Denoising and Error Removal of EEG Signal using DWT and Smoothing Filters

Transcription

1 Deoisig ad Error Removal of EEG Sigal usig DWT ad Smoothig Filters Mahmoud I. Al-Kadi ad Mamu Bi Ibe Reaz Uiversiti Kebagsaa Malaysia Departmet of Electrical, Electroic & Systems Egieerig, Bagi, Selagor 43600, Malaysia ABSTRACT Electroecephalogram (EEG) sigals are sigificatly distorted i case of ay exteral iterferece which ievitably affects moitorig the Depth of Aesthesia (DOA). Durig scoliosis correctio surgeries, the error sample rate ad oise level are remarkably icreased because surgeos use high power electroic equipmet ad several kids of electric had tools. This research ivestigates the mai causes of EEG sigal distortio durig this kid of operatios ad discusses the deoisig process durig differet stages of aesthesia. This paper presets ad tests a ovel EEG deoisig techique for scoliosis correctio surgeries usig a hybrid combiatio of covetioal filters, Wavelet oise removal ad smoothig filters. The Savitzky-Golay (SG) smoothig filter removes the residual oise overlapped with EEG sigals ad also smooth the sharp poits resulted from removig high power oise, high speed samplig rate ad sudde chages i EEG sigals. The parameters of these filters are cotiuously chaged accordig to the oise level ad DOA i order to preserve the sigal characteristics. The experimetal outcomes preset that the resultat EEG data are sigificatly corrected ad fully deoised without ay substatial variatio i their compoets. KEYWORDS Scoliosis correctio surgeries; aesthesia; EEG sigal; errors; deoisig; smoothig. 1 INTRODUCTION Electroecephalogram sigal (EEG) is a data used to measure the activity of the brai usig may electrodes (chaels) arraged i fixed places o the scalp. Normally, the brai sigals are a fuctio of time ad are describable i terms of its amplitude, frequecy, ad phase. These sigals exploited to moitor the depth of aesthesia (DOA) ad diagose some diseases such as Alzheimer's, Parkiso's, Epilepsy, ad Stroke. Aesthesia is a idispesable sectio of surgery where the aesthesiologists moitor the DOA accordig to the variatios of the physiological parameters, such as blood pressure, heartbeat, breathig rate, eye movemet, ad their reactio to the physical excitatio due to the procedure of the surgery. Variatio the levels of aesthesia are strogly affected to the features of the EEG sigals; all these variatios are utilized to moitor the level of awareess [1]. Ucosciousess is defied as a o-resposiveess ad lack of movemet to the paiful stimuli, whereas the severity of operative stimulatio depeds o the type ad duratio of surgery []. There are two reasos lead to iappropriate geeral aesthesia; the first oe caused by uder dosage, which coduce a itraoperative awareess while the secod reaso caused by over dosage which leads to prologed aesthesia ad icreases the risk of postoperative complicatios. Geerally, the commo factor that specifies the appropriate level of geeral aesthesia is the ability to aalyze the brai waves to observe the DOA [3]. I the early eighties, the first stage of EEG sigals represeted by digital filters, which are exploited to suppress udesirable frequecy compoets from the observed EEG sigal. A trai of low ad high pass filters uses to remove the oise comig from the electrical lie [4]. ISBN: SDIWC 17

2 These chais of filters have opeed the way for a ew techique i which is a discrete wavelet trasformig (DWT). This techique depeds o a set of weights to geerate a cross multiplicatio of each elemet of EEG sigal with its eighbors [5]. The DWT is a effective techique to diagose the recorded sigals from several forms of artifacts that overlapped with EEG sigals such as ocular artifacts, iheret oise ad motio artifacts [6]. Other researchers combied the wavelet (WT) with aother techique such as Idepedet Compoet Aalysis (ICA) to produce a ew techique, Wavelet-Idepedet Compoet Aalysis (WICA). This formula showed the best performace to separate the oise from brai waves with miimal iformatio loss [7]. The aim of this paper is to aalyze ad deoise the acquired EEG sigal durig scoliosis correctio surgeries also ivestigates the causes of distortios ad oises durig differet stages of aesthesia. This techique proceeded to four stages to deoise the recorded EEG data. Iitial stage is a automatic error corrector or removal. The secod stage is a set of covetioal filters (Notch filter, IIR bad pass filter) to remove the uwated power ad frequecies. The third stage is a wavelet deoisig process to remove the artifacts that overlapped with recorded EEG sigals. Fially, icostat Savitzky-Golay smoothig filter (SG filter) to deoise the residual oise ad smoothig the chagig poits. The results exhibited that the acquired sigals corrected ad deoised sigificatly with maitaiig the features characteristics of the EEG sigal. METHODOLOGY The EEG sigals have bee recorded for forty hours from two chaels that were acquired from twety patiets who uderwet Scoliosis Correctio Surgeries at the UKM Medical Cetre, Jala Yaacob Latif, Badar Tu Razak, Cheras, Kuala Lumpur. The proposed system tests ad processes the sigal compoets (removig the errors ad deoisig) for each chael separately. This system cosists of four stages as show i Fig. 1. The first stage is proceeded to correct the error samples or delete the whole secod (18 sample) from the recorded waves. The secod stage devoted to removig the effect of AC lie ad uwated frequecies. The third stage used to remove a differet kid of oise, whether it s a extrisic oise or artifacts. The fourth stage is a icostat smoothig filter where the parameter of this filter varies accordig to the oise level. The stage oe ad two used to prepare the EEG sigals ad pipoitig the useful frequecies where stage three ad four used to deoise ad smoothig the prepared EEG sigals. Figure 1. The proposed system for error correctio ad oise removig to the recorded EEG sigals..1 Error Correctio or error cacellatio I this stage, the system checks the probability of error occurrece durig data recordig withi each secod (18 samples). Automatic ISBN: SDIWC 18

3 loop ca correct ad removes the stream of errors accordig to the umber of error samples per secod. The system ca correct the error samples whe the errors are sporadic ad few; this correctio depedig o the values of the previous five samples. If the burst of errors occurs as a result of the use of high-eergy devices, the system uable to correct the error samples; because the recorded sigal embedded iside the trai of errors ad the origial form sigal caot be covered at all; cosequetly, the system will remove the whole secod (18 samples).. Notch (BadStop) ad badpass digital filters The udesirable frequecies that overlapped with the recorded sigal is beig removed by otch filter NF ad badpass filters BPF. The effects of A.C. lie removed by otch filter, while the bad pass filters keepig oly the useful frequecies (0.1Hz to 63 Hz). The term digital filters refer to the mathematical procedures that applied to umeric ad discrete represetatios of cotiuous sigals to atteuate or emphasize certai frequecies. The time domai of discrete filterig of EEG sigals typically ivolves cross multiplyig each oisy data poit ad its eighbors with a group of weights. I fact, the sigal patter is represeted by this set of weights. Recurret cross multiplicatio process for each poit leads to trucate the whole sigal from some artifacts ad removes the D.C. level. The desigs of recursive IIR bad stop (otch filter) ad badpass digital filter takes ito cosideratio the curret ad past samples of the iput ad output values. Fially, the algorithm of this filter attempts to reduce the start-up ad edig trasiets by regulatig the iitial values to match the DC compoet of the sigal..3 Discrete wavelet deoisig process The wavelet deoisig process is assiged to removig the compoets of the various types of artifacts, such as ocular artifacts, iheret oise ad motio artifacts. This techique cosists of three steps: Step oe (decompositio step): fidig the wavelet coefficiets (approximatio ad details) by decomposig the EEG sigals. This decompositio was achieved by a series of lowpass (LPF) ad highpass filters (HPF). To pick up the approximatio coefficiets, the EEG sigal passes through LPF. At the same time, the EEG sigal is passed through a HPF to obtai the detail coefficiets with daw sample by two at each stage [8]. Step two (threshold step): pipoitig the threshold value of the details ad approximatio coefficiets which already have bee fouded i the previous step. The deoisig process has bee achieved by resettig the coefficiets have a absolute value lower tha the threshold level. Step three (recostructio step): recostructs the ew coefficiets by iverse wavelet (IDWT) with up sample by two at each stage [9]. Figure. The similarity betwee the EEG sigal ad Mother Wavelet (a) db4 (b) The variatio of two chaels of EEG sigals. Based o the studies, Daubechies db4, has bee used to aalyze the EEG sigal takig ito accout their optimality i the time-frequecy localizatio properties as show i Fig. (a). Its waveforms are similar to the waveforms to be detected i the EEG sigal as Fig. (b). From these figures, we ca observe distictly that the variace of the recorded EEG sigal ISBN: SDIWC 19

4 has the same variatio of the mother wavelet which leads to deoisig ad artifact removig from the EEG sigal accurately. Accordig to the samplig frequecy, level four is chose based o the domiat frequecy compoets of the EEG sigals..4 Savitzky- Golay smoothig filter The researchers Abraham Savitzky ad Marcel J. E. Golay suggested a method for smoothig the data based o local least squares polyomial approximatio [10]. To desig SG filter, we should defie the order of the polyomial N ad the frame size M which represets to the half width of the of the approximatio iterval. The algorithm of this method is i two stages; firstly, fittig a polyomial to a set of iput samples; secodly, estimate a ew polyomial at a sigle poit withi the approximatio iterval is equivalet to discrete covolutio with a fixed impulse respose. There are several coditios that must be take ito accout whe desigig SG smoothig filter specifically, the order of the polyomial must be strictly less tha the frame size. Moreover, if the order of N is too large, the approximatio problem is badly coditioed ad the solutio will be of o value. I this research, automatic chages i the parameters of the smoothig filter accordig to the oise level ca give the best results..5 Evaluatio criteria To assess the performace of the proposed system at this research; we used four differet criteria: Wavelet Coherece WC (1), Sigal-to- Noise Ratio SNR (), Cross Correlatio Fuctio CCF (3) ad Mea Squared Error MSE (4). The wavelet coherece used to test the performace of the SG filter with the ormal ad oisy EEG sigals. The other three criteria have bee applied to test the performace of deoisig stage (wavelet techique ad SG smoothig filter). 1 xx S ( s W ( )) s R ( S ) -1 x 1 x S(s W ( s) ). S ( S W ( s) ) SNR 10 log CCF MSE xx (1) x( x ( ) ) x( ) N m 1 * xm x ( m) 0 CCF xx ( m) m 0 m 0 () (3) 1 N x( ) x ( ) (4) N 1 where x () is the origi EEG sigal, x () is the deoised EEG sigal, * sig is the complex cojugate ad S is a smoothig operator. The best performace of the proposed system whe achievig the highest value of the SNR, lower value of the MSE ad fially, the highest value of the poits of itersectio of CCF. 3 EXPERIMENTAL RESULTS AND DISCUSSION The Fast Fourier Trasform FFT for each secod to follow chagig i the EEG sigal compoets durig deoisig stages are observed. For the FFT aalysis, the phase spectrum yields iformatio about the phase of compoets relative to the start of the computatio widow. The results of SG filter divided ito two categories; the first oe verifies the coherece betwee the outputs of stage three ad four usig oly oe secod to follow the chages i the EEG sigal clearly, these variatios as a result of usig differet values of the polyomial ad the frame size (smoothig the EEG sigal). The secod form of results, verifies the SNR, MSE ad CCF betwee the outputs of stage three ad four to follow the performace of wavelet deoisig process ad SG filter (deoisig EEG sigal). ISBN: SDIWC 0

5 3.1 Error correctio The occurrece of the errors durig data acquisitio directly affects i measurig the depth of aesthesia. The sample is a error whe its value is egative or much lower tha the value of the overall average of samples per secod. If the umber of errors is sporadic ad few as show i Fig. 3 (a), the error samples have bee estimated by averagig the five previous samples as show i Fig. 3 (b). If the error samples are cosecutive ad exceedig 3 samples (quarter of a secod) as give i Fig. 3 (c), that mea it s a burst of errors ad the system caot estimate these sample values. Thereby, the system automatically removes this secod (18 samples) ad gives a error sig. Figure 3. EEG sigal with the differet kid of errors (a) sporadic error; (b) corrected form; ad (c) burst of errors. figure shows a high level of the cumulative value of the sigal due to the presece of DC level. Figure 4. Two chaels of the EEG sigal (a) ufiltered sigal; (b) filtered sigal by BSF; (c) FFT of the filtered sigals. The badpass filter has bee used to specify the effective frequecies that represet a EEG sigal which varies accordig to the depth of aesthesia, ad are cofied betwee 0.1 ad 63 Hz. I the same cotext; the badpass filter removes some artifacts such as EMG ad ocular which has frequecies more tha 60 Hz; also remove the D.C. level that geerated by the rhythm of breathig or by some of electroic equipmet. 3. Covetioal filters The secod stage icluded two types of filters: badstop filter BSF ad badpass filters BPF. The badstop filter is used i all EEG recordig devices to elimiate the effects of the AC lie ad its harmoics (49-51) Hz. I fact, this filter has bee used to make sure that this type of oise was completely removed from the frequecy of EEG sigals, because this kid of oise specifically overlaps with the Gamma bad which leadig to icrease the amplitude i those frequecies. Fig. 4 (a, b) illustrates a 60 secod of EEG sigal ad the filtered form respectively. We ote that the filtered sigal did ot chage sigificatly from the origial sigal ad the EEG sigals still settled o a high DC level (aroud 1470 µv). Fig. 4 (c) shows the FFT for each secod of the filtered sigal. This Figure 5. (a) Two chaels of the EEG sigal filtered by PSF ad BPF; (b); FFT of the filtered sigals. Fig. 5 (a) shows the EEG sigals after removig the uwated frequecies ad D.C power by BPF where we ote that the brai sigal is oscillate aroud the zero level. Fig. 5 (b) illustrated the FFT of the sigal where remaied oly the useful frequecies ad the amplitude is decreased sigificatly [34]. ISBN: SDIWC 1

6 3.3 Wavelet deoisig techique I the third stage, the system removes the artifacts that overlapped with recorded EEG sigal usig wavelet techique. The wavelet trasform is particularly effective for represetig various aspects of sigals such as treds, discotiuities, ad repeated patters; it is especially powerful for o-statioary sigal aalysis where EEG sigals cotai these evets. Cosequetly, the wavelet trasform is a robust tool for aalyzig trasiet sigals because the iformatio of time ad frequecy ca be obtaied. Furthermore, if the basis wavelet fuctio has a fiite duratio, the the frequecy iformatio obtaied from the wavelet trasform is localized i time. Figure 6. (a) Two chaels of the EEG sigal filtered by BSF, BPF ad WT Techique; (b); FFT of the filtered sigals. Fig. 6 (a, b) shows the EEG sigals ad its FFT after passig through the WT deoisig techique. From this figure, the most of artifacts ad oise that overlapped with recoded sigal are removed without affectig to the details of the EEG sigals. The performace of this stage has bee measured by averagig the SNR, CCF, ad MSE betwee the output of BPF filter ad the output of wavelet deoisig techique i six patiets oly; with 60 secods of EEG sigals for each patiet durig surgery stage as show i Table 1. Add to that, the EEG sigals have bee chose with ormal oise for the patiets 1 ad 6 while we chose EEG sigals with high level of oise for the patiets, 3, 4 ad 5. Table 1. The performace of wavelet deoisig techique Patie ts SNR (db) CCF MSE Ch. 1 Ch. Ch.1 Ch. Ch. 1 Ch From Table 1, we otice that the values of SNR, CCF, ad MSE varied from chael to chael ad from patiet to aother accordig to the oise ad artifacts stregth. The SNR is high whe the EEG sigals have a ormal oise while the wavelet techique gave a low SNR for oisy sigals. The MSE has the same behavior, but it s lowest whe the EEG sigals have a ormal oise ad high from oisy sigals. The good thigs from this result that the CCF gave a good value for all patiets that mea the filtered sigal still correlate with the previous stage. All these results showed that we eed aother stage to remove the remaiig oise ad its effects [35]. 3.4 S-G filter The SG filter is a additioal filter used to remove the residual oise that overlapped with EEG sigals also smoothig the sharp poits that resulted from the removig a high power oise, high speed of samplig ad fially; the sudde chage of the EEG sigals (ature of EEG sigal). ISBN: SDIWC

7 because this order is so eough for the EEG sigals ad does t sped a log processig time while the frame size M was take as a variable odd umber. Figure 7. (a) Two chaels of the EEG sigal filtered by SG filter; (b); FFT of the filtered sigals. Fig. 7 illustrates the filtered EEG sigal ad its FFT by SG filter. At first glace, there is o sigificat chages betwee the stages three ad four (Figs. 6 ad 7) but i fact, there is a sigificat chage i the form of the sigal. These chages were ot appearig clearly i Fig. 6 because the sigal is relatively log ad effect of this filter cocetrated o the details of the EEG sigal. Therefore, we will explai the effect of the SG filter i details i the ext sectios. The value of M ad N is the mai parameters that idetify the fuctio of the SG filter whether smoothig or deoisig. To get a good result for smoothig or deoisig the EEG sigal by SG filter, we should adjust the relatio betwee these parameters accurately. I this research; we chose 5, 9, 13, ad 17 as low values of the frame size ad we chose 31, 35, 39, ad 41 as high values of the frame size. Fig. 8 (a) represets the EEG data before filterig by SG filter (i.e. The data filtered by BSF, BPF, ad WT). Fig. 8 (b ad c) showed the filtered EEG sigal with the low ad high group of the frame size values. This figure shows the chages i the details of the EEG wave with icreasig the differece betwee M ad N. 3.5 Smoothig the EEG sigals usig SG filter To use the SG filter for smoothig the EEG sigal from the sharp poits resultig from the previous filters or high speed samplig rate, there must be a sigificat differece betwee the polyomial order value ad the frame size value. Icreasig the differece betwee M ad N values leads to icrease the smoothig process. I this case, the SG filter dedicated to smoothig the EEG sigal without removig the oise efficietly (high smoothig-low deoisig). To follow these chages i details, we aalyzed oly oe secod for two chaels of EEG sigal as show i Fig. 8. The order of the polyomial N has bee idetified by 3 Figure 8. Oe secod-two chaels of EEG sigal (a) before SG filter (b) after SG filter with low values of M (c) after SG filter with high values of M. Icreasig the value of M makes the sigal more smooth where the sudde iflectio poits begi to disappear as evidet i Fig. 8 (b). Icreasig the value of M too much will chage the shape of the sigal ad may lose valuable details which reflect the accuracy of estimatig the depth of aesthesia as show i ISBN: SDIWC 3

8 Fig. 8 (c). I this research; the frame size 9 or 13 is a good value which gave acceptable results ad keeps the importat features i safe. The coherece betwee the output of stage three ad four has bee used to assess the performace of the SG filter after smoothig the EEG sigal uder various parameters where the coherece has bee represeted by a colorimeter. Fig. 9 illustrates the coherece (aalyzed sigal - wavelet coherece - modulus ad phase) for chael 1 by testig oe secod (18 samples) of oisy ad ormal EEG sigals with a differet frame size. I this research, we aalyzed oly oe chael because chaels 1 ad mostly have same variatios but differet power. The red color represets the highest coherece, while the blue color represets a lack of coherece betwee the tested sigals. The arrows i these figures represet the relative phase betwee the two sigals as a fuctio of scale ad positio. The plot of the relative phases is superimposed o the wavelet coherece. From Fig. 9 (a) represets a oisy EEG sigal; the lowest value of M gave a high coherece betwee filtered ad ufiltered sigals because this value smoothed the whole sigal lightly without effective chage i the filtered sigal as metioed before. The highest value of M lost the coherece as a result of chages i the sigal shape due to the strog smoothig process to the sigal. The directio of the arrows shows that the sigals are i phase except some poits, which meas that the SG filter does't chage the directio of the sigal durig the smoothig process. From this figure; the frame size value 13 gives a acceptable coherece which cofirms the result that obtaied i Fig. 8. Sometimes the EEG sigals do t cotai a high artifact, oise ad iflectio poits. Figure 9. The aalyzed sigal-wavelet coherecemodulus ad phase betwee the outputs of WT deoisig stage ad SG filter stage for oe secod of oisy sigal with a differet frame size (a) oisy sigal (b) ormal sigal. Fig. 9 (b) shows the coherece of oe secod of ormal sigals with a differet frame size. From this figure; icreasig the value of M dramatically does't chage the results affectigly. Therefore, the system ca chage the parameters of the filter automatically by choosig the appropriate value for the frame size accordig to the form of the acquired EEG sigal to reduce the processig time, which leads to estimate the depth of aesthesia quickly ad accurately. The proposed system will automatically chage the parameters of the filter accordig to the values of the three criteria SNR, CCF, ad MSE. Fig. 9 illustrates the criteria which used to assess the performace of the SG filter. From Fig. 10 (a), the SNR was decreased whe the differece betwee the parameters of SG filter icreased. This behavior meas that the deoisig process goig to decrease wheever the differece betwee M ad N have bee ISBN: SDIWC 4

9 icreased, this case compatible with that metioed above (low deoise). bee chose as follows: (3,5), (7,9), (11,13), (15,17), (19,1), (3,5), (7,9), (31,33). Figure. 10. The filter performace for 6 patiets with fixit polyomial N=3 ad differet frame size M (a) SNR i db (b) CCF (c) MSE. Fig. 10 (b ad c) shows that the CCF decreased while MSE icreased with icreasig the differece betwee the SG parameters. Decreasig CCF ad icreasig MSE was came from icreasig the smoothig percetage to the sudde iflectio poits which leadig to retreatig the cogruece betwee the waves that bee tested (before ad after SG filter). We coclude from the foregoig, the parameters of the SG filter must be set very carefully to prevet ay loss to the EEG features that are very importat to determie the depth of aesthesia. 3.6 Deoisig the EEG sigals usig SG filter Whe the differece betwee the values of the filter parameters few (fixed at ); the SG filter is workig as a oise removal to the EEG sigal with little of smoothig (low smoothighigh deoisig). The same criteria have bee used to test the performace of the SG filter for deoisig a sigal with differet groups of parameters. These groups of N ad M have Figure. 11. The filter performace of SG filter with the differet polyomial N ad frame size M (a) SNR i db (b) CCF (c) MSE. Fig. 11 illustrates the performace of the SG filter for chael 1 where chael 1 ad has same variatios but differet i level of the power. Icreasig the parameters of SG filter leads to icrease the values of SNR ad CCF as show i Fig. 11 (a, b). This icrease meas that the SG filter removes the oise from the EEG sigal efficietly without chagig i the sigal form (i.e. High correlatios betwee the filtered ad ufiltered sigals). If the parameters of the filter icreased too much; the values of criteria SNR ad CCF will decrease sharply as show i the last poit i Fig. 11 (a, b). This regressio is a result of the excessive filterig of the EEG sigal it meas we reached to the saturatio poit. I the same cotext; icreasig the values of the parameters leads to decreasig the MSE as show i Fig. 11 (c). This criterio idicates that the errors which resulted from the deoisig process have bee decreased with icreasig the parameters. As well as, icreasig the parameters too much leads to growig the errors as a result of excessive filterig as show i the last poit i Fig. 11 (c). Fially, icrease the values of SNR ad CCF, ad decrease the values of MSE are cofirmig the effectiveess of the SG filter to ISBN: SDIWC 5

10 remove the oise from EEG sigals with avoidace the excessive icreasig the values of N ad M. 4 CONCLUSIONS The brai sigals are ofte distorted due to the oise ad artifacts which have bee raised from may sources. Durig scoliosis correctio surgery, the surgeos, urses, ad physiologist use a lot of had tools ad electroic equipmet that icrease the rate of errors ad oise level to the evets of EEG sigals. This research suggested filterig procedure to trucate the oise ad errors from the acquired wave. Practically; this procedure made up of four steps; Automatic Error Corrector, Covetioal Filters (otch filter ad IIR bad pass filter), Wavelet deoisig process, ad SG filter. The BSF rejects the effects of A.C lie (49-51) Hz, while the BPF passed oly the useful frequecies (0.1-63) Hz. The WT deoised the most of the artifacts that overlapped with recorded sigals. The SG filters smooth the sigal as well as removig the residual artifacts ad oise but with the strict rules. The performace of this filter depeds o the order of the polyomial N ad the frame size M. If the differece betwee these parameters is high; the SG filter dedicated to smoothig the EEG sigal (high smoothig-low deoisig). Whe the differece betwee these parameters are few; the SG filter works as a oise removal to the EEG sigal (low smoothig- high deoisig). To get a good smoothig ad deoisig results usig SG filter, we must balace betwee the values of M ad N. The SNR, CCF, ad MSE gave a good idicatio whe the SG filter used to deoise or smooth the EEG sigal durig a very oisy surgery such as scoliosis correctio surgeries. The suggested procedure chages the parameters of the filter accordig to the stregth of the oise ad the artifacts. This group of filters fudametally works as a local polyomial regressio by settig the smoothed values for each data poit (each secod) idividually. This procedure expressed the high superiority for preservig the features of the data, such as amplitude ad frequecies, which are usually 'washed out' by adjacet averagig. Fially, this research provides a efficiet method to solve the filterig problems for the EEG sigal durig oisy surgeries with log duratio of aesthesia. REFERENCES [1] Musizza B, Ribaric S (010) Moitorig the Depth of Aaesthesia. Sesors, 10: [] Sebel P, Bowdle T, Ghoeim M, Rampil I, Padilla R, Ga T, Domio K (004) The icidece of awareess durig aesthesia: A multiceter Uited States study. Aesth. Aalg., 99: [3] Lågsjö J, Maksimow A, Salmi E, Kaisti K, Aalto S, Oikoe V, Hikka S, Aataa R, Sipilä H, Viljae T, et al. (005) S-ketamie aesthesia icreases cerebral blood flow i excess of the metabolic eeds i humas. Aesthesiology, 103: [4] Va Zae J, Murray MM, Meuli RA, Vesi J-M (013) Adaptive Filterig Methods for Idetifyig Cross-Frequecy Coupligs i Huma EEG. PLoS ONE 8(4): e [5] Al-Kadi M I, Reaz M B I, Mohd Ali M (013) Evolutio of Electroecephalogram Sigal Aalysis Techiques durig Aesthesia. Sesors, 13: [6] Slobouov S, Cao C, Sebastiaelli W (009) Differetial effect of first versus secod cocussive episodes o wavelet iformatio quality of EEG. Cli. Neurophysiol, 10(5): [7] Walters-Williams J, Li Y (011) Performace compariso of kow ICA algorithms to a wavelet- ICA merger. Sigal Process, 5 (3):80-9. [8] Hussai M S, Reaz M B I, Mohd-Yasi F, Ibrahimy M I (009) Electromyography sigal aalysis usig wavelet trasform ad higher order statistics to determie muscle cotractio. Expert Syst., 6(1): [9] Ye H, Deg G, Mauger SJ, Hersbach AA, Dawso PW, et al. (013) A Wavelet-Based Noise Reductio Algorithm ad Its Cliical Evaluatio i Cochlear Implats. PLoS ONE 8(9): e7566. [10] Hargittai S (005) Savitzky-golay least-squares polyomial filters i ECG sigal processig. IEEE Comput. Cardiol. 3: ISBN: SDIWC 6