BioMedical Engineering OnLine

Transcription

1 BioMedical Engineering OnLine BioMed Central Research Treor suppression in ECG Ivan A Dotsinsky* and Georgy S Mihov Open Access Address: Center of Bioedical Engineering, Bulgarian Acadey of Sciences, Acad. G. Bonchev str., bl. 5, 3 Sofia, Bulgaria and Technical University of Sofia, Faculty of Electronic Engineering and Technologies, Klient Ohridski str. 8, Sofia, Bulgaria E-ail: Ivan A Dotsinsky* - iadoc@bas.bg; Georgy S Mihov - gs@tu-sofia.bg; *Corresponding author Published: 9 Noveber 8 Received: 7 May 8 BioMedical Engineering OnLine 8, 7:9 doi:.86/475-95x-7-9 Accepted: 9 Noveber 8 This article is available fro: 8 Dotsinsky and Mihov; licensee BioMed Central Ltd. This is an Open Access article distributed under the ters of the Creative Coons Attribution License ( which perits unrestricted use, distribution, and reproduction in any ediu, provided the original work is properly cited. Abstract Background: Electrocardiogra recordings are very often containated by high-frequency noise usually power-line interference and EMG disturbances (treor). Specific ethod for interference cancellation without affecting the proper ECG coponents, called subtraction procedure, was developed soe two decades ago. Filtering out the treor reains aprioripartially successful since it has a relatively wide spectru, which overlaps the useful ECG frequency band. Method: The proposed ethod for treor suppression ipleents the following three procedures. Containated ECG signals are subjected to oving averaging (cob filter with linear phase characteristic) with first zero set at 5 Hz to suppress treor and PL interference siultaneously. The reduced peaks of QRS coplexes and other relatively high and steep ECG waves are then restored by an introduced by us procedure called linearly-angular, so that the useful high frequency coponents are preserved in the range specified by the ebedded in the ECG instruent filter, usually up to 5 Hz. Finally, a Savitzky-Golay soothing filter is applied for suppleentary treor suppression outside the QRS coplexes. Results: The results obtained show a low level of the residual EMG disturbances together with negligible distortion of the wave shapes regardless of rhyth and orphology changes. Background Electrocardiogra (ECG) recordings are very often containated by residual power-line (PL) interference [-4], base-line drift [5-7], artefacts and EMG disturbances due to involuntary uscle contractions (treor) of the patient [8-]. The base-line drift resulting fro electrocheical processes at the electrode-to-skin barrier [7] is a typical low-frequency noise that distorts the susceptible ST segent [6, 3]. Interference and treor have overlapping frequency bands. Therefore, any algoriths are aied at their coon suppression [4-7] in order to provide an accurate autoatic delineation of the ECG wave boundaries [8]. Specific digital filter for PL interference cancellation, called subtraction procedure, has been developed soe two decades ago and peranently iproved later on [9]. It does not affect the signal frequency coponents around the rated PL frequency. Moving averaging is applied on linear segents of the signal (usually found in the PQ and TP intervals, but also in sufficiently long straight parts of the R and T waves) to reove the interference coponents. They are stored as phase locked corrections and further subtracted fro the signal wherever non-linear segents are encountered, e.g. QRS coplexes or other high and steep waves. Several criteria for linearity have been tested and ipleented Page of

2 BioMedical Engineering OnLine 8, 7:9 depending on the purpose. In general, they are based on the second difference of the signal (atheatical evaluation of the curvature). Filtering out the treor is a priori partially successful since it has a relatively wide spectru, which covers the useful ECG frequency band. One of the first recoendations for ECG instruents [] suggests a low-pass filter with iniu 35 Hz cut-off. However, in this way the aplitudes of sharp QRS waves are reduced. The oving averaging (cob filter with linear phase characteristic) gives siilar results []. ThetieaveragingisoneoftheclassicethodsforECG noise suppression. It is based on the assuption that the ECG signal is repeatable []. As the variability of the ECG orphology is also suppressed, soe authors [3, 4] proposed adaptive triggered filtering. Another way to preserve the ECG individuality is to reduce the nuber of the averaged beats but thus the effect of noise suppression is decreased. The variable ECG orphology, which is related to the respiration, ay be copensated in ultilead recordings by spatial transforations [4]. However, they can not be applied in the case of single channel tie alignent. Kotas [5] published projective filtering of tie-aligned ECG beats. This is an extension of tie averaging, which preserves the variability of the beat orphology. The ethod eploys the rules of principal coponent analysis for the desired ECG reconstruction and ais to retain to soe extent the deviations fro the averaged coponent changes, in the sae tie, rejecting deviations caused by noise. However, the nonlinear projective filtering is coputationally intensive and is known to be sensitive to noise changes. Adaptive filtration has been also attepted but with liited success because the QRS coplexes disturb the adaptationprocessuptotheendofthet-waves[4]. Luo and Topkins [8] obtained faster convergence using additional EMG channel as reference input. Bensadoun et al [9] proposed a ultidiensional ethod but the reduction of sharp Q-waves aplitudes is too high. Clifford et al [6] reported a odel-based filtering ethod. P-, Q-, R-, S- and T-waves are defined by a Gaussian with three paraeters: aplitude, width and relative position with respect to the R-peak. T-wave is described by T + and T - because of its asyetric turning point. Non-linear leastsquares optiization is applied to fit this ECG odel to the observed signal. The authors present one cleanly recorded P- QRS-T interval superiposed by electrode otion noise. The result shows alost total noise suppression but also significant wavefor distortions. However, the locations of the wave peaks atch the uncorrupted signal; the errors around the isoelectric line and the S-T segent are negligible. Thus, uch of the clinical inforation of the beats is captured after the noise reoval. Nevertheless, the error tolerance has to be tested over a set of databases, since non-paraeterized beat will be considered to be an artefact, while soe artefacts ay closely reseble a known beat. An iportant advantage of the ethod is the alost total eliination of series of pulses (artefacts). Saeni et al [7] proposed a nonlinear Bayesian filtering fraework consisting of Extended Kalan Filter (EKF), Extended Kalan Soother (EKS) and Unscented Kalan Filter (UKF) as suboptial filtering schees. They are based on odified dynaic ECG odel thus utilizing a priori inforation about the underlying dynaics of ECG signals. Recordings taken fro the MIT-BIH Noral Sinus Rhyth Database are superiposed by artificially generated noise. They are used for off-line testing EKF, EKS and UKF together with Wavelet denoising technique, adaptive and FIR filtering. A best SNR iproveent (difference between output and input SNR) of about db is obtained with the fraework filters. The authors found that brady- or tachycardia do not considerably affect the filter perforance, while other abnoralities appearing in soe of the ECG cycles ay lead to large errors in the Gaussian functions locations. Besides, neither the odel nor the easureent is reliable for filtering signals with low input SNR. Therefore, an accurate denoising of abnoral ECGs with high orphological changes reains an open proble. Christov and Daskalov [] applied an adopted by Savitzky and Golay [8] soothing procedure, which uses least square approxiation and a special 'wings' function for defining the weighting coefficients. The obtained suppression ration of the EMG artefact is about 6. Low reduction of R and S waves is reported depending ofthewaveshape. Nikolaev and Gotchev [] denoised ECG signals by applying wavelet doain Wiener filtering. They ixed original signals with EMG noise with a SNR = 4 db. Two-stage algorith iproves the traditional technique by involving tie-frequency dependent threshold for calculating the first stage pilot estiate. A SNR over db is obtained together with less than % QRS aplitudes reduction. In another paper Nikolaev et al [] reported an SNR iproveent of ore than db. Another technique for applying the subtraction procedure in the case of treor is reported by Christov [6]. The approach introduces adaptive criterion for linearity detection based on the ratio R between the linear segents length in a selected epoch and its total length usually chosen about s. Norally, the criterion Page of

3 BioMedical Engineering OnLine 8, 7:9 threshold M is a constant, which is set fro to 6 μv [9]. In the referred publication [6], M starts fro a low value of 5 μv andincreasesuntilr reaches a pre-selected value, e.g..9 that corresponds to QRS coplex and free of noise RR interval with noral diensions. The results obtained show a reasonable coproise between treor suppression and QRS aplitudes reduction. Gotchev et al [9] applied Savitzky-Golay filter inside the QRS coplexes and wavelet shrinkage outside the. The first technique gives a good preservation of the RS aplitude of about 3 μv but with low treor suppression, while the second one offers good suppression with 44 μv decreasing in the RS aplitude. The cobined ethod incorporates the features of both approaches. They are switched depending on the value W of the 'wings' function. W < is taken as dynaic order of the Savitzky- Golay filter; a higher value calls the wavelet subroutine. When the cob filter is used as a step of the subtraction procedure [9], the signal inside the QRS coplexes is not subjected to oving averaging. Thus, the QRS peaks are preserved but in the presence of treor the coplexes becoe corrupted and the linear segents are not detected correctly, the last leading to: i) unsuppressed disturbance in false non-linear segents, and ii) rare re-calculation of the phase corrections, which can not follow the changes of the interference aplitudes. These probles are overcoe to soe extent by Dotsinsky and Christov [7], who introduced a parallel buffer. The cob filtering is applied there over the entire signal, thus allowing precise location of the linear segents. However, the possibility of denoising the QRS coplexes by inappropriate treor coponents as a part of the calculated phase corrections still reains. Ai of the study The purpose of this work was to develop real-tie going ethod and algorith for suppressing both treor and PL interference in single- or ultilead ECG regardless of SNR, wave shapes and orphology changes. Methods and aterials The developed ethod for treor suppression in ECG ipleents the following three procedures: Containated ECG signals are subjected to oving averaging (cob filter with linear phase characteristic) with first zero set at 5 Hz to suppress treor and PL interference together. The reduced peaks of the processed signal are then restored by an introduced by us procedure called linearlyangular, thus the useful high frequency coponents are preserved in the range specified by the ebedded in the ECG instruent filter, usually up to 5 Hz. Finally, a Savitzky-Golay soothing filter is applied for suppleentary treor suppression outside the QRS coplexes. About 8 episodes consisting of several RR intervals are extracted fro 5 AHA database recordings [3]. They are preliinary oving averaged to suppress any undefined inherent noise. The obtained signals are called 'conditionally clean'. The sapling rate is 5 Hz, the resolution is 5 μv/bit. In the first part of the study the conditionally clean signals are used for developing the recovery procedure and evaluation of its correctness. For this purpose clean signals are cob filtered and then restored. Input and output signals are copared to assess the distortions introduced by the recovery. In the second part of the study the clean signals are ixed with synthesized 5 Hz PL interference and treor obtained by two ECG electrodes placed on one forear. The ixed signals are subjected to all procedures. The obtained results are analysed to evaluate the treor suppression and PL interference cancellation. In the third part of the study the procedures are applied directly on noisy recordings taken fro the AHA database and MIT-BIH Noise Stress Database. Signal recovery Basic relations between filtered and non-filtered saples The forulae for calculating the iddle ter in oving averaging over n saples for odd n =+ and even n = [9] are presented below: Y X n Y X Xi Xi i = i j = + i = i j n +, ; + n, n =. j= j= ( ) Here is integer, n is equal to the sapling rate divided by the rated interference frequency; i stands for the position of the ongoing averaged saple Y i, which is obtained over surrounding non-averaged saples. Taking in consideration that Xi+ j = Xi+ j + X + X and Xi+ j = X j= j= i j j= j= i i+ j j=, equation () can be expressed by Page 3 of

4 BioMedical Engineering OnLine 8, 7:9 Yi = ( Xi j + Xi j )+ X i n n +, = + j=. Yi = ( Xi j + X Xi Xi i+ j ) X i n n , = j= Substituting Xi = nxi Xi, n = + ; j= X Xi = nxi X i i +, n =, equation () is j= transfored in X i-n X i-k X i-( n-) X i X i- Y i X i+ Y i- Y i+ X i+k X i+ ( n-) X i+n Yi = Xi + ( Xi j Xi + Xi j ) n n +, = + j=. X Yi = Xi + Xi+ j Xi X i X i X + + ( i i j )+ + n = n, j= The polynoial inside the parentheses is a second difference, represents one of the possible versions of the linear criterion [9] and is further denoted as D i,j = X i+j -X i + X i-j =(X i+j - X i )-(X i - X i-j ). Using this equation, the saples X i and Y i can be expressed by Y X D n Y X D Di i = i + i j = + i = i + i j + n,, ;, n,, n =. j= j= The ean signal velocities on the left and the right hand X side of the ongoing saple X i are i X i j v ii, = j and j X i + j X i v i + j i =. They are averaged within the intervals, j [i-j, i] and[i, i+j], since they correspond to the tiecoordinates i+j/ and i-j/. Then, equation (4) is presented as ( ) = + Yi = Xi + j vi+ j, i vi, i j, n n j=. ( v + ) Yi = Xi + j( vi+ j, i vi, i j )+ = n i, i v ii,, n j= Background of the linearly-angular recovery procedure Let us assue that the conditionally clean signal is linear aside fro the ongoing saple X i and has a triangularlike shape (Fig. ). Then v i+j,i = v r, j = i+,..., i+n; v i+j,i = v l, j = i-n,...,i- and the difference v i+j,i - v i,i-j = v r - v l as well as the ratio D i,j /j = D i,k /k, k =,,...,n are constant. Figure Linearly-angular recovery of the signal in the interval [i-n,..., i+n]. The equation (5) is transfored into a unifor expression both for odd and even nuber of averaged saples: Y X v v X Di k i = i + ( r l) kn = i +, ( vr vl) kn, k where the constant n is given by k n n n = j = n = + kn = j + n n, ; n n =, 8 =. 8 j= j= Equation (6) can be written as k X Y D n i = i i k = Yi Di kk k,,, Analogously to the second difference D i,k, a filtered second difference D ik, = Y i+k -Y i + Y i-k is introduced using filtered signal saples. Substituting D i,k = h D ik,, the back filtered saple X* i can be calculated by i i i k X = Y h D, K, The coefficient h is intended to consider the real signal shapes. For the tie being, this study presues that h is very close to. The influence of k on the back filtering error is assessed by experients with k =,,3,4,5;n =5andM =, V. The error coitted is inial with k =,which value is further used. Lower value of k contributes to better shape recovery of rounded peaks, while the steeper ones are sub-copensated. Higher k value restores well steep peaks, but the rounded ones becoe overcopensated. Page 4 of

5 BioMedical Engineering OnLine 8, 7:9 conditionally clean AHA 79d signal conditionally clean AHA 8d signal error -. error Figure Coparisonbetween'clean'andrestoredAHA 79d episode. Figure 4 Coparisonbetween'clean'andrestoredAHA 8d episode. Assessent of the recovery procedure The recovery evaluation is illustrated by episodes of soe AHA signals shown in Fig., 3, 4, 5. They present different ECG rhyth and wave shapes: QRS coplexes + ectopic beats (Fig. ), high and steep QRS coplexes (Fig. 3), high T waves (Fig. 4), high P wave + ST depression (Fig. 5). The two upper traces are clean and processed signals, respectively. The lower traces deonstrate an error coitted in the range of 3%. No loss of clinical inforation is observed. The results obtained with the other episodes taken fro the 5 AHA recordings are identical or better. These episodes are listed in Table with their starting and ending ties. The recovery is assessed without additional suppression outside the QRS coplexes in order to have statistically the sae residual noise all over the episode. Thus, a ore accurate evaluation of the distortions within the coplexes is possible. Actually, the linear segents outside the ventricular beats (see for exaple Fig. and 3) that represent physiological zero-line should be free of any distortions. Obviously, the 'error' there is due to noise coponents of the AHA recordings that have not been totally eliinated by the preliinary oving averaging, since the first lobe of the cob filter [] has an equivalent high-pass cut-off approxiately at 4 Hz. This ipression ay be reinforced by visual inspection of treor episodes after oving averaging followed by soe kind of additional filtering. Consequently, the real errors own to the procedure are considerably saller. One ay speculate that the distortions introduced by the recovery inside the QRS coplexes are within ± 5 μv (seefig.,3,4,5). conditionally clean AHA 5d signal conditionally clean AHA 7d signal error -. error Figure 3 Coparisonbetween'clean'andrestoredAHA 5d episode Figure 5 Coparisonbetween'clean'andrestoredAHA 7d episode. Page 5 of

6 BioMedical Engineering OnLine 8, 7:9 Table : Starting and ending ties of the AHA recordings used for assessent of the recovery procedure. episode taken episode taken episode taken AHA recording starting tie, s ending tie, s AHA recording starting tie, s ending tie, s AHA recording starting tie, s ending tie, s 4d d d d d d d d d 56 d d d d d d d d d d d d 8 5 4d d d d d d d d d d 4 5d d d 4 53d d d d d d d d d d d d d d d d d d d d d 5 8 5d d d 5 8 5d d d d d d d d d d d d d d d d 9 8 7d d d d d d d 5 84 Additional treor suppression in the linear segents Fig. 6 shows Savitzky-Golay frequency responses obtained for 5 Hz sapling rate with different paraeter s. Here the original notation n [8] is substituted by s in order to avoid confusion with the nuber of saples in one PL period. s Yi = CX j i+ j, Cj = 3s + 3s 5j, N= ( s+ ) 4s + 4s 3 / 3 N j= s ( ) Filters with paraeter s < 4 are unusable since their first zero is shifted too far towards the high frequencies that stultify the attepts for treor suppression. In this study s = 5 is used as a coproise between Page 6 of

7 BioMedical Engineering OnLine 8, 7:9 s=7 s=5 s=4. s= [Hz] Figure 6 Savitzky-Golay frequency responses obtained with different paraeter s frequency content of the originally used treor frequency content after oving averaging frequency content after odified Savitzky-Golay filter, s=5 good treor suppression and preserving the P-wave shapes. The expected effect of the additional treor suppression outside the QRS coplexes and soe high T-waves can be seen in Fig. 7 and 8. The first one shows a considerable treor aplitude reduction after the oving averaging and the Savitzky-Golay filter. Therefore, a part of the residual treor in the processed signals deonstrated below, which are taken fro the AHA database, is due to noise coponents in the original recordings. Fig. 8 presents the FFT diagras of the two consecutive filtrations. The observation of the traces in Fig. 7 suggests how to assess the suppression ratio of both procedures. It is quite possible that the axiu peak coupled to a relatively high frequency before filtering is well suppressed after filtering while a lower aplitude lower frequency peak before ay practically preserve its aplitude after that. Therefore, the suppression ratio could be defined as the quotient of the axiu peaks in signals before and after processing. For the oving Figure 8 Frequency content of the treor after oving averaging and Savitzky-Golay filter. [Hz] averaging such ratio is over 6 ties. It becoes about 5 after additional Savitzky-Golay filtering. Results Evaluation of the noise suppression in conditionally clean signals ixed with PL interference and treor Fig. 9 illustrates how the containated signals are obtained. A conditionally clean ECG episode (upper trace) is ixed with treor (second trace) and interference (third trace) to be used further (lower trace) for precise assessing the treor suppression and PL interference cancellation when the three procedures are applied. Noise suppression of the containated AHA 5d episode is presented in Fig.. It is chosen for coparison with Fig. 3, where the sae clean signal is originally used treor treor after oving averaging treor after additional Savitzky-Golay filtering, s= Figure 7 Reduction of the treor aplitudes after oving averaging and Savitzky-Golay filter conditionally clean AHA signal treor power-line interference containated AHA signal Figure 9 Containated episode ('clean' signal + treor+ interference). Page 7 of

8 BioMedical Engineering OnLine 8, 7: processed signal error = processed - clean signals conditionally clean AHA 5d signal extracted treor = containated by treor - clean signals Figure Differences between containated and processed AHA 5d signal. used as input. The traces are as follows: conditionally clean signal; processed signal; error = processed - clean signals; extracted treor = containated by treor - clean signals. The PL interference is totally eliinated [9]. For ore clarity, the containated signal is not shown. The error within the QRS coplexes is the sae as presented in Fig. 3. The treor suppression outside the coplexes is higher due to the additional Savitzky- Golay filtering. The extracted treor is a considerable part of the non-correlated added and residual treors of the clean signal. Again, all clinical inforation (P waves, QRS paraeters, ST segents, T waves) is preserved. This is true also for the other containated and tested AHA database episodes. The next Fig. and deonstrate how the identification arks of soe specific rhyths such atrial and original episode with AF containated by treor episode with AF Figure Treor suppression in episode with AF taken fro MIT-BIH Atrial Fibrillation/Flutter database, 4936 recording, starting at 7 in and 8 s. ventricular fibrillation are preserved (see for exaple the f-wave shapes in Fig. ). Treor suppression in originally noisy recordings The efficiency of the reported ethod and algorith is illustrated below by two originally noisy AHA recordings subjected to the procedures (figures 3 and 4). The two first traces are the original and the processed signals, respectively. The lower traces point out the extracted treor. Fig. 5 and 6 show the result of treor suppression in episodes taken fro the MIT-BIH Noise Stress Database original episode with VF containated by treor episode with VF Figure Treor suppression in episode with VF taken fro AHA 83d recording, starting at 6 s AHA 63d signal started at 744s processed signal extracted treor Figure 3 Noisy suppression in AHA 63d episode. Page 8 of

9 BioMedical Engineering OnLine 8, 7:9 AHA 5d signal started at 64s MIH-BIH Noise Stress Database 8e6d signal - - processed signal processed signal extracted treor. extracted treor Figure 4 Noisy suppression in AHA 5d episode Figure 6 Noisy suppression of 8e6d Noise Stress Database episode, starting at 36 s. Discussion and conclusion The proposed ethod for treor suppression in one- or ultilead ECG is based on oving averaging of the ECG signal followed by a linearly-angular procedure for restoring the affected aplitudes of QRS coplexes and other relatively high and steep ECG waves. Thus, the useful high frequency coponents are preserved in the range specified by the ebedded in the ECG instruent filter, usually up to 5 Hz. Finally, the signal portions outside the QRS coplexes are additionally processed to reduce the treor level by applying a Savitzky-Golay soothing procedure. The results prove the efficiency of the developed ethod. The recovery error of about 5 μv is below the level that ay provoke wrong diagnostic. The interference is totally eliinated. The treor is suppressed approxiately 5 ties. The residual treor does not lead to false ECG interpretation. The procedure - -. MIH-BIH Noise Stress Database 8e4d signal processed signal extracted treor Figure 5 Noisy suppression of 8e4d Noise Stress Database episode, starting at 499 s. efficiency is independent on arrhythia and any other wave shape variations. The algorith is suitable for realtie ipleentation. For the tiebeingtheindividual shape of the restored waves (the coefficient h) is not taken in consideration. This possibility will be further checked up. Copeting interests The authors declare that they have no copeting interests. Authors' contributions GM developed the recovery procedure and deterined the optial order of the Savitzky-Golay filter. ID carried out the other parts of the study. Both authors discussed the ideas accopanying the elaboration of the reported ethod for treor suppression in ECG. They read and approved the final anuscript. References. Huhta JC and Webster JG: 6 Hz interference in electrocardiography. IEEE Trans Bioed Eng 973, :9.. van Metting R, Peper A and Gribergen CA: High-quality recording of bioelectrical events, Part : Interference reduction, theory and practice. Med Biol Eng Coput 99, 8: Ziarani AK and Konrad A: A Nonlinear Adaptive Method of Eliination of Power Line Interference in ECG Signals. IEEE Trans Bioed Eng, 49: Ju-Won Lee and Gun-Ki Lee: Design of an adaptive filter with a dynaic structure for ECG signal processing. Intern J Contr Autoat Syst 5, 3: Pottala EW, Bailey JJ, Horton MR and Gradwohl JR: Suppression of baseline wander in the ECG using a bilinearly transfored, null phase filter. J Electrocardiol 989, Suppl: Frankel RA, Pottala EW, Browser RW and Bailey JJ: Afilterto suppress ECG baseline wander and preserve ST-segent accuracy in a real-tie environent. J Electrocardiol 99, 4 (4): McAdas ET and Jossinet J: Nonlinear transient response of electrode-electrolyte interface. Med Biol Eng Coput, 38 (4): Page 9 of

10 BioMedical Engineering OnLine 8, 7: Luo S and Topkins WJ: Experiental study: Brachial otion artefact reduction in the ECG. Cop Cardiol 995, Bensadoun Y, Novakov E and Raoof K: Multidiensional adaptive ethod for cancellation EMG signals fro the ECG signal. Proceedings of the 7th Annual International Conference on the IEEE Engng in Med and Biol Soc 995: Montreal Roberge FA, Kearney RE 995, Christov II and Daskalov IK: Filtering of electroyogra artifacts fro the electrocardiogra. Med Eng Phys 999, : Nikolaev N and Gotchev A: ECG signal denoising using wavelet doain Wiener filtering. Proceedings of the European Signal Processing Conference EUSIPCO- Tapere, Finland;, Nikolaev N, Gotchev A, Egiazarian K and Nikolov Z: Suppression of electroyogra interference on the electrocardiogra by transfor doain denoising. Med Biol Eng Coput, 39 (6): Dotsinsky I and Stoyanov T: Optiization of bi-directional digital filtering for drift suppression in electrocardiogra signals. J Med Eng Technol 4, 8(4): Thakor NV and Zhu Y: Applications of adaptive filtering to ECG analysis: noise cancellation and arrhythia detection. IEEE Trans Bioed Eng 99, 38: Roanca M and Szabo W: Electrocardiogra pre-processing for the reoval of high frequency and power-line frequency noise. Proceedings of the 6th International Conference on Optiization of Electrical and Electronic Equipents: May 998; Braşov 998, Christov II: Dynaic power-line interference subtraction fro biosignals. JMedEngTechn, 4: Dotsinsky I and Christov I: Power-line interference subtraction fro the electrocardiogra in the presence of electroyogra artifacts. Electrotechnika Elektronika, 37( ):8. 8. Sahabi JS, Tandon SN and Bhatt RKP: Quantitative analysis of errors due to power line interference and base line drift in detection of onsets and offset in ECG using wavelets. Med Biol Eng Coput 997, 35: Levkov C, Mihov G, Ivanov R, Daskalov I, Christov I and Dotsinsky I: Reoval of Power-line Interference fro the ECG: a Review of the Subtraction Procedure. BioMed Eng OnLine 5, 4:5.. Subcoittee on Instruentation Coittee on Electrocardiography Aerican Heart Association, Recoendation for instruents in electrocardiography and vectorcardiography. IEEE Trans Bioed Eng 967, 4: Lynn PA: Online digital filters for biological signals: soe fast designs for a sall coputer. Med Biol Eng Coput 977, 5: Jane R, Rix H, Cainal R and Laguna P: Alignent ethods for averaging of high resolution cardiac signals: A coparative study of perforance. IEEE Trans Bioed Eng 99, 38(6): Laguna P, Jane R, Meste O, Poon PW, Cainal P, Rix H and Thakor NV: Adaptive filter for event-related bioelectric signals using ipulse correlated reference input: Coparison with signal averaging techniques. IEEE Trans Bioed Eng 99, 39: Aströ M, Carro Santos E, Sörno L, Laguna P and Wohlfar B: Vectorcardiographic loop alignent and the easureent of orphologic beat-to-beat variability in noisy signals. IEEE Trans Bioed Eng, 47(4): Kotas M: Projective filtering of tie-aligned ECG beats. IEEE Trans Bioed Eng 4, 5: Clifford GD, Shoeb A, McSharry PE and Janz BA: Model-based filtering, copression and classification of the ECG. Intern J Bioelectroagnetis 5, 7(): Saeni R: A nonlinear Bayesian filtering fraework for ECG denoising. IEEE Trans Bioed Eng 7, 54(): Savitzky A and Golay M: Soothing and differentiation of data by siplified least square procedures. Anal Che 964, 36: Gotchev A, Christov I and Egiazarian K: Denoising the electrocardiogra fro electroyogra artifacts by cobined transfor-doain and dynaic approxiation ethod. proceedings of the International Conference on Acoustics Speech and Signal Processing ICASSP': Orlando USA 3 7 May, Aerican Heart Association (AHA) arrhythia ECG database. Eergency care Research Institute 5 Butler Pike, Plyouth Meeting, PA 946 USA. Publish with BioMed Central and every scientist can read your work free of charge "BioMed Central will be the ost significant developent for disseinating the results of bioedical research in our lifetie." Sir Paul Nurse, Cancer Research UK Your research papers will be: available free of charge to the entire bioedical counity peer reviewed and published iediately upon acceptance cited in PubMed and archived on PubMed Central yours you keep the copyright BioMedcentral Subit your anuscript here: Page of