2014 IEEE Conference on Biomedical Engineering and Sciences, 8 - 10 December 2014, Miri, Sarawak, Malaysia

Selection of Mother Wavelets Thresholding Methods in Denoising

Multi-channel EEG Signals during Working Memory Task
Noor Kamal Al-Qazzaz, Sawal Ali, Siti Anom Ahmad, Md. Shabiul Islam,
Mohd Izhar Ariff
Abstract The aim of this pilot study was to select the most
similar mother wavelet function and the most efficient
threshold in order to use with wavelet basis function for the
human brain electrical activity during working memory task. A
60 seconds was recorded from the scalp using the
Electroencephalography (EEG). 19 electrodes were placed over
different sites on the scalp where analyzed for one control
subject and one post-stroke patients in the first week of his
stroke onset. In this study, forty-five mother wavelet basis
functions from orthogonal families with four thresholding
methods were used. The selection of mother wavelet functions
like Daubechies (db), symlet (sym) and coiflet (coif) and the
thresholding methods these are sqtwolog, rigrsure, heursure
and minimax are to check mother wavelet functions similarity
with the recorded EEG signals during working memory task.
The test have been done using four evaluating criteria, namely
signal to noise ratio (SNR), peak signal to noise ratio (PSNR)
mean square error (MSE) and crosscorelation method (xcorr).
Symlet mother wavelet of order 9 (sym9) is the most compatible
for all the 19 channels for both EEG datasets that selected to be
examined and the best results have been obtained by using the
rigrsure thresholding method.

I. INTRODUCTION
The electroencephalogram (EEG) records brain electrical
activity in a non-invasive way using electrodes. EEG has
been widely used in clinical diagnosis due to its high
temporal resolution, which is crucial to study the brain
electrical changes associated with post-stroke dementia [1].
Typically, the clinical EEG waveforms have an amplitude
around (10-100) v and the frequency range of (1 to 100) Hz.
Based on its frequency band, the EEG can be classified into
five rhythms. Alpha waves (), are the common in healthy
adults. It occurs in case of wakefulness, relax and eye closed,
and it disappears in pathological cases like coma and sleep. It
has a frequency range of ( 8 to 12 Hz) and its voltage range is
approximately 20-200 v. Beta waves () appear in case of
Noor Kamal Al-Qazzaz and Sawal Ali are with the Department of
Electrical, Electronic and Systems Engineering, Faculty of Engineering and
Built Environment, Universiti Kebangsaan Malaysia; UKM Bangi, Selangor
43600, Malaysia (corresponding author to provide phone: 006-012(email:
6480895;
e-mail:
noorbmemsc81@yahoo.com),
sawal@eng.ukm.my)
Siti Anom Ahmad is with the Department of Electrical and Electronic
Engineering, Faculty of Engineering, Universiti Putra Malaysia;UPM
Serdang, Selangor 43400, Malaysia (email: sanom@upm.edu.my)
Md. Shabiul Islam is with the Institute of Microengineering and
Nanoelectronics (IMEN); Universiti Kebangsaan Malaysia; UKM Bangi,
Selangor 43600, Malaysia (shabiul@ukm.edu.my)
Mohd Izhar Ariff is with Neurology Unit, Department of Medicine,
UKM Medical Centre, Cheras, Kuala Lumpur 56000, Malaysia
(izhar.ariff@ppukm.ukm.edu.my)
Noor Kamal Al-Qazzaz is with the Department of Biomedical
Engineering, Al-Khwarizmi College of Engineering, Baghdad University,
Baghdad 47146, Iraq

978-1-4799-4084-4/14/$31.00 2014 IEEE

extra excitation of the central nervous system with

frequencies in the range of ( 13-30 Hz) and amplitudes in the
range of 5-10 v . The frequencies of the theta waves ()
range from ( 4 to 7 Hz). It is observed in case of sleep,
arousal in older children and adults, emotional stress and
idling with an amplitude range of 5-10 v. Delta waves (),
are the slowest wave with frequency ( less than 3.5 Hz ) and
high amplitude of 20-200 v. It occurs in deep sleep, in
infancy and serious organic brain diseases [2]. Finally, the
Gamma waves () are the fastest wave with a frequency
range of ( 30-100 Hz). It appears in special cases such as
intensive concentration [3]. However, EEG is affected by
non-cerebral sources called artifacts that may mimic the brain
cognitive or pathological activity and therefore influence the
analysis. Most the artifacts that contaminated the EEG signal
are overlapped with the EEG frequencies. They are
physiological artifacts like muscle activity, pulse and eye
blinking or non-physiological artifacts like power line
interference noise and sweat [4, 5]. Numerous methods have
been used to deal with artifacts that affected the EEG
recording. Wavelet is a time-frequency analysis that used to
denoise the non-stationary signals such as EEG [6]. The
selection of mother wavelet function is the most significant
part of wavelet analysis. In this context, this paper address
these challenges with 19 channels ( Fp1, Fp2, F7, F3, Fz, F4,
F8, T3, T5, T4, T6, P3, Pz, P4, C3, Cz, C4, O1, and O2 for
the left and right frontal, temporal, parietal, central and
occipital regions on the scalp).
In the current study, forty-five mother wavelets were
tested to select the more similar mother wavelet, these
include daubechies (db1-db20), symlet (sym1-sym20) and
coiflet (coif1-coif5) and a comparative study of the
performance of four well-known standered thresholding
methods these are sqtwolog, rigrsure, heursure and minimax
thresholding methods were chacked. The similarity of mother
wavelet functions with the recorded EEG data using four
thresholding methods are evaluated using signal to noise ratio
(SNR), peak signal to noise ratio (PSNR), mean square error
(MSE) and crosscorelation method (xcorr). Furthermore, the
selected mother wavelet with thresholding method have been
exploited to denoise the EEG signals which are useful in
revealing the hidden EEG characteristics and enhances the
recorded signal that helps in quantifying the abnormalities in
the spectral of the post-stroke patients EEG.
II. METHODS AND MATERIALS
The EEG dataset was acquired originally from 19 sites on
the scalp using cap electrodes. All the 19 channels from this
dataset were selected in the current work. The flowchart of
our proposed method is shown in Fig.1

214

2014 IEEE Conference on Biomedical Engineering and Sciences, 8 - 10 December 2014, Miri, Sarawak, Malaysia

noise and a band pass filter of (0.5-64 Hz) frequency range

has been used to limit the band of the recorded EEG signals.

Start

A. Wavelet analysis
Wavelet transforms is a popular denoising technique. It
has been introduced to process the non-stationary signals, as
EEG and electromyography (EMG). Zikov et al, Krishnaveni
et al and other researchers have been used wavelet to remove
ocular artifacts [6, 7]. The mathematical equation of the
continuous wavelet transform is shown in 1. The basic
concept of wavelet transform is based on the selection of the
mother wavelet function which is considered as the core
component of the wavelet transform. Mother wavelet
window function (t) is shifted by the location parameter (b)
and dilated or contracted by scaling parameter (a) , as in 2, to
compute the correlation with the signal f(t), the continuous
wavelet transform (CWT) is defined by [8]:

EEG Data Acquisition

during working memory
task

Select Optimal MWT

Select decomposition
Level L=4

,
Select thresholding method

Is Min MSE

No

Yes

(3)

Where a0 and b0 values are set to 2 and 1, respectively.

Based on Mallats algorithm, the DWT decomposed the
signal into different frequency bands this can be described by
the two quadrature mirror filters (QMF), in term of finite
impulse response (FIR) [9], where g is a high pass filter
(HPF) and h is the low pass filter (LPF), the filter h is related
to the scaling function, while filter g is related to the mother
wavelet as given in 4, 5 and 6 [10]:

No

Yes
Is Max PSNR

(2)

The discrete wavelet transform (DWT) can be processed

by obtaining the discrete value of the parameters a and b as in
3.
,

Is Max SNR

(1)

No

Yes
Is Max
xcorr

(4)

(5)

(6)

Yes

The QMF output is characterized as shown in 7 and 8:

Calculate IDWT for the EEG

denoised signal

Stop
Figure 1. The block diagram of the proposed method.
The EEG data were obtained during working memory
task in order to investigate the association and the similarity
between brain waves during mental activation through EEG
measures and the mother wavelet basis functions using
Matlab program. Conventional filtering are used as a first
stage to process the 19 channels EEG data. Notch filter at (50
Hz) has been used to remove the power line interference

2
2

(7)
(8)

When the signal x(n) act as LPF it convolved with h(n2L), otherwise it acts as HPF and convolved with (n-2L).
The result is transforming the original signal into two sub
bands. It is significant that the HL are the approximation
components (A) and it represents the lowest resolution
components, and GL is the details decomposition components
(D) that describes the high resolution components.
Discrete wavelet transform (DWT) is used for denoising
purposes, the first step is selecting the optimal mother
wavelet function based on its compatibility with the noisy
EEG signal, the efficient orthogonal wavelet family like
daubechies, symlet and coiflets have been used because the

215

2014 IEEE Conference on Biomedical Engineering and Sciences, 8 - 10 December 2014, Miri, Sarawak, Malaysia

EEG signals have high compatibility with these families to

introduce the best result based on segments decomposition
into non-overlapping sub-band frequency [11, 12]. The
second step is selecting the suitable mother wavelet and
threshold to be suitable for the denoising procedure.
B. Mother wavelet optimal selection
In order to find the suitable mother wavelet function for
EEG denoising by wavelets among forty-five mother wavelet
from different families including daubechies (db1-db20),
symlet (sym1-sym20) and coiflet (coif1-coif5). It is necessary
to select the mother wavelet function. Daubechies, symlet
and coiflets were selected to test their similarity to the EEG
signals based their best result in most cases of EEG signals
[13, 14]. The selection of mother wavelet has performed by
segmenting each one of the recorded 19 channels length of
(15360 samples) into (60 segments) length of each segment
was (256 samples). Forty-five mother wavelets were applied
to check their similarity to the recorded EEG datasets.

the Signal to Noise Ratio (SNR), Mean Square Error (MSE)

and Peak Signal to Noise Ratio (PSNR) were used. Assuming
that the EEG original signal x(n) and the denoised signal is
x(n), and N is the length of the selected window SNR,
PSNR, MSE and xcorr can be evaluated using 12- 15
respectively [13, 16-18]:

10

20

(12)
(13)

(14)
(15)

C. Threshold selection method

To select a properate threshold, a comparative analysis of
four wavelet thresholding methods have been performed
using Square-Root-Log (sqtwolog), (minimaxi) using
minimax principle, (Heursure) and (rigrsure) using the
principle of Steins Unbiased Risk Estimate (SURE). These
methods are briefly described as follows [15]:

Here RMSE is the root-mean-square error, and x and y

are two random variables with expected values of x and y
and standered deviation of x and y.

E. Subjects and EEG recording procedure

For the present pilot study, two EEG datasets were
examined to check the compatibility of the mother wavelet
functions with the EEG signals that obtained from two
different health state of subjects. The first EEG dataset was
for a control subject (right-handed male, 57 years old)
whereas the second EEG dataset was for a (right-handed
male, 59 years old) post-stroke patient in the first week after
stroke onset. Control subjects were matched with the stroke
patients regarding the age and level of education. The poststroke patient satisfied the National Institutes of Health
Stroke Scale (NIHSS) [19] to evaluate the acuity of his
stroke. The patient was recruited from the stroke ward in
Pusat Perubatan Universiti Kebangsaan Malaysia
(PPUKM), Malaysia. CT/MRI scans, medical history,
clinical, neurological, psychological examinations (including
Mini-Mental State Examination (MMSE), Montreal
Cognitive Assessment (MoCA) and laboratory studies were
performed to diagnose the post-stroke patient [20]. All the
study experiment protocols had been approved by the Human
Ethics Committee of the National University of Malaysia.
Information consent form was signed by the patient. EEG
activities were recorded from 19 sites using electrodes set of
NicoletOne systems (V32) designed and manufactured by
VIASYS Healthcare Inc., USA. 19 electrodes plus ground
and system reference electrodes positioned according to the
10-20 international system ( Fp1, Fp2, F7, F3, Fz, F4, F8, T3,
T5, T4, T6, P3, Pz, P4, C3, Cz, C4, O1, and O2 which
illustrated in Fig. 2. The EEG was recorded for 60 seconds
during working memory task as shown in Fig 3. Patients
were asked to memorise five words for 10 seconds [21]. Then
he was asked to remember the five words while the EEG was
recorded with the eyes closed. After 1 min he was asked to
open the eyes and enumerate the five words he could
remember. NicoLet EEG system is sampled at 256 Hz

Sqtwolog criterion: it can be calculated using universal

thresholding method using the median absolute deviation
() ) and the length Nj of the noisy signal at the jth scale
as in 9 and 10:
| |
.

(9)
(10)

Where represent the wavelet coefficient at scale j.

Rigrsure criterion: soft threshold method evaluator

of unibiased risk. It can be calculated using equation 11:
(11)

Where b is the bth squared wavelet coefficient at minimum

risk chosen from the vector W=[1, 2, .., N]
contains the square wavelet coefficient values that
arranged from small to large and is the standered
deviasion of the noisy signal.

Heursure criterion: is a combination of both Sqtwolog

and Rigrsure methods. In case of signal to noise ratio is
very small the SURE methods estimation is poor
whereas the Sqtwolog methods gives better threshold
estimation.

Minimax criterion: it based on the minimax principle

that used in statistics. A fixed threshold is used to yield
the minimax performance for mean square error against
an ideal procedure.

D. Evaluation criteria of artifacts removal

In order to evaluate the performance of the selected
mother wavelet functions using four thresholding methods,

As an indicator, the higher (SNR, PSNR and xcorr) and

the smaller MSE predicts better compatibility and provides
higher similarity between the recorded EEG datasets and the
mother wavelet functions.

216

2014 IEEE Conference on Biomedical Engineering and Sciences, 8 - 10 December 2014, Miri, Sarawak, Malaysia

sampling frequency, impedance of electrodde/skin was below

10 kilo ohms, sensitivity of 100 v/cm, low cut of 0.5 Hz and
high cut of 70 Hz using referential montage.

0.5

-0.5

-1

-1.5

10

12

14

16

18

Figure 4. Symlet mother wavelett function of order 9.

TABLE 1: EVALUATION CRITERIA TO SELECT
T MOTHER WAVELET FUNCTION
FOR FP1 CHANNEL OF THE CON
NTROL SUBJECT.

Figure 2. The electrodes placement sites baased on the 10-20

system.

Thresholding
Methods

MWT

SNR
R

MSE

PSNR

db4

08
-6.0

9.86

38.19

Sqtwolog

sym9

-5.6
61

8.86

38.65

coif2

-6.0
03

9.76

38.23

db4

48
-3.4

5.42

40.79

sym9

-2.3
37

4.20

41.90

coif2

-3.4
40

5.33

40.86

db4

65
-5.6

8.93

38.62

sym9

-4.3
38

6.67

39.89

coif3

-5.6
60

8.84

38.66

db4

09
-5.0

7.85

39.18

sym9

-4.2
26

6.48

40.01

coif2

-4.9
93

7.57

39.34

RigrSure

HeurSure

Minimaxi

TABLE 2: CROSS-CORRELATION CRITERIA TO

T SELECT MOTHER WAVELET
FUNCTION FOR FP1 CHANNEL OF TH
HE CONTROL SUBJECT.

Figure 3. The original signals during workiing memory task

for stroke patient.

Thresholding
Methods
Sqtwolog

ON
III. RESULTS AND DISCUSSIO

In this study, the results in Tables 1 shhows that among

forty-five mother wavelet functions, symlet of order 9 (sym9)
provides the best fit to the tested 19 channells of the two EEG
datasets. This might due to the shape of sym
m function and the
denoised signal as shown in Fig. 4. It providded pronouncedly
the highest (SNR, PSNR, and xcorr) and tthe lowest (MSE)
when applied with Rigrsure thresholdingg method, which
reveal the best similariy properties based onn these evaluation
criteria for both control subject and poost-stroke patient
respectively as shown in Fig. 5. Fig. 6 shoows the similarity
between the recorded EEG signal during w
working memory
task and the sym9 mother wavelet functiion for the poststroke patient.

217

MWT

Xcorr

db14

0.89

sym9

0.91

RigrSure

coif4
db2

0.86
0.96

HeurSure

sym9
coif2
db16

0.95
0.97
0.93

sym9

0.93

coif2
db16

0.89
0.93

sym9

0.94

coif2

0.92

Minimaxi

2014 IEEE Conference on Biomedical Engineering and Sciences, 8 - 10 December 2014, Miri, Sarawak, Malaysia

Rigrsure was used as the main thresholding method for

this comparative study. It is a soft thresholding and it perform
well with minimum mean square error comparing with other
thresholding methods as shown in Fig 5.
10

[1]

[2]

8
MSE

REFERENCES

Sqtwolog

RigrSure

HeurSure

Minimaxi

[3]

[4]

Sqtwolog RigrSure HeurSure Minimaxi

Thresholding Methods

Figure 5. Comparative plot of the four thresholding

methods.

[5]
[6]

F p1

original siganl
filtered signal

-100
-200
200

400

600

800

1000

1200

200

400

600

800

1000

1200

200

400

600

800

1000

1200

200

400

600

800

1000

1200

200

400

600

800

1000

1200

Fp2

0
-100
-200

T5

40
20
0
-20

[7]

T6

40
20
0
-20

O1

40
20
0
-20
40
20
0
-20

O2

[8]
200

400

600
sample number

800

1000

1200

Figure 6. The similarity between the recorded EEG

dataset of stroke patient during working memory task and
sym9.

[9]
[10]

VI. CONCLUSION
In the current study an optimal wavelet function has been
selected to check the compatibility of the EEG datasets of
both control subject and the post-stroke patient with mother
wavelet function using four thresholding methods. 19
channels from different regions on the scalp were tested with
forty-five mother wavelet functions. Symlet mother wavelet
of order 9 is the most similar function for both EEG datasets
which was recording during a working memory task using
Rigrsure thresholding method. The selection is based on the
maximum (SNR and PSNR) and minimum MSE. The current
study provides useful information on the optimal selection
and expectation for mother wavelet functions and
thresholding methods that may help in the increase prediction
accuracy of dementia that may follow a stroke diagnostic.

[11]

[12]

[13]

218

N. K. Al-Qazzaz, S. H. B. M. Ali, S. A. Ahmad,

K. Chellappan, M. S. Islam, and J. Escudero,
"Role of EEG as Biomarker in the Early
Detection and Classification of Dementia,"2014.
W. J. G., Medical Instrumentation Application
and Design. New York: Wiley, 1998.
M. I. Al-Kadi, M. B. I. Reaz, and M. A. M. Ali,
"Evolution of Electroencephalogram Signal
Analysis Techniques during Anesthesia,"
Sensors, vol. 13, pp. 6605-6635, 2013.
T.-P. Jung, S. Makeig, M. Westerfield, J.
Townsend, E. Courchesne, and T. J. Sejnowski,
"Removal of eye activity artifacts from visual
event-related potentials in normal and clinical
subjects," Clinical Neurophysiology, vol. 111,
pp. 1745-1758, 2000.
I. M. B. Nez, "EEG artifact dtection," 2010.
T. Zikov, S. Bibian, G. A. Dumont, M.
Huzmezan, and C. Ries, "A wavelet based denoising technique for ocular artifact correction
of the electroencephalogram," in Engineering in
Medicine and Biology, 2002. 24th Annual
Conference and the Annual Fall Meeting of the
Biomedical Engineering Society EMBS/BMES
Conference, 2002. Proceedings of the Second
Joint, 2002, pp. 98-105.
V. Krishnaveni, S. Jayaraman, S. Aravind, V.
Hariharasudhan, and K. Ramadoss, "Automatic
identification and Removal of ocular artifacts
from EEG using Wavelet transform,"
Measurement Science Review, vol. 6, pp. 45-57,
2006.
A. Shoeb and G. Cliord, "Chapter 16 - Wavelets;
Multiscale Activity in Physiological Signals," in
Biomedical Signal and Image Processing, ed,
2005.
M. a Unser, Wavelets in medicine and biology:
CRC press, 1996.
S. G. Mallat, "A theory for multiresolution
signal
decomposition:
the
wavelet
representation," Pattern Analysis and Machine
Intelligence, IEEE Transactions on, vol. 11, pp.
674-693, 1989.
C.-J. Chang, "Time frequency analysis and
wavelet transform tutorial," Time frequency
analysis for biomedical engineering, National
Taiwan University.
L. Yu, "EEG De-noising Based on Wavelet
Transformation," presented at the ICBBE
International Conference, Beijing, 2009.
J. Rafiee, M. Rafiee, N. Prause, and M. Schoen,
"Wavelet basis functions in biomedical signal
processing," Expert Systems with Applications,
vol. 38, pp. 6190-6201, 2011.

2014 IEEE Conference on Biomedical Engineering and Sciences, 8 - 10 December 2014, Miri, Sarawak, Malaysia

[14]

[15]

[16]

[17]

[18]

[19]

[20]

[21]

M. I. Al-kadi, M. Reaz, and M. Ali,

"Compatibility of mother wavelet functions
with the electroencephalographic signal," in
Biomedical
Engineering
and
Sciences
(IECBES), 2012 IEEE EMBS Conference on,
2012, pp. 113-117.
N. Verma and A. Verma, "Performance
analysis of wavelet thresholding methods in
denoising of audio signals of some Indian
Musical Instruments," International Journal of
Engineering Science and Technology, vol. 4,
pp. 2040-2045, 2012.
L. Yu, "EEG De-Noising Based on Wavelet
Transformation," in Bioinformatics and
Biomedical Engineering, 2009. ICBBE 2009.
3rd International Conference on, 2009, pp. 1-4.
J. Walters-Williams and Y. Li, "BMICAindependent component analysis based on Bspline mutual information estimator," Signal
and Image Processing, vol. 3, pp. 33-52, 2012.
N. Sriraam and C. Eswaran, "Performance
evaluation of neural network and linear
predictors for near-lossless compression of
EEG signals," Information Technology in
Biomedicine, IEEE Transactions on, vol. 12,
pp. 87-93, 2008.
T. Brott, H. Adams, C. P. Olinger, J. R. Marler,
W. G. Barsan, J. Biller, et al., "Measurements
of acute cerebral infarction: a clinical
examination scale," Stroke, vol. 20, pp. 864870, 1989.
N. K. Al-Qazzaz, S. H. Ali, S. A. Ahmad, and
S. Islam, "Cognitive assessments for the early
diagnosis
of dementia
after
stroke,"
Neuropsychiatric disease and treatment, vol.
10, p. 1743, 2014.
N. K. Al-Qazzaz, S. H. Ali, S. A. Ahmad, S.
Islam, and K. Mohamad, "Cognitive
impairment and memory dysfunction after a
stroke diagnosis: a post-stroke memory
assessment," Neuropsychiatric disease and
treatment, vol. 10, p. 1677, 2014.

219