, which is faster than LSM, but the drawback is high
calculation cost due to additional sensors to provide reference inputs [23] .
3.5.2. Wiener Filtering
7
First International Conference on Advances in Physical Sciences and Materials IOP Publishing
Journal of Physics: Conference Series 1706 (2020) 012093 doi:10.1088/1742-6596/1706/1/012093
In signal processing, the wiener filter solves the signal estimation problem. It is capable of estimating
the target process by time-invariant filtering of an observed noisy signal process of presumptuous
stationery and noise spectra. However, it is a statistical filtering method to examine and inspect the
true EEG data, employs a linear invariant filter to minimize the mean square error between the pure
EEG signal and estimated signal [24]. However, the limitation of extra reference channels overcome
by wiener filter but a bit complex in computation.
4. Hybrid Methods
To avail the benefit from established approaches, novel research approaches came into the scenario,
the conjunction of various methods known as hybrid methods, which means a combination of two or
more methods [25-27]. The fundamental hybrid strategies are discussed below.
Figure 7. Process flow of the EMD-BSS method.
Figure 7 represents the process flow of the EMD-BSS method, in which the raw contaminated EEG
data is decomposed by the EMD algorithm, combined with BSS to remove artifacts and to reconstruct
the clean EEG by applying the reverse algorithms. This hybrid method is the most powerful tool in
the removal of EMG artifacts from EEG signals; other side complexity is more in computing and
identifying the artifactual components.
4.1. Wavelet -BSS
Whilst WT failed, in the case, if the artifacts overlap with the spectral properties in the spectral
domain, this remained as the limitation of ICA, the conjunction of ICA with wavelet method has been
proposed to avoid the short comes of both and to pick the positive [28-31].
Figure 8. Block diagram of Wavelet –BSS Method.
Figure 8 shows the block diagram of the Wavelet-BSS method. Initially, the EEG signal data is
decomposed with WT, and EEG data is decomposed by WT, and then contaminated data components are
fed into the selected ICA algorithm. However, at the endpoint, artifacts are removed and reconstructed
using preserved wavelet components and disposed components.
4.2. BSS and Support Vector Machine
A further extended method of BSS in the hybrid model is BSS-SVM. Using BSS, the recorded raw
EEG data were decomposed into multiple components. In the subsequent level, extraction of
numerous features takes place such as temporal, spatial etc. Then the extracted features are given as
input to SVM classifier to examine the artifacts and to reconstruct the signal with the remaining
8
First International Conference on Advances in Physical Sciences and Materials IOP Publishing
Journal of Physics: Conference Series 1706 (2020) 012093 doi:10.1088/1742-6596/1706/1/012093
components as artifact-free signals. Figure 9 shows a schematic of the BSS-SVM method. Shocker et
al. [32] first applied BSS with SVM to eliminate artifacts related to eye blinking.
Figure 9. Block diagram of BSS –SVM Method.
5. Comparative Analysis
The different methodologies summarized in this paper are most widely used, removing the artifacts of
the EEG signal. Few techniques implemented to remove eye blinking and eye movement’s artifacts.
The majority of EEG related applications are real-time applications like bio medical-signal processing
can be booming with the help of the reference signal. Regression and filtering algorithms can be
efficiently executed to eliminate the various artifacts. Filter methods have more practically opted
methods for specific reference artifacts. The requirement of the reference channel that detains the
adaptive and regression algorithms in removal of the unsurpassed artifacts. WT transforms not make
the grade to identify artifacts that overlap with spectral properties. BSS techniques ensure flexibility
because it does not require any preceding data and extra reference channels.
Along with these BSS methods, ICA the separation is done, where the signal is converted into independent
components as ICCs with the decomposing process, and by discarding the artifacts, the new signal is
reconstructed. In CCA algorithm was implemented to separate the EMG potential from EEG
potential, and the computational time is less. Due to this CCA algorithm is applicable in real-time
application. In such circumstances, the BSS is not suitable due to which it works on a more significant
number of channels, and the accuracy in obtaining the truth is high. Zou et al. [33] following the ICA
algorithm was extended by a grouping of ICCs components; therefore, physiological and non-physiological artifacts
are identified from the EEG signal data. BSS method is incorporated with advantages of both ICA and CCA
into one single carcass vector analysis, which is used to reject the muscle artifact by extracting the
sources with maximal independence and maximal autocorrelation [34, 35]. Contrasting the convention
algorithms, machine learning-based approaches became new research projections to identify artifacts
using massive datasets.
6. Conclusion
EEG potentials are originated from the cerebral cortex, which is picked up with scalp electrodes. EEG
signals, however, continuously merge with unwanted surrounding signals. In fact, a wide variety of
techniques have been implemented to remove the undesired signal components known as artifacts,
and these artifact removal methods still require high accuracy and efficiency. This review paper
recaps the basic approaches and conclusions set by the researcher’s literature. The pros and cons are
emphasized for each method. Consequently, there is no most excellent preference method for
removing of all categories artifacts. As a result, the main focused future objective is to implement a
specific algorithm for sufficient attenuation of artifacts with better accuracy and efficiency.
In conclusion, there is a future scope where machine learning and traditional methods may
combine to develop new distinguished algorithms to achieve an effective removal of artifacts in EEG
signal analysis.
References
[1] Somers B and Bertrand A 2016 Removal of eye blink artifacts in wireless EEG sensor
networks using reduced-bandwidth canonical correlation analysis Journal of Neural
Engineering 13 066008
[2] Urigüen J A and Garcia-Zapirain B 2015 EEG artifact removal-state-of-the-art and
guidelines J Neural Eng 12 031001
9
First International Conference on Advances in Physical Sciences and Materials IOP Publishing
Journal of Physics: Conference Series 1706 (2020) 012093 doi:10.1088/1742-6596/1706/1/012093
[3] Islam M K, Rastegarnia A and Yang Z 2016 Methods for artifact detection and
removal from scalp EEG: A review Neurophysiologie Clinique/Clinical Neurophysiology
46 287-305
[4] Hamal A Q and bin Abdul Rehman A W 2013 Artifact Processing of Epileptic EEG
Signals: An Overview of Different Types of Artifacts. In: 2013 International Conference
on Advanced Computer Science Applications and Technologies: IEEE) pp 358-61
[5] Goncharova I I, McFarland D J, Vaughan T M and Wolpaw J R 2003 EMG
contamination of EEG: spectral and topographical characteristics Clinical
Neurophysiology 114 1580-93
[6] Shackman A J, McMenamin B W, Slagter H A, Maxwell J S, Greischar L L and
Davidson R J 2009 Electromyogenic Artifacts and Electroencephalographic Inferences
Brain Topography 22 7-12
[7] Klados M A, Papadelis C, Braun C and Bamidis P D 2011 REG-ICA: A hybrid
methodology combining Blind Source Separation and regression techniques for the
rejection of ocular artifacts Biomedical Signal Processing and Control 6 291-300
[8] Hillyard S A and Galambos R 1970 Eye movement artifact in the CNV
Electroencephalography and Clinical Neurophysiology 28 173-82
[9] Whitton J L, Lue F and Moldofsky H 1978 A spectral method for removing eye
movement artifacts from the EEG Electroencephalography and Clinical Neurophysiology
44 735-41
[10] Kumar P S, Arumuganathan R, Sivakumar K and Vimal C 2008 Removal of ocular
artifacts in the EEG through wavelet transform without using an EOG reference
channel Int. J. Open Problems Compt. Math 1 188-200
[11] Safieddine D, Kachenoura A, Albera L, Birot G, Karfoul A, Pasnicu A, Biraben A,
Wendling F, Senhadji L and Merlet I 2012 Removal of muscle artifact from EEG data:
comparison between stochastic (ICA and CCA) and deterministic (EMD and
wavelet-based) approaches EURASIP Journal on Advances in Signal Processing 2012
127
[12] Dabbakuti J K and Ratnam D V 2016 Characterization of ionospheric variability in
TEC using EOF and wavelets over low-latitude GNSS stations Advances in Space
Research 57 2427-43
[13] Neeli R, Dabbakuti J K, Chowdhary V R, Tripathi N K and Devanaboyina V R 2018
Modeling of local ionospheric time varying characteristics based on singular value
decomposition over low-latitude GPS stations Astrophysics and Space Science 363 182
[14] Dabbakuti J K, Devanaboyina V R and Kanchumarthi S R 2016 Analysis of local
ionospheric variability based on SVD and MDS at low-latitude GNSS stations Earth,
Planets and Space 68 94
[15] Dabbakuti J K, Mallika Y, Rao M V, Rao K R and Ratnam D V 2019 Modeling of GPS-
TEC using QR-decomposition over the low latitude sector during disturbed
geomagnetic conditions Advances in Space Research 64 2088-103
[16] Dabbakuti J R K K and Venkata Ratnam D 2017 Modeling and analysis of GPS-TEC
low latitude climatology during the 24th solar cycle using empirical orthogonal
functions Advances in Space Research 60 1751-64
[17] Berg P and Scherg M 1991 Dipole modelling of eye activity and its application to the
removal of eye artefacts from the EEG and MEG Clinical Physics and Physiological
Measurement 12 49
[18] Jung T-P, Makeig S, Bell A J and Sejnowski T J 1998 Central auditory processing and
neural modeling: Springer) pp 189-97
[19] Jung T-P, Makeig S, Westerfield M, Townsend J, Courchesne E and Sejnowski T J
2000 Removal of eye activity artifacts from visual event-related potentials in normal
and clinical subjects Clinical Neurophysiology 111 1745-58
10
First International Conference on Advances in Physical Sciences and Materials IOP Publishing
Journal of Physics: Conference Series 1706 (2020) 012093 doi:10.1088/1742-6596/1706/1/012093
[20] De Clercq W, Vergult A, Vanrumste B, Van Paesschen W and Van Huffel S 2006
Canonical correlation analysis applied to remove muscle artifacts from the
electroencephalogram IEEE transactions on Biomedical Engineering 53 2583-7
[21] Yang B, Zhang T, Zhang Y, Liu W, Wang J and Duan K 2017 Removal of
electrooculogram artifacts from electroencephalogram using canonical correlation
analysis with ensemble empirical mode decomposition Cognitive Computation 9 626-
33
[22] Xu X, Chen X and Zhang Y 2018 Removal of muscle artefacts from few-channel EEG
recordings based on multivariate empirical mode decomposition and independent
vector analysis Electronics Letters 54 866-8
[23] He P, Wilson G and Russell C 2004 Removal of ocular artifacts from
electroencephalogram by adaptive filtering Medical and biological engineering and
computing 42 407-12
[24] Somers B, Francart T and Bertrand A J J o n e 2018 A generic EEG artifact removal
algorithm based on the multi-channel Wiener filter 15 036007
[25] Dabbakuti J R K K and G B L 2019 Application of Singular Spectrum Analysis Using
Artificial Neural Networks in TEC Predictions for Ionospheric Space Weather IEEE
Journal of Selected Topics in Applied Earth Observations and Remote Sensing 12 5101-7
[26] Dabbakuti J R K K, Peesapati R, Yarrakula M, Anumandla K K and Madduri S V
2020 Implementation of storm-time ionospheric forecasting algorithm using SSA–
ANN model. In: IET Radar, Sonar & Navigation: Institution of Engineering and
Technology) pp 1249-55
[27] Dabbakuti J R K K, Jacob A, Veeravalli V R and Kallakunta R K 2020 Implementation
of IoT analytics ionospheric forecasting system based on machine learning and
ThingSpeak. In: IET Radar, Sonar & Navigation: Institution of Engineering and
Technology) pp 341-7
[28] Azzerboni B, Carpentieri M, La Foresta F and Morabito F 2004 Neural-ICA and
wavelet transform for artifacts removal in surface EMG. In: 2004 IEEE International
Joint Conference on Neural Networks (IEEE Cat. No. 04CH37541): IEEE) pp 3223-8
[29] Winkler I, Brandl S, Horn F, Waldburger E, Allefeld C and Tangermann M 2014
Robust artifactual independent component classification for BCI practitioners Journal
of neural engineering 11 035013
[30] Huppert T J, Diamond S G, Franceschini M A and Boas D A 2009 HomER: a review
of time-series analysis methods for near-infrared spectroscopy of the brain Applied
optics 48 D280-D98
[31] Calcagno S, La Foresta F and Versaci M 2014 Independent component analysis and
discrete wavelet transform for artifact removal in biomedical signal processing
American Journal of Applied Sciences 11 57
[32] Shoker L, Sanei S and Chambers J 2005 Artifact removal from electroencephalograms
using a hybrid BSS-SVM algorithm IEEE Signal Processing Letters 12 721-4
[33] Zou Y, Nathan V and Jafari R 2014 Automatic identification of artifact-related
independent components for artifact removal in EEG recordings IEEE journal of
biomedical and health informatics 20 73-81
[34] Chen X, Peng H, Yu F and Wang K 2017 Independent vector analysis applied to
remove muscle artifacts in EEG data IEEE Transactions on Instrumentation and
Measurement 66 1770-9
[35] Chen X, Liu A, Chen Q, Liu Y, Zou L and McKeown M J 2017 Simultaneous ocular
and muscle artifact removal from EEG data by exploiting diverse statistics Computers
in biology and medicine 88 1-10
11
