FORM 2
THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003
COMPLETE SPECIFICATION (See Section 10 and Rule 13)
Title of invention:
BLINK ARTIFACT REMOVAL FROM EEG SIGNALS USING MULTIVARIATE VARIATIONAL MODE DECOMPOSITION
Applicant
Tata Consultancy Services Limited A company Incorporated in India under the Companies Act, 1956
Having address:
Nirmal Building, 9th floor,
Nariman point, Mumbai 400021,
Maharashtra, India
Preamble to the description
The following specification particularly describes the invention and the manner in which it is to be performed.

TECHNICAL FIELD [001] The disclosure herein generally relates to removal of blink artifacts from electroencephalogram (EEG) signals, and, more particularly, to systems and methods for removal of blink artifacts using a Multivariate Variational Mode Decomposition (MVMD) method.
BACKGROUND [002] Electroencephalogram (EEG) is a non-invasive technique for recording a conglomeration of electric potential generated in neurons. Currently, this technique is being used for various medical and non-medical applications such as brain-computer interfaces. EEG signals provide excellent time resolution compared to other methods like Functional magnetic resonance imaging (fMRI). However, EEG signals are easily contaminated by other signals like electromagnetic radiations or power line noise that creates inductive currents in the cables which are connected to a subject. Another major artifact source is the electrophysiological responses of other organs like eye, heart, muscles and so on. Presence of these artifacts degrades the performance of EEG-based applications and analysis. Eye-blinks and eye movements are the most problematic artifact that affects EEG signals. Removal of blink artifacts is challenging as these are non-stationary and non-linearly mixed with EEG. Furthermore, conventional approaches for blink artifact removal are dependent on experts for manual identification of the blinks and also result in a decrease in quality of reconstructed EEG signal as compared to the raw EEG signal.
SUMMARY
[003] Embodiments of the present disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems.
[004] In an aspect, there is provided a processor implemented method comprising the steps of: decomposing, via one or more hardware processors, a raw electroencephalogram (EEG) signal having one or more channels into a plurality of

Intrinsic Mode Functions (IMFs) using a Multivariate Variational Mode Decomposition (MVMD) method; identifying, via the one or more hardware processors, a dominant frequency for each IMF in the plurality of IMFs associated with one of the one or more channels; and detecting one or more blink artifacts in one or more IMFs having the dominant frequency in an empirically determined range of frequencies based on peaks in the one or more IMFs and width of the peaks, wherein the peaks represent blink peaks and the width of the peaks represent blink duration respectively; wherein the detected one or more blink artifacts in one or more IMFs associated with one of the one or more channels is indicative of the presence of the one or more blink artifacts in the other channels of the one or more channels.
[005] In another aspect, there is provided a system comprising: one or more data storage devices operatively coupled to one or more hardware processors and configured to store instructions configured for execution via the one or more hardware processors to: decompose a raw electroencephalogram (EEG) signal having one or more channels into a plurality of Intrinsic Mode Functions (IMFs) using a Multivariate Variational Mode Decomposition (MVMD) method; identify a dominant frequency for each IMF in the plurality of IMFs associated with one of the one or more channels; detecting one or more blink artifacts in one or more IMFs having the dominant frequency in an empirically determined range of frequencies based on peaks in the one or more IMFs and width of the peaks, wherein the peaks represent blink peaks and the width of the peaks represent blink duration respectively, wherein the detected one or more blink artifacts in one or more IMFs associated with one of the one or more channels is indicative of the presence of the one or more blink artifacts in the other channels of the one or more channels.
[006] In yet another aspect, there is provided a computer program product comprising a non-transitory computer readable medium having a computer readable program embodied therein, wherein the computer readable program, when executed on a computing device, causes the computing device to: decompose a raw electroencephalogram (EEG) signal having one or more channels into a plurality of Intrinsic Mode Functions (IMFs) using a Multivariate Variational Mode

Decomposition (MVMD) method; identify a dominant frequency for each IMF in the plurality of IMFs associated with one of the one or more channels; detecting one or more blink artifacts in one or more IMFs having the dominant frequency in an empirically determined range of frequencies based on peaks in the one or more IMFs and width of the peaks, wherein the peaks represent blink peaks and the width of the peaks represent blink duration respectively, wherein the detected one or more blink artifacts in one or more IMFs associated with one of the one or more channels is indicative of the presence of the one or more blink artifacts in the other channels of the one or more channels.
[007] In accordance with an embodiment of the present disclosure, the one or more hardware processors are configured to obtain the one or more channels from regions proximate to one or more of frontal, temporal and central lobe regions of a subject being monitored.
[008] In accordance with an embodiment of the present disclosure, the one or more hardware processors are configured to identify the dominant frequency by: performing a Discrete Fourier transform on each IMF in the plurality of IMFs to obtain frequencies up to fs/2 and power associated thereof, wherein fs is a sampling rate of the raw EEG signal; and identifying a frequency from the obtained frequencies corresponding to a maximum power as the dominant frequency for the associated IMF.
[009] In accordance with an embodiment of the present disclosure, the empirically determined range of frequencies is 2 to 4 Hz.
[010] In accordance with an embodiment of the present disclosure, the one or more hardware processors are configured to remove the detected one or more blink artifacts from the raw EEG signal to obtain one or more blink free channels by: interpolating one or more non-blink regions preceding and succeeding the detected one or more blink artifacts to replace the detected one or more blink artifacts and obtain one or more interpolated IMFs in the plurality of IMFs associated with the one or more channels; and reconstructing the one or more channels associated with the detected one or more blink artifacts by adding the one

or more interpolated IMFs and remaining IMFs in the plurality of IMFs to obtain the one or more blink free channels.
[011] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.
BRIEF DESCRIPTION OF THE DRAWINGS
[012] The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles:
[013] FIG.1 illustrates an exemplary block diagram of a system for blink artifact removal from an electroencephalogram (EEG) signal, in accordance with some embodiments of the present disclosure.
[014] FIG.2 illustrates a comparison between a blink removal processing pipeline of a state-of-the-art approach and the present disclosure.
[015] FIG.3A and FIG.3B illustrate an exemplary flow diagram of a computer implemented method for blink artifact removal from an EEG signal, in accordance with some embodiments of the present disclosure.
[016] FIG.4 illustrates decomposition of a bivariate signal consisting of a mixture of tones using Multivariate Variational Mode Decomposition (MVMD) method as known in the art.
[017] FIG.5 illustrates a raw EEG signal (a) with example Intrinsic Mode Functions (IMFs) having different dominant frequencies Ø in (b), (c) and (d) respectively as known in the art.
[018] FIG.6 illustrates a sample plot of raw and processed EEG signals after removing blink artifacts by the Independent Component Analysis (ICA) method and the method in accordance with some embodiments of the present disclosure.
[019] FIG.7 illustrates performance metrics for the method in accordance with some embodiments of the present disclosure on synthetic data when compared with the ICA method.

[020] FIG.8 illustrates percentage change in powers of theta, alpha and beta bands from raw EEG and the filtered EEG in the non-blink regions using the method of the present disclosure when compared with the ICA method.
[021] FIG.9 illustrates a boxplot of correlation values obtained in the non-blink regions for a high-end EEG signal using the method of the present disclosure and the ICA method.
DETAILED DESCRIPTION OF EMBODIMENTS [022] Exemplary embodiments are described with reference to the accompanying drawings. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. Wherever convenient, the same reference numbers are used throughout the drawings to refer to the same or like parts. While examples and features of disclosed principles are described herein, modifications, adaptations, and other implementations are possible without departing from the scope of the disclosed embodiments. It is intended that the following detailed description be considered as exemplary only, with the true scope being indicated by the following claims.
[023] Presence of ocular artifacts such as eye blinks and other eye movement related artifacts in electroencephalogram (EEG) signals degrades the overall performance of any EEG-based applications. Removal of blink artifacts is challenging as these are non-stationary and non-linearly mixed with EEG. In some cases, electrooculography (EOG) signals have been used as a reference for detecting blink artifacts in the EEG signal. However, this requires a separate sensor for recording horizontal and vertical eye movements. Broadly two approaches have been used by state-of-the art: a) band-pass or adaptive filtering-based approaches where the blink region is first detected and then corrected and b) multi-channel algorithms which compute the correlation between EEG signals recorded at different scalp positions. Conventional bandpass filtering fails to effectively remove undesired blink components from the EEG signal as they are oscillated with time-varying frequency and are non-linearly generated. The adaptive filter-based

approach reported does not require detection of blink region. However, detection of exact locations of the blink segments is challenging.
[024] Multi-channel approaches like Independent component analysis (ICA) are currently widely used to remove blink artifacts. ICA does not detect the blink regions precisely since it assumes the blink components to be linearly located but can work without a reference channel. It decomposes the raw EEG signal into several independent components based on the spatio-temporal characteristics of the signal. Higher the number of EEG channels, better is the performance of ICA. Also, the components corresponding to eye-blinks needs to be identified manually by experts, which is time consuming. There is also literature available that show a decrease in quality of the reconstructed EEG data through ICA when compared to the ground truth for low resolution EEG devices in a systematic way. An extension of multivariate empirical mode decomposition (MEMD) for blink removal by using the EOG as a reference using a separate EOG sensor has been proposed in a literature. However, the MEMD approach cannot be used directly as it fails in terms of non-mode alignment and mode-mixing.
[025] Applicant has addressed these problems by providing systems and methods for firstly automatic detection of blink artifacts using the mode-alignment property of the Multivariate Variational Mode Decomposition (MVMD) method and Intrinsic Mode Functions (IMFs) generated using the MVMD method. Secondly, interpolation is used to reconstruct the raw EEG signal post removal of the automatically detected blink artifacts such that EEG related information in the non-blink region of the EEG signal is relatively unaffected resulting in higher correlation between the raw and processed EEG signals.
[026] Referring now to the drawings, and more particularly to FIG. 1 through FIG.9, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments and these embodiments are described in the context of the following exemplary system and/or method.
[027] FIG.1 illustrates an exemplary block diagram of a system 100 for blink artifact removal from an electroencephalogram (EEG) signal, in accordance

with some embodiments of the present disclosure. In an embodiment, the system 100 includes one or more processors 104, communication interface device(s) or input/output (I/O) interface(s) 106, and one or more data storage devices or memory 102 operatively coupled to the one or more processors 104. The one or more processors 104 that are hardware processors can be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, graphics controllers, logic circuitries, and/or any devices that manipulate signals based on operational instructions. Among other capabilities, the processor(s) are configured to fetch and execute computer-readable instructions stored in the memory. In the context of the present disclosure, the expressions ‘processors’ and ‘hardware processors’ may be used interchangeably. In an embodiment, the system 100 can be implemented in a variety of computing systems, such as laptop computers, notebooks, hand-held devices, workstations, mainframe computers, servers, a network cloud and the like.
[028] I/O interface(s) 106 can include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like and can facilitate multiple communications within a wide variety of networks N/W and protocol types, including wired networks, for example, LAN, cable, etc., and wireless networks, such as WLAN, cellular, or satellite. In an embodiment, the I/O interface(s) can include one or more ports for connecting a number of devices to one another or to another server.
[029] The memory 102 may include any computer-readable medium known in the art including, for example, volatile memory, such as static random access memory (SRAM) and dynamic random access memory (DRAM), and/or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes. In an embodiment, one or more modules (not shown) of the system 100 can be stored in the memory 102.
[030] FIG.2 illustrates a comparison between a blink removal processing pipeline 200 of a state-of-the-art approach and the present disclosure. The state-of-the-art approach depicted in a box with broken line is the commonly used ICA

method while the box with solid line depicts the method of the present disclosure at a high level. It may be noted that in both the approaches a C-channel raw EEG signal is processed for blink artifact removal. In the ICA method, the steps involved are: a) decomposition of a time series EEG signal into independent components using a EEGLAB toolbox b)
identification of components (intermediate representation of channel) corresponding to eye blinks - performed manually by an EEG expert by analyzing scalp topology plots and the corresponding power spectrum; c) removal of the identified blink components which also include information in non-blink regions within the components; and d) reconstruction of the EEG signal to obtain a processed EEG signal without the blink components wherein the degree of correlation between the raw and processed EEG signal is low considering removal of the blink components. In the method of the present disclosure, the steps involved are: a) decomposition of the time series EEG signal into intermediate modes or Intrinsic Mode Functions (IMFs); b) identification of blink artifacts in the IMFs automatically without dependency on human experts; c) removal of only the blink artifacts and d) reconstruction of the EEG signal to obtain a processed EEG signal without the blink artifacts by interpolating some non-blink regions in the proximity of the blink artifacts. Considering only the blink artifacts are removed, there is no loss of information in the non-blink region, thereby resulting in a relatively higher correlation between the raw and processed EEG signal compared to the art.
[031] FIG.3A and FIG.3B illustrate an exemplary flow diagram of a computer implemented method 300 for blink artifact removal from an EEG signal, in accordance with some embodiments of the present disclosure. In an embodiment, the system 100 includes one or more data storage devices or memory 102 operatively coupled to the one or more processors 104 and is configured to store instructions configured for execution of steps of the method 300 by the one or more processors 104. The steps of the method 300 will now be explained in detail with reference to the components of the system 100 of FIG.1. Although process steps, method steps, techniques or the like may be described in a sequential order, such processes, methods and techniques may be configured to work in alternate orders.

In other words, any sequence or order of steps that may be described does not necessarily indicate a requirement that the steps be performed in that order. The steps of processes described herein may be performed in any order practical. Further, some steps may be performed simultaneously.
[032] Accordingly, in an embodiment of the present disclosure, the one or more processors 104, are configured to decompose, at step 302, the raw EEG signal having one or more channels into a plurality of Intrinsic Mode Functions (IMFs) using the MVMD method. In an embodiment, one or more channels are obtained from regions proximate to one or more of frontal, temporal and central lobe regions of a subject being monitored. Typically, a single channel EEG signal or a multi-channel EEG signal obtained from the regions mentioned includes blink artifacts in all the channels. The MVMD method is an extension of variational mode decomposition (VMD) method. In VMD, a one-dimensional signal is decomposed into K number of modes uk (t) as,

such that the sum of the bandwidths of all the modes is minimized and the signal gets reconstructed at least in least square sense or ideally fully, by summing up the K modes together. The MVMD extends this approach of VMD to multivariate data x(t) = [x1(t), x2(t),x3(t),….xm(t)] by extracting K multivariate modulated oscillations uk(t) on similar grounds as in equation (1) with uk(t) = [u1(t), u2(t), u3(t),….um(t)]. The resulting cost function is given by,

where is a complex valued signal with signal frequency wkcomponent across
M channels, subject to the constraint that with m =
1,2,3, …M. The method is beneficial due to i) mode alignment property, ii) quasi-orthogonality across modes, (iii) separation of multivariate modulated oscillations

inherent in the data and (iv) robustness to noise. FIG.4 illustrates decomposition of a bivariate signal consisting of a mixture of tones using Multivariate Variational Mode Decomposition (MVMD) method as known in the art. The ability of MVMD to identify and align principal modulated oscillations present in the data across multiple channels and modes may be noted in the figure. The input data was a bivariate signal whose individual components were a mixture of a 36-Hz sinusoid that was common to both the data channels (Channel 1 and Channel 2); a 2-Hz tone in the Channel 1 and a 24-Hz tone in the Channel 2. The K = 3 number of modes obtained by applying MVMD are shown. It may be observed that all modes are aligned in terms of their frequency content: the 36 Hz tone present in all data channels is localized in a single mode u3. The 2-Hz signal is located in the Channel 1 of u1 while the 24-Hz tone is localized in the Channel 2 of u2.
[033] In an embodiment of the present disclosure, the one or more processors 104, are configured to identify, at step 304, a dominant frequency for each IMF in the plurality of IMFs associated with one of the one or more channels. In an embodiment, firstly at step 304a, a Discrete Fourier transform is performed on each IMF in the plurality of IMFs to obtain frequencies up to fs/2 and power associated thereof, wherein fs is a sampling rate of the raw EEG signal. The Discrete Fourier transform may be Fast Fourier Transform (FFT) in one embodiment. Once the frequencies and associated power are obtained, at step 304b, a frequency that corresponds to a maximum power is identified as the dominant frequency for the associated IMF.
[034] Upon computing the dominant frequency for each of the IMFs of the selected single channel, it is observed that not all the IMFs with dominant frequency below 4 Hz can be used to detect blink artifacts automatically. So, a threshold value of IMFs with a dominant frequency range is empirically derived based on analysis of some samples of the EEG signal. Accordingly, in an embodiment of the present disclosure, the one or more processors 104, are configured to detect, at step 306, one or more blink artifacts in one or more IMFs having the dominant frequency in an empirically determined range of frequencies. In an embodiment, the empirically determined range of frequencies is 2 to 4 Hz. The step of detecting the one or more

blink artifacts does not require human intervention, as in the art, and is based on peaks in the one or more IMFs and width of the peaks, wherein the peaks represent blink peaks and the width of the peaks represent blink duration, respectively.
[035] In an embodiment of the present disclosure, to determine peaks in an IMF, the IMFs are converted to a positive IMF using the formula IMF[n] = |IMF[n]| on all samples of the IMFs where |x| gives an absolute value of x (i.e. x = x, if x ≥ 0 and x = -x otherwise). Then, the local maxima of the positive IMFs are located. For each maximum, the nearest troughs on either side of the maximum are determined. The prominence of this maximum is the lower of the two differences between the maximum and the two troughs. Finally, the maxima that has a prominence of at least 4 times the median value of the positive IMFs are considered peaks of blink artifacts. The time-locations on either side of the maximum, where the positive IMF reaches half the prominence below the peak height, are determined. Width of each blink artifact is twice the difference between these two locations. In an embodiment, the MATLAB® function findpeaks may be used suitably to obtain the peaks and their widths as described above. It may be noted that on account of the use of the MVMD, the detected one or more blink artifacts in one or more IMFs associated with one of the one or more channels is indicative of the presence of the one or more blink artifacts in the other channels of the one or more channels.
[036] Once the blink artifacts are detected, the blink regions in all the IMFs below 4 Hz are replaced with zeros. This is treated as missing data and few data points before blink start and few data points after blink end are used to fit a linear trend in the blink region which is known as interpolation. In other words, data from preceding and succeeding non-blink regions is used to interpolate the missing blink artifact portion. Further, all the IMFs of the single selected channel are added together to get the original signal without blinks. The process of detecting blink artifacts may be performed on a single channel and then the blink artifact regions are processed by replacing the blink artifacts with interpolated points is repeated for all the channels. Accordingly, in an embodiment of the present disclosure, the one or more processors 104, are configured to remove, at step 308, the detected one

or more blink artifacts from the raw EEG signal to obtain one or more blink free channels that may be used effectively for any neuro-engineering based analysis or applications like brain computer interfaces. In an embodiment, at step 308a, one or more non-blink regions preceding and succeeding the detected one or more blink artifacts are interpolated to replace the detected one or more blink artifacts resulting in one or more interpolated IMFs in the plurality of IMFs associated with the one or more channels. The interpolated IMFs and the remaining IMFs in the plurality of IMFs are then added at step 308b, to reconstruct the one or more channels associated with the detected one or more blink artifacts to obtain the one or more blink free channels.
[037] An algorithm for the blink artifact removal method 300 of the present disclosure using MVMD may be represented as provided below. Algorithm: Blink artifact removal using MVMD Input: Raw EEG data: C channels each at sampling rate fs. Let each channel last from time-index n = 0 to n= N-1. Output: EEG data, of the same dimensions as input
without blink artifacts Procedure: 1: Decompose the given multi-channel EEG data into K IMFs using MVMD.
Denote the Cth channel’s kth IMF by 2:
3: for k = 0 to K-1 do
4: if
5:
6: break
7: end if
8: end for
9: Detect peaks and peak widths in b [n] 10: I←{ }
11: for each detected peak (location p, width w) do 12:

13: end for
14: for c = 0 to C-1 do
15: for k = 0 to K-1 do
16: if Ø(uk,1[.]≤∆ then
17: uk,c [n], n ∈ I ← Data interpolated from uk,c [n], n ∉ I
18: end if
19: end for
20: vc[n] =∑Kk-1=0 u�k,c[n],n∈{0,….N - 1}
21: end for
[038] As explained above, the algorithm uses two thresholds, δ = 2Hz and ∆= 4Hz. The function Ø(u[. ]) returns the dominant frequency of the signal (or IMF). The square brackets indicate that the algorithm is applied on digital signals. First the multi-channel EEG signals are decomposed into NXC IMFs using MVMD, where N is the number of samples. Normally ocular artifacts like eye blinks and eye movements are represented as low frequency (below 4 Hz) signals. Hence a straightforward approach would be to reject all IMFs having dominant frequency below 4 Hz. However, this would also remove some low frequency and slow varying useful components of the EEG signal. Thus, in accordance with the present disclosure, the blink artifacts are identified from the IMFs. Eye blinks have higher amplitude compared to normal EEG signal but conventional peak detection algorithms on the raw EEG signal produce many false positives. However, the IMFs that correspond to blinks are smoother due to the band-limited spectral content. As a result, peak detection works better on the IMFs. Due to the mode-alignment property of MVMD, the dominant frequency for a given IMF is similar across channels and hence it is sufficient to consider the dominant frequency identified for a single channel.
[039] FIG.5 illustrates a raw EEG signal (a) with example Intrinsic Mode Functions (IMFs) having different dominant frequencies ∅ in (b), (c) and (d) respectively as known in the art. It may be noted that the IMFs having lower dominant frequency are smoother than the original signal, hence peak detection works better. It is also evident from the illustration that blinks are most prominent

in IMFs having dominant frequency closer to 4 Hz (c). Thus, in accordance with the present disclosure, the IMFs that have the dominant frequency in the range � = 2, ∆ are chosen for peak detection. Steps 3 through 8 in the algorithm above depict selection of IMFs while step 9 enables peak detection. For each detected peak, the corresponding width is used to determine the duration of the blink artifact as shown in steps 10 through 13 of the above algorithm. All IMFs that have a dominant frequency below ∆ Hz are modified as follows: samples outside the blink artifacts are linearly interpolated to replace the samples inside the blink artifacts. The reconstruction involves addition of the modified IMFs or the interpolated IMFs with the remaining IMFs resulting in a blink free EEG signal as shown in steps 14 through 21 of the above algorithm.
EXPERIMENTAL EVALUATION
Datasets used: 3 different datasets were used - Synthetically generated EEG data, High end EEG (Covert Shift Dataset) and CogBeacon Dataset.
[040] Synthetically generated EEG data: Synthetic EEG signal was generated using the tools provided by Yeung N et al. (“Simulated EEG data generator 2019). First a clean EEG data is generated and then blinks are added at known locations. 10 such 4-channel EEG data are created with a sampling frequency of 220 Hz and various SNR values. Thus, in this data set, blink start and end locations are known. This effectively helps in validation of blink removal algorithms.
[041] High end EEG (Covert Shift Dataset): This dataset provided by Treder et al. in “Brain computer interfacing using modulations of alpha activity induced by covert shifts of attention” in Journal of neuro-engineering and rehabilitation 2011 aims to study whether visual attention shifts in different pairs of directions can be differentiated via alpha wave activity in the brain. 8 healthy subjects were asked to fixate and covertly shift attention alternately. A 60-channel actiCAP® EEG device was used to record brain activations at a sampling rate of 1000 Hz, along with 2 - channel EOG. This dataset

was used to compare the performance of the method of the present disclosure for blink artifact removal and using EOG channels as ground truth for blink positions. [042] CogBeacon Dataset: CogBeacon is a publicly available multimodal dataset provided by Papakostas et al. in “CogBeacon: A multi-modal dataset and data collection platform for modeling cognitive fatigue 2019”. It consists of 76 sessions of EEG data collected from 19 male and female users performing different versions of the Wisconsin Card Sorting Test (WCST) provided by Monchi et al. in “Wisconsin card sorting revisited: distinct neural circuits participating in different stages of the task identified by event-related functional magnetic resonance imaging 2001” and Berg in “A simple objective technique for measuring flexibility in thinking” for testing the ability to display flexibility in thinking. The system provides feedback on whether a particular match is right or wrong. The matching rule changes frequently, and the user has to figure out the rule based on the feedback given. Two modified versions (V1 and V2) were also created based on the number of available options for the user to choose the cards, in each turn of this game, so as to increase the computational demands of the task. The raw EEG data was captured using a consumer grade Muse headset 2, sampled
at a frequency of 220 Hz; from 4 EEG electrodes placed at locations AF7, AF8, TP9, and TP10 respectively, as per the International Standard 10-20 system of EEG electrode scalp locations. This was accompanied by a fatigue self-report - indicated by a push button placed in front of the subject while performing the task. The dataset also contains facial key points captured using a camera and user performance statistics during the task. This information was used to classify the FATIGUE and NO-FATIGUE states of a subject with fatigue self-report taken as ground truth. The method of the present disclosure was used on this dataset to remove blink artifacts and compare the overall classification accuracy at the end.
EVALUATION METRICS

[043] Signal-to-error ratio (SER): SER is a measure of how much the non-blink EEG regions are getting altered by the noise cleaning technique and is expressed as,

where M denotes the number of EEG channels, x is the raw EEG data and d is the error in processed EEG data in the non-blink region given by d = (raw EEG -filtered EEG). Ideally in the non-blink regions, the processed EEG should be same as the raw EEG giving a value of d = 0. E{.} is the power of the signal and p is the weight obtained from each channel defined as,

Higher values of SER indicate better noise cleaning performance.
[044] Correlation: the correlation between the raw EEG and the processed EEG in both blink and blink-free segments are also considered as a metric.
[045] Variance-based metric (V): It is defined as the ratio of variance in the processed blink region to the variance in the blink region. Lesser the value of V, better is the performance of the noise cleaning approach.
[046] Percentage change in band power: This is calculated for theta, alpha and beta bands. Theoretically, these bands are devoid of blink artifacts and hence, any blink removal approach should not alter these band powers.
[047] Classification accuracy: The classification accuracy of detecting FATIGUE and NO-FATIGUE states was computed using the CogBeacon dataset. This helps to establish the importance of blink removal and its contribution in the overall assessment of cognitive states from the EEG signal.
RESULTS
[048] Performance on the synthetically generated EEG data: In synthetically generated EEG signal, the exact blink start and end locations are

known. FIG.6 illustrates a sample plot of raw and processed EEG signals after removing blink artifacts by the Independent Component Analysis (ICA) method and the method in accordance with some embodiments of the present disclosure (PD). It is observed that the signal obtained by the method of the present disclosure correlates well with the raw signal. Similar plots were obtained for all the datasets. FIG.7 and FIG.8 show evaluation metrics calculated on the synthetically generated EEG data. Particularly, FIG.7 illustrates performance metrics for the method in accordance with some embodiments of the present disclosure (PD) on synthetic data. It is observed that SER is high for the method of the present disclosure compared to ICA which is indicative of the better performance of the method of the present disclosure. The correlation values obtained for both the blink and non-blink regions are also shown in FIG.7. It is noted that for both the regions, correlation is higher in case of the method of the present disclosure compared to that in the ICA based method. The lesser value of variance-based metric (V) in FIG.7 for the method of the present disclosure is also indicative of better performance in comparison to the ICA based method. FIG.8 illustrates percentage change in powers of theta, alpha and beta bands from raw EEG and the filtered EEG in the non-blink regions using the method of the present disclosure (PD) and the ICA method. It is observed that the change is less in the method of the present disclosure in comparison to the ICA method. This can be attributed to the fact that the ICA method rejects the whole component identified as blink component and hence leads to loss of valuable data in the frequency bins outside of the blinks also.
[049] Performance on high end EEG dataset: FIG.9 illustrates a boxplot of correlation values obtained in the non-blink regions for a high-end EEG signal using the method of the present disclosure (PD) and the ICA method. The correlation values are averaged over 57 channels. The blink regions are identified using the vertical EOG data. A total of one-minute data from each of the 8 participants were taken for the sake of brevity and similar results were found across other windows in the entire dataset. It
is seen that the correlation obtained is good in both the cases; however, the method of the present disclosure performs slightly better than ICA.

ICA: 0.9788 (+0.0133) Max: 0.9954, Min: 0.9546 and
Method of the present disclosure: 0.9869 (+0.0094) Max: 0.9989, Min: 0.9690).
[050] Performance on CogBean dataset: The raw EEG data is processed using the method of the present disclosure and the ICA method separately. The raw EEG data consists of 4 channels giving independent components using ICA. Hence, a maximum of 2 components were removed to avoid loss of valuable data during the reconstruction phase. In the method of the present disclosure, the IMFs were generated using MVMD and peaks were detected as explained in Algorithm 1. Post reconstruction of the blink-free signal, non-overlapping windows of duration 2 seconds were used to compute a total of 80 features (20 features per channel). These features consisted of 9 morphological features such as 5 EEG band powers (delta, theta, alpha, beta and gamma) along with 4 band power ratios of alpha, beta and theta bands. Remaining 11 are statistical features such as mean, variance, standard deviation, kurtosis, skewness, maximum, minimum and three Hjorth parameters. In total 4749 NO -FATIGUE and 2193 FATIGUE instances were obtained. A total of 3 such feature sets were generated using 3 different processing pipelines i.e. a) without blink removal method; b) blink removal using the ICA method; c) blink removal using the method of the present disclosure. Class imbalance was then handled using Synthetic Minority Over-sampling Technique (SMOTE). After performing SMOTE, the new feature set consisted of 9135 instances of which 4386 are FATIGUE instances (about 48% of total instances). This feature set was then used to learn a Random Forest classifier and different metrics were calculated for performance evaluation.
[051] Table I reports the classification accuracy and f-score obtained using various approaches. Table I: Comparative classification accuracy on CogBeacon dataset.

Approach Classification accuracy f-score
Without blink removal 84.94+0.95 0.89
Blink removal using ICA 89.94+0.92 0.93

Blink removal using the 87.92+0.96 0.91
method of the present
disclosure
Without blink removal 87.93+0.89 0.88
Blink removal using ICA 91.97+0.85 0.92
Blink removal using the 90.12+0.76 0.90
method of the present
disclosure
The upper half of the table corresponds to the original, unbalanced dataset and the bottom half to the balanced dataset obtained through SMOTE algorithm. Without any blink removal, a classification accuracy of 84.94%, which is quite high compared to that reported in the state of the art (Papakostas et al.), i.e. 67%. One major difference is that the state of the art has used band power values provided by MUSE device itself, whereas, the present evaluation uses derived beta, delta and gamma band powers from the raw EEG. From Table I it is evident that blink removal using the ICA method performs well in all cases, where as method of the present disclosure is also at par in all respects.
[052] Based on the above description and the experimental evaluations of the approach of the present disclosure, it may be concluded that the method of the present disclosure is able to successfully clean the eye blink artifacts from the raw EEG data. The method of the present disclosure outperforms the state of the art, ICA-based approach on synthetically generated EEG dataset and low-resolution EEG dataset. The classification accuracy obtained on the publicly available CogBeacon dataset is comparable with that obtained using the state-of-the-art approach (ICA). However, the method of the present disclosure is preferable over the ICA-based method for the following reasons: i) the major concern in the ICA-based method is that, once the independent components are obtained, it requires manual intervention of an expert to identify the blink-related components. The method of the present disclosure enables automatic detection of blink artifacts from IMFs and hence, manual intervention is not required. ii) Since the number of independent components that contribute to form a complete EEG signal is not

known, the number of distinct components obtained post ICA is typically set as the number of input EEG channels. Hence, for low resolution EEG devices having small number of input channels (for example 4-channel MUSE device), the number of components is small. In such a scenario, it is difficult to correctly identify the artifact components and hence the ICA does not perform well for such devices. iii) ICA rejects the whole artifact component which may contain other useful EEG components as well. Removal of such components leads to removal of the accompanied EEG data as evident from FIG. 6 and FIG. 8. In accordance with the method of the present disclosure, only the blink artifacts are detected and removed followed by reconstruction of the EEG signal. Hence, other useful information contained in alpha, beta or theta bands are not affected. Lastly, iv) ICA is a source separation-based algorithm, hence, requires the spatial locations of the electrodes (i.e. spatial distribution of the individual signal sources) as well a large number of channels to perform well. On the other hand, the method of the present disclosure is not restricted to any such constraint and can be applied to a variety of EEG devices available in the market.
[053] The written description describes the subject matter herein to enable any person skilled in the art to make and use the embodiments. The scope of the subject matter embodiments is defined by the claims and may include other modifications that occur to those skilled in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims or if they include equivalent elements with insubstantial differences from the literal language of the claims.
[054] It is to be understood that the scope of the protection is extended to such a program and in addition to a computer-readable means having a message therein; such computer-readable storage means contain program-code means for implementation of one or more steps of the method, when the program runs on a server or mobile device or any suitable programmable device. The hardware device can be any kind of device which can be programmed including e.g. any kind of computer like a server or a personal computer, or the like, or any combination thereof. The device may also include means which could be e.g. hardware means

like e.g. an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of hardware and software means, e.g. an ASIC and an FPGA, or at least one microprocessor and at least one memory with software processing components located therein. Thus, the means can include both hardware means and software means. The method embodiments described herein could be implemented in hardware and software. The device may also include software means. Alternatively, the embodiments may be implemented on different hardware devices, e.g. using a plurality of CPUs.
[055] The embodiments herein can comprise hardware and software elements. The embodiments that are implemented in software include but are not limited to, firmware, resident software, microcode, etc. The functions performed by various components described herein may be implemented in other components or combinations of other components. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can comprise, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
[056] The illustrated steps are set out to explain the exemplary
embodiments shown, and it should be anticipated that ongoing technological
development will change the manner in which particular functions are performed.
These examples are presented herein for purposes of illustration, and not limitation.
Further, the boundaries of the functional building blocks have been arbitrarily
defined herein for the convenience of the description. Alternative boundaries can
be defined so long as the specified functions and relationships thereof are
appropriately performed. Alternatives (including equivalents, extensions,
variations, deviations, etc., of those described herein) will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein. Such alternatives fall within the scope of the disclosed embodiments. Also, the words “comprising,” “having,” “containing,” and “including,” and other similar forms are intended to be equivalent in meaning and be open ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items, or meant to be limited to only the listed item or items. It must also be

noted that as used herein and in the appended claims, the singular forms “a,” “an,” and “the” include plural references unless the context clearly dictates otherwise.
[057] Furthermore, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present disclosure. A computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor may be stored. Thus, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor(s) to perform steps or stages consistent with the embodiments described herein. The term “computer-readable medium” should be understood to include tangible items and exclude carrier waves and transient signals, i.e., be non-transitory. Examples include random access memory (RAM), read-only memory (ROM), volatile memory, nonvolatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, and any other known physical storage media.
[058] It is intended that the disclosure and examples be considered as exemplary only, with a true scope of disclosed embodiments being indicated by the following claims.

We Claim:
1. A processor implemented method (300) comprising the steps of:
decomposing, via one or more hardware processors, a raw electroencephalogram (EEG) signal having one or more channels into a plurality of Intrinsic Mode Functions (IMFs) using a Multivariate Variational Mode Decomposition (MVMD) method (302);
identifying, via the one or more hardware processors, a dominant frequency for each IMF in the plurality of IMFs associated with one of the one or more channels (304); and
detecting one or more blink artifacts in one or more IMFs
having the dominant frequency in an empirically determined range
of frequencies based on peaks in the one or more IMFs and width of
the peaks, wherein the peaks represent blink peaks and the width of
the peaks represent blink duration respectively (306);
wherein the detected one or more blink artifacts in one or more IMFs
associated with one of the one or more channels is indicative of the presence
of the one or more blink artifacts in the other channels of the one or more
channels.
2. The processor implemented method of claim 1, wherein the one or more channels are obtained from regions proximate to one or more of frontal, temporal and central lobe regions of a subject being monitored.
3. The processor implemented method of claim 1, wherein the step of identifying a dominant frequency is performed by:
performing a Discrete Fourier transform on each IMF in the plurality of IMFs to obtain frequencies up to fs/2 and power associated thereof, wherein fs is a sampling rate of the raw EEG signal (304a); and

identifying a frequency from the obtained frequencies corresponding to a maximum power as the dominant frequency for the associated IMF (304b).
4. The processor implemented method of claim 1, wherein the empirically determined range of frequencies is 2 to 4 Hz.
5. The processor implemented method of claim 1 further comprising removing the detected one or more blink artifacts from the raw EEG signal to obtain one or more blink free channels (308) by:
interpolating one or more non-blink regions preceding and succeeding the detected one or more blink artifacts to replace the detected one or more blink artifacts and obtain one or more interpolated IMFs in the plurality of IMFs associated with the one or more channels (308a); and
reconstructing the one or more channels associated with the detected one or more blink artifacts by adding the one or more interpolated IMFs and remaining IMFs in the plurality of IMFs to obtain the one or more blink free channels (308b).
6. A system (100) comprising:
one or more data storage devices (102) operatively coupled to one or more hardware processors (104) and configured to store instructions configured for execution via the one or more hardware processors to:
decompose a raw electroencephalogram (EEG) signal having one or more channels into a plurality of Intrinsic Mode Functions (IMFs) using a Multivariate Variational Mode Decomposition (MVMD) method;
identify a dominant frequency for each IMF in the plurality of IMFs associated with one of the one or more channels; and

detect one or more blink artifacts in one or more IMFs having
the dominant frequency in an empirically determined range of
frequencies based on peaks in the one or more IMFs and width of
the peaks, wherein the peaks represent blink peaks and the width of
the peaks represent blink duration respectively,
wherein the detected one or more blink artifacts in one or more IMFs
associated with one of the one or more channels is indicative of the presence
of the one or more blink artifacts in the other channels of the one or more
channels.
7. The system of claim 6, wherein the one or more processors are further configured to obtain the one or more channels from regions proximate to one or more of frontal, temporal and central lobe regions of a subject being monitored.
8. The system of claim 6, wherein the one or more processors are further configured to identify the dominant frequency by:
performing a Discrete Fourier transform on each IMF in the plurality of IMFs to obtain frequencies up to fs/2 and power associated thereof, wherein fs is a sampling rate of the raw EEG signal; and
identifying a frequency from the obtained frequencies corresponding to a maximum power as the dominant frequency for the associated IMF.
9. The system of claim 6, wherein the empirically determined range of frequencies is 2 to 4 Hz.
10. The system of claim 6, wherein the one or more processors are further configured to remove the detected one or more blink artifacts from the raw EEG signal to obtain one or more blink free channels by:

interpolating one or more non-blink regions preceding and succeeding the detected one or more blink artifacts to replace the detected one or more blink artifacts and obtain one or more interpolated IMFs in the plurality of IMFs associated with the one or more channels; and
reconstructing the one or more channels associated with the detected one or more blink artifacts by adding the one or more interpolated IMFs and remaining IMFs in the plurality of IMFs to obtain the one or more blink free channels.