SOBI-RO for Automatic Removal of Electroocular Artifacts from EEG Data-Based Motor Imagery

Transcription

1 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 s: Abstract Signals from eye movements and blinks can be orders of magnitude larger than braingenerated 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 signalprocessing 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. 2014), September 2014, Bandung, INDONESIA. (e-isbn ). Organized by 180

2 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 years participated in motor imagery experiment. Brain signal was recorded using MITSAR EEG 202 and the electrodes placement is shown in the Fig. 2. Figure 2:Electrodes placement. 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 Organized by 181

3 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 (p 1, p 2, p k ) are defined as: R x p i = 1 N N k=1 x k x T k p i = QR x p i Q T (2) and then, the orthogonalized mixing matrix A = QH, perform Joint Aproximation Diagonalization (JAD): R x p i = QR x p i Q T = AR s p i A T = UD i U T 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 D i p i. Finally, the estimated of source signals as [13]: Ŝ k = U T Qx 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 k s i k 2 SDR db = 10log 10 s i k ŝ i k 2 (4) k where s i k is the pure motor imagery signal and ŝ i 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. Organized by 182

4 Figure3: Recorded EEG Signal ofthe first subject. Figure 4: Preprocessed Signal by using BPF for first subject 2014), September 2014, Bandung, INDONESIA. (e-isbn ). Organized by 183

5 Independent Component µV Time [s] 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. 2014), September 2014, Bandung, INDONESIA. (e-isbn ). Organized by 184

6 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 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 %. And the highest SDR value is 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 ) through the Bandung Technical Management Unit for Instrumentation Development (Deputy for Scientific Services) and the competitive program (No ) 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, 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, Organized by 185

7 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 , Turnip, A. and Kusumandari, D. E., Improvement of BCI performance through nonlinear independent component analisis extraction, Journal of Computer, vol. 9, no. 3, pp , March 2014 (JCP, ISSN X). 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, , 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 (6), January, Fisch, B.J.,Artifacts. In: Fisch, B.J. (Ed.), Spehlmann s EEG Primer, 2nd edition. Elsevier, Amsterdam, The Netherlands, pp Z. J. Koles. The quantitative extraction and topographic mapping of the abnormal components in the clinical EEG. Electroencephalogr. Clin. Neurophysiol., 79: , 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 , Ren Sin Tung, Wai Yie Leong, Processing Obstructive sleep apnea syndrome (OSAS) data. J. Biomedical Science and Engineering, 6, ,2013. Saeid Sanei dan Chambers, J.A. Chambers. EEG Signal Processing. Center of Digital Signal Processing. Cardiff University, UK. John Wiley & Sons, Ltd Aapo Hyvärinen, Juga Karhunen dan Erkki Oja. Independent Component Analysis. John Wiley & Sons, Inc Andrzej Cichocki and Shun-ichi Amari.Adaptive Blind Signal and Image Processing. John Wiley & Sons 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 Organized by 186