SOBI-RO for Automatic Removal of Electroocular Artifacts ... · Figure 1:BCI block diagram (From...
Transcript of SOBI-RO for Automatic Removal of Electroocular Artifacts ... · Figure 1:BCI block diagram (From...
Proceeding of the International Conference on Artificial Intelligence and Computer Science (AICS 2014), 15 - 16 September 2014, Bandung, INDONESIA. (e-ISBN 978-967-11768-8-7). Organized by http://WorldConferences.net 180
SOBI-RO for Automatic Removal of Electroocular Artifacts from EEG Data-Based
Motor Imagery
Arjon Turnip and Fajar Budi Utomo
Technical Implementation Unit for Instrumentation Development,
Indonesian Institute of Sciences, Bandung, Indonesia
Emails: [email protected], [email protected]
Abstract—Signals from eye movements and blinks can be orders of magnitude larger than brain-
generated electrical potentials and are one of the main sources of artifacts in electroencephalographic
(EEG) data. This article presents a method based on blind source separation (BSS) for automatic
removal of electroocular artifacts from EEG datain amotor imagery experiment. BBS is a signal-
processing methodology that includes independent component analysis (ICA)using second order blind
identification with robust orthogonalization (SOBI-RO) is proposed.Simulation results shows that the
ocular artifacts are significantly removed and the sources of the brain activity are clearly identified.
The identification performance using signal to distortion ratio value about 68.88% is achieved.
Keywords—EEG signal, Ocular Artifact, SOBI-RO.
I. INTRODUCTION
The Brain-Computer Interface (BCI) provides an additional output channel from brain, and uses the
neuronal activity of brain to control effectors such as robotic arm or wheel chair; or to restore motor
abilities of paralyzed or stroke patients [1-4]. The core components of a BCI system [1-3] are brain
signal acquisition, pre-processing, feature extraction, classification, translation and feedback control of
external devices. Based on the type of sensors used for the data acquisition, BCI systems can be
invasive or non-invasive. The BCI scheme is shown on the Fig. 1.As Fig. 1 shows, BCIs can be seen
as a pattern recognition system [1]. Its aim is to translate brain activities into commands for a devices
control. In order to achieve this goal, firstly signals from the brain are acquired by electrodes mounted
on the scalp or in the head and subsequently the specific features of these signals will be extracted
(e.g, amplitudes of evoked potentials, band powers or power spectral density values). Then these
features are classified and translated into commands to control a device. In this paper, we focus on one
kind of neurophysiological signals, namely electroencephalogram (EEG) signals that are electrical
brain activities recorded from electrodes placed on the scalp.
EEG is a widely used non-invasive BCI due to its low expense and high temporal resolution. The
EEG data acquisition is followed by a pre-processing stage which attenuates the artifacts and noises
present in the brain signal, to enhance the relevant information.The EEG signals contain not only
desired signal from brain electrical activity but also undesired electrical brain activity. The undesired
signals come from recorded signals that are non-cerebral in origin (they are called artifacts).Ocular
artifacts occur when the subject blinks the eye and creates significant electrical potential during EEG
recording. They are featured by high amplitude, but the high amplitude peaks are mainly seen on the
frontopolar channels in the combination with the occipital channels. These peaks areconsidered as one
of the most considerable artifacts in EEG studies [7-9]. Due to the presence of ocular artifacts, it is
difficult to analyze EEG signal because of their spikes. The undesirable signals must be eliminated or
attenuated from the EEG to ensure a correct classification. The removal artifacts in EEG signal is a
challenge and a crucial task.
Proceeding of the International Conference on Artificial Intelligence and Computer Science (AICS 2014), 15 - 16 September 2014, Bandung, INDONESIA. (e-ISBN 978-967-11768-8-7). Organized by http://WorldConferences.net 181
Figure 1:BCI block diagram (From data acquisition to Ocular artifacts removal are the main focus in
this research).
Independent Component Analysis (ICA) has the capability to remove the ocular artifact from brain
signal. ICA is a one of other popular method to separate the brain signal from ocular artifact in EEG
signal [10].The concept of ICA lies in the fact that the signals may be decomposed into their
constituent independent component. The mixed sources signal can be assumed independent from each
other, this concept plays crucial role in separation source signal from mixed signal [11]. In this paper,
ICA based Second Order Blind Identification with Robust Orthogonalization (SOBI-RO) is used to
remove ocular artifacts. The performance of the separation and extraction is measured through Signal
to Distortion Ratio (SDR) value. The SDR is needed to have information about how accurate SOBI-
RO in separating brain signal from ocular artifact in EEG signal.
II. MOTOR IMAGERY EXPERIMENT
Six healthy men with age of 21-22 years participated in motor imagery experiment. Brain signal
was recorded using MITSAR EEG 202 and the electrodes placement is shown in the Fig. 2.
MITSAR EEG 202 has 19 channels and 2 reference channels on the electrocap. Since this research
focus on ocular artifact removal, then the main observation is made for channel Fp1, Fp2, O1, O2 and
C3. In the experiment, the subjects were instructed to imagine of right hand movement and
blinkedtheir eyes in the same time to generate ocular artifactwhen the stimulus is displayed in the
monitor for two seconds. When the right arrow displayed on the monitor the subjects were instructed
to imagine of right hand movement and blink their eyes.For thirty seconds recording time, three
stimulus weredisplayed and five seconds for relaxing time between two stimulus.
III. OCULAR ARTIFACT REMOVAL METHODS
The method uses time structure when the independent components (ICs) are time signals, this is in
contrast to basic of ICA model which is mixed random variables. ICs may contain more structure than
Figure 2:Electrodes placement.
Proceeding of the International Conference on Artificial Intelligence and Computer Science (AICS 2014), 15 - 16 September 2014, Bandung, INDONESIA. (e-ISBN 978-967-11768-8-7). Organized by http://WorldConferences.net 182
simple random variables such as the autocovariances (covariances over different time delays) of the
ICs [12], the standard mixing model:
𝑥 = 𝐻𝑠 𝑘 (1)
where 𝑥 𝑘 is mixed signals and H is mixing matrix.
Before setting time delayed covariance matrices of mixed signals, formulating the robust
orthogonalization 𝑥 𝑘 = 𝑄𝑥 𝑘 must be done first. By using several time lags, up to 100 number of
time lags, the time delayed covariance matrices of mixed signal for preselected time delays
(𝑝1 ,𝑝2 , ……𝑝𝑘) are defined as:
𝑅𝑥 𝑝𝑖 = 1
𝑁 𝑥 𝑘 𝑥 𝑇 𝑘 − 𝑝𝑖
𝑵
𝒌=𝟏
= 𝑄𝑅𝑥 𝑝𝑖 𝑄𝑇 (2)
and then, the orthogonalized mixing matrix 𝐴 = 𝑄𝐻, perform Joint Aproximation Diagonalization
(JAD):
𝑅𝑥 𝑝𝑖 = 𝑄𝑅𝑥 𝑝𝑖 𝑄𝑇 = 𝐴𝑅𝑠 𝑝𝑖 𝐴
𝑇 = 𝑈𝐷𝑖𝑈𝑇 3
for 𝑖 = 1,2,3,… . 𝐿, ) JAD reduces the probability of un-identifiability of a mixing matrix caused by an
unfortunate choice of time delay 𝑝. Then orthogonal mixing matrix can be estimated as  = 𝑄Ĥ = 𝑈
and diagonal matrix 𝐷𝑖 𝑝𝑖 . Finally, the estimated of source signals as [13]:
Ŝ 𝑘 = 𝑈𝑇𝑄𝑥 𝑘 (4)
and the mixing matrix asĤ = 𝑄 + 𝑈.
IV. RESULT
The recording EEG signalusing winEEG software at 250 Hz sampling rate from motor imagery is
shown on the Fig. 3. The first sessionabout 30 seconds recorded EEG signal has three stimulus which
is shown by square mark. The masked signal according to the given stimulus has high amplitude in
several channels. When the subjects imagine the right handmovement and blink their eyes, there is a
spike in short period of time, it can reach upper than 100 µV. The spike is predicted as a result of the
subjects blink.
Inthe preprocessing, the signals are filtered using band pass filter (BPF) with the frequency range
from 0.5 Hz to 30 Hz. The frequency under 0.5 Hz is related to respiration and upper 30 Hz is related
to fast beta wave.Then the next step is to remove ocular artifact by using SOBI-RO. After removing
ocular artifact, the expected result is to get brain signal without ocular artifacts. Calculating SDR value
is one way to measure accuracy of separation of the proposed method. The preprocessed and the
extracted (using SOBI-RO) signals are shown in the Fig.4 and Fig. 5, respectively.In the Fig. 5(the
separated signals), the diffrent amplitude for each channels which indicated the higher one as a
accumulation of the artifact are shown. Those statement is proved from the mapping (i.e., to
showwhere is exactly the location of brain activity) of the separated signals as shown in the Fig. 7 and
Fig. 9 (the first subject) respectively.Good separation is marked by red color focusingon one channel
such as in the eleventh channel. Since this research only focus the channels Fp1, Fp2, O1,O2 and C3
(see Fig. 2), then each channel can be found in the channels 17, 1, 13, 10, and 12, respectively (see
Fig. 7). The channel 2 does not give a good separationfor channel Fp2 since that channel still
contaminated by others sources. Generally for all subject, most of the focusing channels are clearly
identified.
To evaluate te performance of the separation, one method called Signal to Distortion Ratio (SDR)
value is calculated, which is defined as
𝑆𝐷𝑅 𝑑𝐵 = 10𝑙𝑜𝑔10 𝑠𝑖 𝑘
2𝑘
𝑠𝑖 𝑘 − ŝ𝑖 𝑘 2
𝑘
(4)
where 𝑠𝑖 𝑘 is the pure motor imagery signal and ŝ𝑖 𝑘 is the estimation brain signal [14]. In this
research, the recorded signals without ayes blink is used as a pure motor imagery signal. Those
signals(see Fig. 6) are processed by using ICA algorithm Infomax in winEEG software which is
embedded with EEG system.
Proceeding of the International Conference on Artificial Intelligence and Computer Science (AICS 2014), 15 - 16 September 2014, Bandung, INDONESIA. (e-ISBN 978-967-11768-8-7). Organized by http://WorldConferences.net 183
Figure 4: Preprocessed Signal by using BPF for first subject
Figure3: Recorded EEG Signal ofthe first subject.
Proceeding of the International Conference on Artificial Intelligence and Computer Science (AICS 2014), 15 - 16 September 2014, Bandung, INDONESIA. (e-ISBN 978-967-11768-8-7). Organized by http://WorldConferences.net 184
Figure 5: Estimated Source for first subject in trial 1 (10-12 seconds).
Figure 6: Brain mapping from SOBI-RO in first subject for trial 1.
Figure 7: Pure motor imagery signal in first subject processed by ICA Infomax in winEEG software.
Time [s]
Inde
pend
ent
Com
pone
nt
5µV
1 501
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
1 2
19
Proceeding of the International Conference on Artificial Intelligence and Computer Science (AICS 2014), 15 - 16 September 2014, Bandung, INDONESIA. (e-ISBN 978-967-11768-8-7). Organized by http://WorldConferences.net 185
Then the SDR value are shown in Table 1. From Table 1, it can be concluded that not all observed
channels have SDR value (marked by dash)which indicated that the separation are not fully success.
Table 1: SDR value of six subjects
The best SDR value about 2.3413 dB is achieved in the channel 13 (indicated channel O1). As
shown in Fig. 7 and Fig. 6, channel 8 is very clear and lower amplitude compared with others. This
result indicated that that the ratio between original brain signal and its error is very small. The higher
SDR value gives the better separation accuracy of the signal from ocular artifacts.
V. CONCLUSION
The results presented in this study is from 30 seconds recording of EEG signal during Motor
imagery experiment using SOBI-RO algorithm. The proposed algorithm is success to remove ocular
artifacts for 62 trials from total 90 trials, with percentage 68.88 %. And the highest SDR value is
2.3413 dB in first subject for channel O1. It means that the estimation signal from SOBI-RO has a
little difference with the pure motor imagery signal that processed by winEEG software. Moreover, the
proposed algorithmdescribed herein can isolate correlated electroocular components with a high
degree of accuracy. Although the focus is on eliminating ocular artifacts in EEG data, the approach
can be extended to other sources of EEG contamination.
VI. ACKNOWLEDGMENT
This research was supported by the thematic program (No. 3425.001.013) through the Bandung
Technical Management Unit for Instrumentation Development (Deputy for Scientific Services) and
the competitive program (No. 079.01.06.044) through the Research Center for Metalurgy (Deputy for
Earth Sciences) funded by Indonesian Institute of Sciences, Indonesia.
VII. REFERENCES
A. Turnip, K.-S. Hong and M.-Y. Jeong, Real-time feature extraction of P300 component using
adaptive nonlinear principal component analysis, BioMedical Engineering OnLine, vol.10,
no.83, 2011.
Turnip, A., Soetraprawata, D., and Kusumandari, D. E., "A Comparison of EEG Processing Methods
to Improve the Performance of BCI, International Journal of Signal Processing Systems, vol.
1, No. 1, 2013.
Proceeding of the International Conference on Artificial Intelligence and Computer Science (AICS 2014), 15 - 16 September 2014, Bandung, INDONESIA. (e-ISBN 978-967-11768-8-7). Organized by http://WorldConferences.net 186
U. Hoffmann, J.-M. Vesin and T. Ebrahimi, An efficient P300-based brain-computer interface for
disabled subjects, Journal of Neuroscience Methods, vol.167, no.1, pp.115-125, 2008.
Turnip, A. and Kusumandari, D. E., Improvement of BCI performance through nonlinear
independent component analisis extraction, Journal of Computer, vol. 9, no. 3, pp. 688-695,
March 2014 (JCP, ISSN 1796-203X).
Mingai Li, Yan Cui and Jinfu Yang, Automatic Removal of Ocular Artifact from EEG with DWT
and ICA Method”, journal of Appl. Math. Inf.Sci 7, No.2, 809-816, 2013.
Turnip, A., Haryadi, Soetraprawata, D., and Kusumandari D. E., A Comparison of Extraction
Techniques for the rapid EEG-P300 Signals, Advanced Science Letters, vol. 20, no. 1, pp. 80-
85(6), January, 2014.
Fisch, B.J.,Artifacts. In: Fisch, B.J. (Ed.), Spehlmann’s EEG Primer, 2nd edition. Elsevier,
Amsterdam, The Netherlands, pp. 108 – 124.1991.
Z. J. Koles. The quantitative extraction and topographic mapping of the abnormal components in the
clinical EEG. Electroencephalogr. Clin. Neurophysiol., 79:440–447, 1991.
Turnip, A. and Sutraprawata, D., “Feature Extraction of EEG-P300 Signals Using Nonlinear
Independent Component Analysis”, International Journal of Mechanical & Mechatronics
Engineering, vol. 13, no. 2, pp. 38-45, 2013.
Ren Sin Tung, Wai Yie Leong, Processing Obstructive sleep apnea syndrome (OSAS) data. J.
Biomedical Science and Engineering, 6, 152-164,2013.
Saeid Sanei dan Chambers, J.A. Chambers. EEG Signal Processing. Center of Digital Signal
Processing. Cardiff University, UK. John Wiley & Sons, Ltd.2007.
Aapo Hyvärinen, Juga Karhunen dan Erkki Oja. Independent Component Analysis. John Wiley &
Sons, Inc. 2001.
Andrzej Cichocki and Shun-ichi Amari.Adaptive Blind Signal and Image Processing. John Wiley &
Sons.2002.
Janett Walter-Williams and Yan Li. BMICA-Independent Component Analysis Based On B-Spline
Mutual Information Estimastion Signals. Canadian Journal on Biomedical Engineering &
Technology vol.3 no.4, May 2012.