Int J Pharm Bio Sci 2016 July; 7(3) : (B) 1149 - 1153

Original Research Article

Biomedical

International Journal of Pharma and Bio Sciences

ISSN
0975-6299

ASSESSMENT OF COGNITIVE STATUS FROM EEG DATA USING

INDEPENDENT COMPONENT ANALYSIS
M. SUNDAR RAJ1, K. ADALARASU2,*, M. JAGANNATH3 AND R. BALA SARANYA2
1

Department of Biomedical Engineering, Bharath University, Chennai, Tamilnadu, India

Department of ECE, PSNA College of Engineering & Technology, Dindigul, Tamilnadu, India
3
School of Electronics Engineering, VIT University, Chennai, Tamilnadu, India
2

ABSTRACT
Biomedical signal processing aims at extracting significant information from physiological signals
recorded from the human body. The various biomedical signals such as electroencephalography (EEG),
electrocardiography (ECG), magnetoencephalography (MEG) etc. can be analyzed by using both
independent component analysis (ICA) and principal component analysis. The main objective of this
study is to assess the cognitive status of the individual using independent component analysis (ICA). The
methodology of this study is to compare the different ICA algorithms and identify better algorithm for EEG
signal processing; extract the feature parameter from the EEG signal and identify the cognitive status of
subjects using a feature extracted parameter. Thus, finally the extracted parameter such as alpha, theta
and beta band values is used to analyze the cognitive status of the individual. Our study concluded that
wavelet packet decomposition of EEG showing increased beta activity and decrease in alpha activity with
increase in theta activity during task performance is a good approach to assess mental fatigue and
alertness level.
KEYWORDS

: Cognitive status, Electroencephalography (EEG), Stress, Independent component analysis (ICA).

Dr. K. ADALARASU
Department of ECE, PSNA College of Engineering & Technology, Dindigul,
Tamilnadu, India

*Corresponding Author

This article can be downloaded from www.ijpbs.net

B - 1149

Int J Pharm Bio Sci 2016 July; 7(3) : (B) 1149 - 1153

INTRODUCTION
NO human being is exempted from stress. Stress
causes a number of biological changes and is intended
to activate the body's fuel reserves. The stress in the
army personnel can be attributed to aspects like risk to
life, threat of an enemy, hardships faced, feeling of
loneliness, family worries, and loss of sleep and rest,
lack of opportunity for change of routine, etc. Military
psychology studies human behavior in the combat
situation and attempts to solve the problems faced by
the defense personnel. Active cognition during complex
and sustained operations is a critical component for
1
success in current and future military operations . The
brain is responsible for processing information, making
decisions and initiating actions on the external
1,2
environment . Cognitive effort involves sustained
vigilance, selective attention, complex decision making
and perceptual-motor control skills. Real-time monitoring
of individuals for cognitive performance capacity via an
approach based on sampling multiple neurophysiologic
signals and integrating those signals with performance
prediction models potentially provides a method of
supporting warfighters' and commanders' decision

making and other operationally relevant mental

processes and is consistent with the goals of augmented
cognition. Measures of the electrical activity of the brain
are seen by many to be direct ways of determining the
cognitive demands placed upon an operator. Basic
research studies in human cognitive neurophysiology
currently include the use of brain imaging techniques
such as positron emission tomography (PET), functional
magnetic
resonance
imaging
(fMRI),
mass
spectroscopy, near-infrared spectroscopy (NIRS),
magneto-encephalography
(MEG),
and
electroencephalography to measure brain activity
changes at specific cortical (by surface electrodes) and
subcortical sites (by implanted electrodes) during
perceptual and cognitive task performance. EEG
provides good observational data of variability in mental
status because of its high temporal resolution. EEG
waveform (amplitude and frequency) depends on the
conscious level of the person. Short episodes (3-15 sec)
of theta and decreased alpha activity in the EEG are
often used as indicators of micro-sleeps. Eyes-closed
resting EEG shifts from alpha to theta activity during
transition to sleep.

Figure 1
Rhythms of human brain
It has often been hypothesized that during changes in
alertness or onset of fatigue, the EEG spectral signal
would include an increase of low frequency EEG waves,
such as delta (0.5-4 Hz), theta (4-8 Hz) and alpha (8-13
Hz) while higher frequency waves such as beta rhythms
(14-32 Hz) will decrease (Fig. 1). Increased EEG alpha
wave activity has often been associated with an
increased level of relaxation and hence its decrease
occurs during stress. Theta increases with workload and
is associated with numerous processes, including
working memory, problem solving, encoding, or self2,9
monitoring . Brain is one of the largest and most
complex organs in the human body. There are five major
brain waves, distinguished by their different frequency
ranges. Frequency bands from low to high frequencies
respectively Delta (), Theta (), Alpha (), Beta () and
Gamma (). Fig. 1 denotes the various rhythms of the
human brain.

LITERATURE REVIEW

parametric nonlinearity to any kind of density of sources.

Constrained ICA (cICA) algorithm can be one of the best
tools for EM brain signal analysis, with an initial
application in automating artifact extraction in EEG and
4,5
6
MEG . Van Dun et al. study concludes that ICA is able
to detect auditory steady-state responses (ASSR) for
reliable hearing threshold estimation at audiometric
frequencies. In this they achieved significant time
reduction for both single channel and multi-channel
signal recordings. The temporal de-correlation method
provides better performance for analyzing physiological
7
signal analysis such as sEMG, EEG etc. Acharya et al.
concluded
that
constrained
genetic
algorithm
optimization based on ICA overcomes the long standing
permutation ambiguity and recovers the independent
8
components in a fixed order. Wei-Chung Huang et al.
used ICA technique with information maximization
(Infomax) algorithm in on-line EEG signal separation.
They concluded that this system provides faster design
and very good correlation.

Jeong-Won Jeong et al. showed that mixture density

ICA provides better separation by matching flexible
This article can be downloaded from www.ijpbs.net
B - 1150

Int J Pharm Bio Sci 2016 July; 7(3) : (B) 1149 - 1153

EXPERIMENTAL PROTOCOL
EEG refers to the recording of the brains spontaneous
electrical activity over a short period of time, usually 20
40 minutes, as recorded from multiple electrodes placed
on the scalp. The raw EEG data is given as the input.
The EEG signal considered here is collected by means
of 19 electrodes for 28 minutes from 14 participants.
Thus the collected raw data is preprocessed before
applying it to an ICA algorithm. Fig. 2 shows the work
flow of EEG signal processing. In order to reduce the
noise in the signal the filtering process is done by two
cases: Band pass (0.4-30 Hz) and Band stop (49-51
Hz).

A. Independent Component Analysis (ICA)

Independent component analysis is one of the most
commonly used algorithms in blind source separation. It
solves the problem of finding unknown, unobserved or
hidden structure in high dimensional data. Independent
component analysis (ICA) is a technique of data
transformation that finds independent sources in
recorded mixtures of sources. It does not require any
information on incoming signals. Since it utilizes only the
statistical independence of the incoming signals, this
separation problem is known as blind signal separation.
Such techniques have been applied to many fields,
including biology, biomedical signal processing, digital
communication.

.
Figure 2
Work flow of EEG signal processing
1) FAST ICA
The algorithm is based on a fixed-point iteration scheme
maximizing non-Gaussianity as a measure of statistical
independence. It can also be derived as an
approximative Newton iteration. To estimate n
independent components, run these algorithm n times.
To ensure that we estimate each time a different
independent component, we only need to add a simple
orthogonalizing projection inside the loop. The column of
the demixing matrix B is orthonormal because of the
sphering. Thus we can estimate the independent
components one by one by projecting the current
solution w(k) using equation (1) on the space orthogonal
to the columns of the demixing matrix B previously
found. Define the matrix B as the matrix whose columns
are the previously found columns of B.
(1)
The FICA algorithm and the underlying contrast
functions have a number of desirable properties when
compared with existing methods for ICA. The following

are the advantages of this algorithm: very fast

convergence; easy to use; increases performance and
little memory space.
2) Efficient FAST ICA
Efcient version of Fast ICA is based on the following
observations: The symmetric FICA algorithm can be run
with different nonlinearity for different sources. The
algorithm remains to be asymptotically efcient if the
theoretically optimum auxiliary constants in the algorithm
are replaced by their consistent estimates.
3) Comparison between FICA and EFICA
There are many ICA algorithms for separation, so to
predict the better algorithm the convergence time of the
signals based on ICA algorithm is calculated. Table 1
shows the comparison of FICA and EFICA algorithms. In
which the overall analysis proved that FICA has better
convergence time and this algorithm is selected for the
separation process.

Table 1
Comparison of convergence time between FICA and EFICA algorithms
Algorithms
FICA
EFICA

B. Wavelet Decomposition And Feature Extraction

The object of wavelet analysis is to decompose signals
into several frequency bands. In this work the number of
decomposition levels was chosen to be 6. The number
of decomposition levels is chosen based on the
dominant frequency components of the signal. The
levels are chosen such that those parts of the signal that
correlate well with the frequencies necessary for
classification of the signal are retained in the wavelet
coefficients. In this study, the number of decomposition

Convergence Time (sec)

27.03
103.4

levels was chosen as 6 since the EEGs have little useful

information above frequency 30 Hz. Daubechies wavelet
of order 4 (db4) was selected because its smoothing
feature was suitable for detecting changes of the EEG
signals. Frequency bands corresponding to six
decomposition levels corresponding to EEG band for the
sampling frequency of 500 Hz. In order to extract the
feature parameters the duration of 10 sec in each signal
is extracted based on the time interval of 2 minutes. For
the extracted signal the relative energy can be

This article can be downloaded from www.ijpbs.net

B - 1151

Int J Pharm Bio Sci 2016 July; 7(3) : (B) 1149 - 1153
calculated. Thus, the relative wavelet energy (RWE) is
defined in equation (2).
(2)
where j is the level of decomposition.

RESULTS
The raw EEG data is given as the input. In order to
reduce the noise in the signal the filtering process is
done by two cases. The filtered output is a mixed signal
data in order to separate the mixed signal the

independent component analysis algorithm is used for

separation.
There are many ICA algorithms for
separation so to predict the better algorithm the
convergence time of the signals based on ICA algorithm
is calculated. Based on overall analysis FASTICA has
better convergence time and this algorithm is selected
for the separation process. In order to analyze the
cognitive status in the total of 19 channel only 7 (Fp1,
Fp2, Fz, Cz, P3, P4, Pz) channels are selected. The
relative wavelet energy (RWE) can be calculated using
equation (2) for different frequency bands.

Table 2
RWE slope values of alpha band
Electrode
Placement
Fp1
Fp2
Fz
Cz
P3
Pz
P4

Participant
Participant 1
Participant 2
0.1730
0.3206
0.3330
0.1650
0.0761
0.1010
0.0428
0.1710
0.3444
0.3183
0.1602
0.1729
0.2661
0.2630

The regression plot denotes the accuracy of matching

between the time and the alpha values of signals
obtained. From the regression plot the slope value is
calculated to determine whether the signal has either
positive inclination or negative. From the calculated

value the cognitive status of each subject can be

analyzed. Table 2, 3 and 4 shows the slope value of
alpha, theta and beta band value of two participants and
Table 5 shows the average slope value of all three
bands.

Table 3

RWE slope values of theta band

Electrode
Placement
Fp1
Fp2
Fz
Cz
P3
Pz
P4

Participant
Participant 1
Participant 2
0.1817
-0.4369
-0.3949
0.1932
-0.0281
-0.0844
-0.0730
-0.1371
0.0416
0.0343
0.0971
0.0883
0.1466
0.1437

Table 4
RWE slope values of beta band
Electrode
Placement
Fp1
Fp2
Fz
Cz
P3
Pz
P4

Participants
Participant 1
Participant 2
-0.3001
0.0964
0.0413
-0.3225
-0.0694
-0.0303
0.0269
-0.0490
-0.3739
-0.3696
-0.2065
-0.2306
-0.3903
-0.3811

Table 5
Average slope values of three bands
Electrode
Placement
Fp1
Fp2
Fz
Cz
P3
Pz
P4

Alpha
-0.3342
0.9415
0.6927
0.2515
0.0061
0.2009
1.3726

Average Slope Value

Theta
1.291
0.837
0.5477
0.0925
1.8156
0.7879
1.3706

Beta
-0.8236
-1.5532
-1.2067
-0.4067
-1.7104
-0.9563
-2.574

The cognitive status of each participant can be analyzed based

on the slope values calculated. Table 6 denotes the cognitive
status of each participant.

This article can be downloaded from www.ijpbs.net

B - 1152

Int J Pharm Bio Sci 2016 July; 7(3) : (B) 1149 - 1153
Table 6
Cognitive status of the participants
Participant
Participant 1
Participant 2
Participant 3
Participant 4
Participant 5
Participant 6
Participant 7
Participant 8
Participant 9
Participant 10
Participant 11
Participant 12
Participant 13
Participant 14

Drowsiness
Drowsiness
Alert
Drowsiness
Alert
Alert
Drowsiness
Drowsiness
Very Alert
Drowsiness
Alert
Drowsiness
Drowsiness
Drowsiness

CONCLUSION
Component separation algorithms can be used in health
care, such as diagnosis of diseases, extracting important
features of physiological signals such as EEG, ECG,
EMG etc, and medical imaging. In the present study, we
have explained how the ICA algorithm is used in the
separation of EEG signal, and based on their
convergence time better algorithm is predicted. In this
various features such as alpha, beta and theta bands
are extracted by means of wavelet decomposition and
relative wavelet energy. The extracted feature value is
analyzed by means of regression plot. This provides the
slope values of each signal for various bands. This value
is compared to analyze the cognitive status of the

Cognitive Status
Mental Fatigue
Mental Fatigue
No Fatigue
No Fatigue
No Fatigue
Mental Fatigue
No Fatigue
Mental Fatigue
Not Fatigue
Mental Fatigue
No Fatigue
No Fatigue
Mental Fatigue
No Fatigue

No Stress
Stress
No Stress
Stress
Stress
No Stress
Stress
No Stress
No Stress
No Stress
Stress
Stress
No Stress
Stress

various subjects. Thus, we conclude that this work

provides the better way for separation and analyzing the
cognitive status of the humans.

ACKNOWLEDGEMENT
The authors would like to acknowledge all the
participants involved in the study. We also wish to thank
the reviewers for their valuable comments to enhance
the quality of the manuscript.

CONFLICT OF INTEREST
Conflict of interest declared none.

REFERENCES
1.

2.

3.

4.

5.

Hedman E, Andersson E, Lekander M, Ljtsson

B. Predictors in internet-delivered cognitive
behavior
therapy
and
behavioral
stress
management for severe health anxiety. Behaviour
Research and Therapy, 2015, Jan; 64: 49-55.
Konopka LM, Zimmerman EM. Neurofeedback
and psychopharmacology: designing effective
treatment based on cognitive and EEG effects of
medications. Clinical Neurotherapy, 2014: 55-84.
Jeong JW, Kim TS, Singh M. Independent
component analysis with mixture density model
and its application to localize the brain alpha
activity in fMRI and EEG. Nuclear Science
Symposium Conference Record, IEEE, 2001; 4:
1897-1901.
James CJ, Gibson OJ. Temporally constrained
ICA: an application to artifact rejection in
electromagnetic brain signal analysis. IEEE
Transaction on Biomedical Engineering, 2003;
50(9): 1108-1116.
Parmar SD, Sahami JS. A comparative survey on
removal of MECG artifacts from FECC using ICA

6.

7.

8.

9.

algorithms. Intelligent Sensing and Information

Processing, 2004, p. 88-91.
Van Dun B, Wouters J, Moonen M. 2007.
Improving
auditory
steady-state
response
detection using independent component analysis
on multichannel EEG data. IEEE Transactions on
Biomedical Engineering, 2007, July; 54(7): 12201230.
Acharya DP, Panda G, Lakshmi YVS.
Constrained genetic algorithm based independent
component analysis. IEEE Congress on
Evolutionary Computation, 2007, p. 2443-2449.
Wei-Chung Huang, Shao-Hang Hung, Jen-Feng
Chung, Meng-Hsiu Chang, Lan-Da Van, ChinTeng Lin. FPGA implementation of 4-channel ICA
for on-line EEG signal separation. IEEE
Biomedical Circuits and Systems Conferences,
2008, November 20-22, p. 65-68.
Ahmed MI, Subramanian DV. A review of brain
machine interface techniques, tools and
applications. International Journal of Pharm Bio
Sciences, 2015, July; 6(3): 179-185.

This article can be downloaded from www.ijpbs.net

B - 1153