Analysis of EEG data Using ICA and Algorithm Development for Energy Comparison

IJSRD - International Journal for Scientific Research & Development| Vol. 1, Issue 3, 2013 | ISSN (online): 2321-0613
All rights reserved by www.ijsrd.com 585
Analysis of EEG data Using ICA and Algorithm Development for Energy
Comparison
Hiral Gandhi1
Kiran Trivedi2
2
Associate Professor
1, 2
Department of Electronics and communication
1, 2
Shantilal Shah Engineering College, Bhavnagar, Gujarat, India
Abstract—This Electroencephalogram (EEG) signal analysis
very useful in clinical research and brain computer interface
application. EEG signal (brain wave) recordings are highly
susceptible from artifacts which are originated from the non-
cerebral origin of the brain. EEG detection and rejection of
artifacts are necessary for acquiring correct information
from EEG signal. Emotiv, Epoc headset can record 16
channels from the scalp of the electrode. EEGLAB allows
analysis of EEG signal through Event related potential
(ERP) analysis, Independent component analysis (ICA), and
time/frequency analysis. Independent component analysis
(ICA) may be suitable method for detecting artifacts. We
analyzed EEG data which are recorded using emotiv epoc in
a different situation for a single person. EEG data are
preprocessed by EEGLAB and decomposes the data by the
ICA. Using statistical method, analyzed the all the dataset
and finding the relationship among the dataset. T- Test
shows that EEG pattern is unique in a person. EEG data is
divided into different frequency band to find the relationship
between the dataset. Also develop the algorithm for
calculating energy of dataset for each channel. Comparing
the energy for each dataset and each channel to find the
maximum and minimum value of energy. In higher
frequency range (13-100 Hz) dataset D (meditation)
contains maximum value of energy for most channels
among all datasets.
Keywords: EEG signal, EEG artifacts, EEGLAB, ICA, Emotiv
epoc headset, Energy
I. INTRODUCTION
EEG data analyses are most currently the subject of
research. Practical applications based on these analyses have
a wide range of applications. It is useful for developing BCI
application as well as in clinical research. The main problem
of analyzing EEG data is to remove the artifact from the
EEG data. Artifacts are high amplitude compare to normal
EEG data. They are present all the time in EEG data. For
correct analysis we have to remove artifact as much as
possible.
The intensities of brain waves recorded from the
surface of the scalp range from 0 to 200 microvolts, and
their frequencies range from few 0 Hz to 200 Hz. Most of
the time, the brain waves are irregular, and there is no
specific pattern can be recognized in the EEG. There are
mainly seven types of Brain waves: Epsilon waves (< 0.4
Hz) Delta waves (0.4-4 Hz), Theta waves (4-8 Hz), Alpha
waves (8-13 Hz), Beta waves (13-30 Hz), Gamma waves
(30-100 Hz), and Lambda waves (100-200 Hz). Delta waves
generate in sleeping adults, premature babies or if there is
any sub cortical lesions and is found in the frontal region of
brain in adults and posterior region in children. Theta waves
generate in children, in adults when they are in emotional
stress or they have deep midline disorders and is found in
parietal and occipital region. Alpha waves generate in quiet
resting state but not sleep and found in the occipital region.
Beta waves generate in active, busy, active concentration or
anxious thinking state and are found in the frontal and
parietal region. Gamma waves generates in certain cognitive
or motor functions. Epsilon and Lambda waves are recently
classified. They are related to the higher level of
consciousness.
Artifacts are somewhat more difficult to remove,
because they are not present all the time, and not in all
electrodes. Artifacts over other noise are the relative high
voltage as compared to the normal EEG signals. However,
since artifacts contaminate the signal very much, they are a
huge problem when using EEG signals. Fortunately, for
same reason many solutions for the have been suggested,
from very simple to mathematically complex. But which
method is suitable for particular EEG data analysis, it is not
specified. Artifacts remove is mostly based on the researcher
visual inspection. Therefore, the good experience in the
analysis of the EEG data is required. EEG signal is recorded
using Emotiv epoc for different situation in a person.
Applying the statistical method apply to find out
relationship among the dataset. Calculating the energy of
each channel for these data and observe the energy variation
in a different situation. To illustrate this, we record the EEG
signal using Emotiv Epoc. These data are preprocessed by
EEGLAB. This paper provides an introduction to the
EEGLAB and various methods for the purposes analysis
EEG data. To extract the information from the EEG data,
applying various ICA algorithms to these data. These dataset
are checked by statistical methods. EEG data is divided into
bands of frequency and they are analyzed independently.
Also calculating the energy of each channel of all dataset.
II. EEG DATA RECORDING AND ANALYSIS OF
DATA
There is continuously electrical activity in the brain. This
activity can be recorded on the scalp of head by putting
electrode.
A. EEG data recording
For recording EEG data, we used wireless emotive epoch
headset. We have research edition v1.5.1.2 (License version)
of emotive epoch wireless headset. Emotiv epoch is wireless
headset to record electrical activity on the scalp of the brain.

Analysis of EEG data Using ICA and Algorithm Development for Energy Comparison
(IJSRD/Vol. 1/Issue 3/2013/0048)
All rights reserved by www.ijsrd.com
586
It has 14 channels plus two channels for reference purposes.
Fig. 1 shows the pictorial view of Emotiv. Epoc headset.
Emotiv epoc consists of 14 channels (AF3, F7, F3, FC5, T7,
P7, O1, O2, P8, T8, FC6, F4, F8, and AF4) for recording the
EEG signal. It is wireless device so the recording of signal
become easier compare to other methods in which placing of
electrodes using wires
Fig. 1: Emotiv epoc headset [9]
B. EEG dataset
EEG signal are recorded in a single person for the situation
given in table 1. Emotiv Epoc can save the data in the
form .EDF (European Data Format) extension. It is standard
file format for exchange and storing EEG (medical data).
EDF file has a header and one or more data records. The
header contains general information like subject
identification, start time etc. and stores multichannel data,
allowing different sample rates for each signal.
C. EEGLAB and ICA
EEGLAB is interactive software for processing continuous
and event-related EEG, MEG and other electrophysiological
data using independent component analysis (ICA),
time/frequency analysis, and other methods including
artifact rejection. It is developed by the Swartz Center for
Computational Neuroscience, Institute for Neural
Computation, University of California SanDiego. EEGLAB
is freely available at reference [8] under the GNU public
license for non-commercial use and open source
development.
ICA is originally proposed for speech signals. Scott
Makeig proposed ICA for EEG source separation.
Therefore, in order to understand ICA, it is essential to
understand independence. At an intuitive level, if two
variables x1 and x2 are independent then the value of one
variable provides absolutely no information about the value
of the other variable. There are several algorithm developed
for ICA algorithms.
1) Infomax
2) Extended infomax
3) Second order blind identification(SOBI)
4) Joint Approximation Diagonalization Eigen Matrices
(JADE)
5) Fast ICA
All the ICA algorithm based on the principal that ICA
decomposes such that component maximally independent.
ICA algorithms start with the random approximation. Let
the EEG data be represented by the vector of time varying
electrode potentials ( )X t , and let the source activities be
( )iS t where i = 1, 2….... n
       1 2 ............ nS t S t S t S t    (1)
Let the scalp maps (patterns of potential) be represented by
vectors, ( )iA t i = 1, 2……… n
The EEG data is the sum of different source signals:
1 1 2 2( ) ( ) ( ) ( ) ( ) ............. ( ) ( )n nX t A t S t A t S t A t S t    (2)
Fig. 2: ICA operation
We have a measured signal vector X and to find the source
signal vector S by approximating value of A vector. Fig. 2
shows the basic operation of ICA.
III. STATISTICAL ANALYSIS AND ENERGY
COMPARISON
In this section we deals with statistical method which is used
for the purposes of comparing EEG dataset and finding
energy of each dataset.
A. T-test
T-test used for comparing responses of the groups. In this
test each group is considered to be a sample from distinct
population. The responses of each group are independent of
those in other group. T-test is available in statistical toolbox
under MATLAB. This test suggests accept or reject null
hypothesis.
We compare the all the dataset using t-test. T-test
applicable for the one sample pair and unpair two samples.
In this analysis we used paired unequal variances for the
sample data. Mathematically we can test data following way
for two sample assuming unequal variances. The t static to
test whether the population means are different or not and is
calculated as
1 2
1 2
X X
X X
t
S 

 (3)
Where; 1 2
2 2
1 2
1 2
X X
S S
S
n n
 
T-test tells that to accept or reject null hypothesis
for the given sets of data.
B. Wavelet analysis
Wavelet analysis represents a windowing technique
with variable-sized regions. Wavelet analysis allows the use
of long time intervals where we want more precise low
frequency information, and shorter regions where we want
high frequency information. In MATLAB wavelet analysis
can be done. Wavelet can be applied for multidimensional
signal. EEG data are multidimensional data, so the wavelet
can directly be applied. We find energy of the each channel
for all datasets using multidimensional data.

587
IV. METHODS AND RESULTS
EEG data is recorded using Emotiv Epoc for different
condition in a single subject. These data are recorded using
Emotiv epoc. Then these data are pre-processed using
EEGLAB. Here we have recorded data for the following
situation.
A 5th floor in neutral condition
B Normal driving condition
C Ground floor
D Meditation
E
Lift is moving from ground floor to seventh floor
(Approximately 85 ft.)
F Lift is moving from seventh floor to ground floor
G ST bus stands with 65 dB noise.
H Noisy environment approximately 75 dB noise
I Driving at 80 km/h
Table. 1 : EEG signal data
We observe all the 16 channels of the datasets. If
the artifact detected by observing the characteristics of
channel, remove that from the channel. These data are
decomposed into the independent component by ICA and
remove the component which contains artifacts. Here,
rejection should not be exceeding 10-15% of the data. If we
remove the more component then the original information
may be lost from the dataset. Fig 3 shows the flow of EEG
signal analysis for energy comparison.
Fig. 3: Flow of EEG signal analysis
After removing the artifact from the dataset, these
datasets are compared by statistical method. Here we used
the T-test to compare the dataset. This test tells that datasets
have different or same. Applying T-test to the pair of dataset
using MATLAB. This test gives the following result
A B C D E F G H I
A 1 0 0 1 1 1 0 0 1
B 0 1 0 0 0 0 0 0 0
C 0 0 1 0 0 0 0 0 0
D 1 0 0 1 1 1 0 0 1
E 1 0 0 1 1 1 0 0 1
F 1 0 0 1 1 1 0 0 1
G 0 0 0 0 0 0 1 0 1
H 0 0 0 0 0 0 0 1 0
I 1 0 0 1 1 1 0 0 1
Table. 2 : T-test among the all datasets
The result shows that EEG pattern is unique in a
single person when the outside noise is less.
For comparing energy of the dataset we develop
the algorithm to find the energy. We have artifact free data.
These Dataset is divided into different frequency bands of
theta waves (0-4 Hz), delta waves (4-8 Hz), alpha waves (8-
13 Hz), beta waves (13-30 Hz) and gamma waves (30-100
Hz).Therefore, each dataset is divided into five groups of
data. We apply these methods to all EEG dataset which are
recorded. Objective to divide these data is that we can
analyse data in each band independently to see the effect of
different condition in EEG data.
All EEG Data are divided in to five frequency
band. Wavelet analysis is applied to all these data. Here
each data contain 14 channels, so the multi-dimensional
wavelet analysis applied to these data under MATLAB.
Then we find energy of each channel for all the dataset
using MATLAB. After calculating energy of each dataset,
find the maximum and minimum value of threshold value of
energy and comparing the threshold value of energy for each
dataset.
Recorded
Data
0-4 HZ 4-8 HZ
Max. Min. Max. Min.
A FC6 P8 FC6 P8
B FC6 P8 P7 P8
C AF3 P8 FC6 P8
D P7 P8 AF4 P8
E T8 P8 FC6 P8
F FC6 P8 FC6 P8
G FC6 P8 FC6 P8
H FC5 P8 F7 P8
I FC5 P8 F8 P8
Table. 3: Channel containing threshold value of energy in
recorded data for 0-4 Hz and 4-8 Hz frequency
Recorded
Data
8-13 HZ 13-30 HZ
Max. Min. Max. Min.
A FC6 P8 T8 P8
B P7 P8 P7 P8
C F8 P8 T7 P8
D AF4 P8 T7 P8
E FC6 P8 FC6 P8
F FC6 P8 FC6 P8
G FC6 P8 FC6 P8
H F7 P8 T8 P8
I F8 P8 P7 P8
Table. 4 : Channel containing threshold value of energy in
recorded data for 8-13 Hz and 13-30 Hz frequency
Table 3 shows the comparing the maximum and
minimum value of energy for all dataset. In this table we
show the channel name which contains maximum and

588
minimum value of energy. Comparison is given for 0-4 Hz
and 4-8 Hz frequency range. We can see that most data set
channel FC5 contain maximum value of energy.
Table 4 shows the comparison for 8-13 Hz and 13-
30 Hz frequency. In A, E, F and G dataset FC6 channel
contain maximum value of energy for 8-13 Hz and 13-30 Hz
frequency. In B, C, D, H, I dataset, channel P7, F8, AF4, F7,
F8 contain maximum value of energy respectively for 13-30
Hz frequency range.
Recorded
Data
30-100 HZ DATA
Max. Min. Max. Min.
A T8 P8 FC6 P8
B T8 P8 P7 P8
C T7 P8 F8 P8
D T8 P8 AF4 P8
E FC6 P8 FC6 P8
F FC6 P8 FC6 P8
G FC6 P8 FC6 P8
H T8 P8 F7 P8
I T7 P8 F8 P8
Table. 5: Channel containing threshold value of energy in
recorded data for 30-100 Hz frequency and all frequency
Table 5 shows comparison of energy for 30-100 Hz
frequency and dataset (contain all frequency).In 30-100 Hz
frequency range channel FC6, T7, T8 contain maximum
energy. For dataset for all frequency most dataset channel
FC6 contains maximum value. In all frequency range P8
contain minimum energy value.
V. CONCLUSIONS
We are recorded the EEG data for a single person in a
different situation. EEGLAB is suitable for analyzing the
EEG signal. We remove some amount of the artifact by
visually inspecting the 14 EEG channel from the dataset.
ICA algorithm decompose EEG signal into the maximal
independent component. Performing t-test among all the
EEG decomposed data. Results of this test shows that EEG
Pattern is unique in a single person. EEG data is divided into
the bands of frequency and analyze the data. By multi
wavelet analysis observing the energy of each channel for
all the datasets. Result shows that in dataset C (ground floor)
and D (meditation) consists maximum energy among all
dataset for frequency range 0-4 Hz.Dataset I (driving at
speed 80 km/h) have maximum value of energy among all
data for frequency range 4-13 Hz range. In higher frequency
range (13-100 Hz) dataset D (meditation) contains
maximum value of energy compare to other EEG data.
REFERENCES
[1]. Arnaud Delorme, Scott Makeig, “EEGLAB: an open
source toolbox for analysis of single-trial EEG
dynamics including independent component analysis”,
Swartz Center for Computational Neuroscience,
Institute for Neural Computation, University of
California San Diego (Journal of Neuroscience Methods
134 (2004), 9-21)
[2]. Grega Repovs “Dealing with Noise in EEG Recording
and Data Analysis” (Informatica Medica Slovenica
2010; 15(1))
[3]. Arnaud Delorme, Terrence Sejnowski1, Scott Makeig
“Enhanced detection of artifacts in EEG data using
higher-order statistics and independent component
analysis”, Computational Neurobiology Laboratory,
Salk Institute for Biological Studies,This article is
reformatted version of the original article published in
NeuroImage 34 (2007) 1443–1449
[4]. ChinTeg Lin, Ruei-Cheng Wu ,” Estimating Driving
Performance Based on EEGSpectrum Analysis”Brain
Research Center, University System of Taiwan, Taipei
112, Taiwan Department of Electrical and Control
Engineering, National Chiao-Tung University, Hsinchu
300, Taiwan(EURASIP Journal on Applied Signal
Processing 2005:19, 3165–3174c 2005 Hindawi
Publishing Corporation)
[5]. Ivan Manuel, Benito Nunez, “EEG Artifact Detection”
Department of CyberneticsCzech Technical University
in Prague
[6]. Ali S.AlMejrd, “Human Emotions Detection using
Brain Wave Signals: A Challenging”, European Journal
of Scientific Research ISSN 1450-216X Vol.44 No.4
(2010), pp.640-659
[7]. V. Matic, W. Deburchgraeve, S. Van Huffel ,
“comparison of ica algorithms for ecg artifact removal
from eeg signals”, University of Belgrade, Department
of Signals and Systems, Serbia Katholieke Universiteit
Leuven, Department of Electrical Engineering ESAT-
SCD, Leuven, Belgium (IEEE-EMBS Benelux Chapter
Symposium, November 9-10, 2009).
[8]. http://www.sccn.ucsd.edu/eeglab/)
[9]. http://www.emotiv.com
[10]. http://www.mathworks.com
[11]. Saeid Sanei and J.A. Chambers “EEG SIGNAL
PROCESSING” John Wiley &. Sons, Ltd Chapter 1-2.
[12]. James V. Stone ‘’INDEPENDENT COMPONENT
ANALYSIS” A Bradford Book the MIT Press
Cambridge, Massachusetts London, England, Chapter
7.

Analysis of EEG data Using ICA and Algorithm Development for Energy Comparison

More Related Content

What's hot (18)

Viewers also liked (8)

Similar to Analysis of EEG data Using ICA and Algorithm Development for Energy Comparison (20)

More from ijsrd.com (20)

Recently uploaded (20)

Analysis of EEG data Using ICA and Algorithm Development for Energy Comparison