Reduction of Noises In ECG Signal by Various Filters

Reduction of Noises In ECG Signal by Various Filters

Stalin Subbiap, Dr. Rajkumar Patro2, Dr. K. Rajendran3

1Lecturer, Department of Electronics, Guru Nanak Engineering College, Andhrapradesh, India.

2Associate Professor, Guru Nanak Engineering College, Andhrapradesh, India

3Assistant Professor, Department of Electronics, Government Arts College Ayyarmalai Kulithalai, Tamilnadu, India

Abstract

In this work, different type of filters are proposed for improving the removal of noise in ECG signals. This type of filtering procedure is applied to the Several ECG patient signals. That ECG signals are collected from MIT-BIH database. The ECG signal from MIT- BIH arrhythmia database may contain artifacts, noise and baseline wanders. Therefore it is necessary to denoise the ECG signal to remove all these unwanted parts of the signal. The noise filtering progress is applied to the 30 ECG signals, that the filtering process are Median filter, Finite impulse response filter, Butterworth filter, Gaussian filter. Two type of performance metrics are evaluated for this noise removal technique. The metrics are Mean Square error, Peak Signal to Noise Ratio.

Keywords: ECG, MIT-BIH Database, Median filter, FIR filter, Gaussian filter, Butterworth filter.

1. Introduction:

   The electrocardiogram (ECG) is one of the bio- electric signals and it is used to take the electrical activities of the heart. The signal of the ECG is characterizes by the PQRS Waveform. This type of bioelectrical signal mostly used in clinical side to identified the heart diseases in the Patients. The surface electrodes are used to measure the ECG signal. Some time noises are produced through this surface electrode, that the electrodes are not placed properly in the patient. Some of the steps to remove the artifact basically they are

   • Check the patient skin, carry out and preparing good skin for the patient.

   • Check the electrode gel.

   • In the time of checking the patient should be warm and not to speak and move.

   • Check the chest lead that is placed in proper position or not. If it is not placed properly means it leads to false diagnosis of infarction.

   • Check the Patient cable connection to the ECG device

     These are some of the points that lead to produce noise in the ECG signal.

     By this type of interference some type of noises are present in the ECG signal. These types of noises are baseline wander, Power line interference, motion artifacts, electrode contact, and instrument noise [1]. To remove these type of noises in the ECG signal with the help of novel filters in the time of processing the method of preprocessing.

     In this paper that the preprocessing method provides the removal of noise in the ECG signal and provides the very effective way to detect the heart diseases in the patient. The base line wander noises are produced by the amplitude and frequency variations in the ECG signal [1]. It is observed that the 50% of base line noises are present in the amplitude of the ECG signal. Another most important type on noise, which mostly appeared in ECG signal, is power line interference and it is caused by the 60HZ in the sinusoidal and its harmonics. During signal measurement some type on noises are detected that are electrode motion and muscle contraction [2]. Low-pass filter and high-pass filters are the usual filters to remove the noise in the ECG signal. But these types of filters produce artifacts on the QRS complex [4] and some type of filter banks also produces these artifacts [5, 6]. Some type of filters algorithm together with the ECG signal modelling used to remove the noise reduction in ECG signal [7, 8, and 9]. Extended Kalman filters techniques are used to remove the noise in ECG signal [7]. Some of the techniques such as mean shift algorithm [8] state vectors with time delay [10] and empirical mode decomposition [11] are used for ECG signal noise reduction.

     Modern techniques such as Principle Component Analysis, Neural Network, and Wavelet transform are

     used to remove the noise in the ECG signal, especially it remove the high frequency from their wave it have a disadvantage that their bandwidth is not constant. [12, 13, 14]. Another one method to remove the baseline wanders by using wavelet based cascaded adaptive filter. Here they introduce the energy ratio; if the energy ratio is greater than the given threshold then the noise removal is taken by cubic spline estimation otherwise it must be filtered by discrete Meyer wavelet filter and the cubic spline estimation [15].

     ECG signal are collected from the MIT-BIH

     From this relation, the heat equation also suggests a convenient way to calculate the Gaussian filter. For digital image

     i,j

     un = u ih, jh, n (8)

     Denote the digital sample of a continuous image sequence u x1, x2, t on a Cartesian grid of resolution h with temporal sampling rate. Under the central difference scheme, the forward Euler scheme in the temporal direction leads to the digital Gaussian filtering formula.

     un+1 = un + un + un + un +

     Arrhythmia database. This database includes 48 half-

     i,j

     i+1,j

     n

     i1,j

     i,j+1

     i,j1

     hour recordings of two leads of ECG signals. These

     1 4 ui,j

     (9)

     ECG signals are recorded at a sampling frequency of 360 Hz with 11 bit resolution over a 10mv range [3].

     Several method has developed to remove the noise in the ECG signal depend on filter banks [16-21] they are principal component analysis (PCA), independent component analysis (ICA), neural networks (NNs), adaptive filtering, empirical mode decomposition EMD, Wavelet Transform.

2. Type of filters Median filter:

   This is a filter that makes possible for the elimination of a divergent value by changing the divergent value in a finite series with the medium value in the same series [22]. When it is of two dimensions, the MF for images would be developed as follows

   m k = med w k =

   med xn k , . . , x1 k , x0 k , x1 k , . . , xn(k) (1)

   Where = 2p. For stability reason it is required that

   1

   4

   The key issue arising from diffusion-based

   denoising schemes is the problem of optimal stopping time, that is, to decide when to stop the process to achieve a well-balanced performance of suppressing the noise while retaining the fidelity of the target signal. Without termination at certain finite time T, by the conservation law of adiabatic diffusion, as t

   , u(x, t) will go to the mean of z over the whole image domain. A general way of choosing optimal stopping time usually depends on a cost or risk function, which properly defined to faithfully reflect human or machine visual perception.

   Low-Pass Butterworth Filter:

   Assume an Nth-order transfer function of analog low-pass Butterworth prototype is

   Gaussian filter:

   H s = 1

   (10)

   Gaussian filter is one of the crucial important for

   N

   n =0

   An sn

   both the theory and application compared to other linear filters. In two dimensional this filter is given by

   Through the mapping between the analog and digital domain, its corresponding digital low-pass and high-

   g x =

   1 x 2

   2 e 2 2 . (2)

   pass filters are in the form

   1 + b01Z1 1 + b11Z1 + b12Z2

   2

   In this equation denotes the scale level and it

   control the filter width,

   H Z =

   1 + a01Z1

   1 + a11Z1

   + a12Z2

   .

   So the denoised image

   p>u = g Z. (3)

   1+bK 1Z1+bK 2Z2

   1+aK 1Z1+aK 2Z2

   K = N (11)

   2

   Then the denoising error can be decomposed as

   u u 2 g u u 2 + g n 2, (4)

   Here the controls the tradeoff between removing noise.

   Gaussian filter is remarkably associated with the

   Where b01 = a01 = 0 , if N is an even integer.

   K is the largest integer not greater than N

   2

   For designing digital band-pass and band-pass

   filters, s is quadratic in Z, so the mapping leads to a 2Nth-order transfer function in Z given by

   linear heat diffusion equation:

   ut x, t = 1 x, t = 1 ux x + ux x (5)

   H Z = 1+b01 Z1 +b02 Z1

   1+a01 Z1 +a02 Z1

   (12)

   2 2 1 1 2 2

   1 + b Z1 + b Z2 + b

   Z3 + b

   Z4

   u x, t = Z(x) (6)

   11 12

   13 14 .

   Its solution is

   u x, t = g

   z(x). (7)

   1 + a11 Z1 + a12 Z2 + a13 Z3 + a14 Z4

   1+bK 1Z1+bK 2Z2 +bK 3Z3+bK 4Z4 , K = N (13)

   t 1+aK 1Z1 +aK 2Z2 +aK 3Z3 +aK 4Z4 2

   Whereb01 = b02 = a01 = a02 = 0, if N is an even integer.

   The transfer function of digital filters can be split into a gain factor and a filter transfer function itself as

   H Z = GH (Z) (14)

   Where

   are picked from the MIT-BIH database. Then applied different kind of filters to these signals to remove the different noises are baseline wander, power line interference, muscle artifact noise, and electrode motion artifact noise and compared the filters output by performance metrics are Mean Square Error (MSE),

   H Z = N(Z)

   D(Z)

   Window based FIR Filter:

   (15)

   Peak Signal to Noise Ratio (PSNR).

   Performance metrics:

   Removal of noise in the ECG signal using Median filter, Finite impulse response filter, Gaussian filter,

   In this method start with the desired frequency

   response specification Hd(w) and the corresponding unit sample response hd(n) is determined using inverse Fourier transform. The relation between Hd w and hd(n) is as follows:

   Butterworth filter. For this performance some type of metrics has taken placed in this paper and compared them to show the result of best performance for noise removal in the ECG signal. Compare the type of filters by MSE, PSNR.

   Hd w =

   hd (n)ejwn

   (16)

   Let us take

   i=0

   jwn

   x = double(input image)

   hd n

   = Hd(w)e dw

   (17)

   The impulse response hd(n) obtained from the above equation is of infinite duration. So, it is truncated at some point, say n=M-1 to yield an FIR filter of length M (i.e. 0 to M-1). This truncation of hd(n) to length M-1 is done by multiplying hd(n) with an window. Here the design is explained by considering the rectangular window, define as

   y = double(filtered image) z = abs(x y)

   MSE = mean(mean(z. ^2)))) (21) MSE is denoted as Mean Square Error. That the largest value of MSE means that the ECG signal is poor

   quality

   .

   w n = 1 n = 0,1,2 . M 1

   0 otherwise

   (18)

   PSNR = 20 log10 255 (22)

   MSE

   Thus, the impulse response of the FIR filter becomes

   h n = hd n w(n)

   PSNR is denoted as Peak signal to Noise Ratio. PSNR is opposite to MSE, that if the small value of

   = hd n n = 0,1,2 . M 1

   0 otherwise

   (19)

   PSNR means that the removal of noise in the ECG signal doesnt provide better result.

   Now the multiplication of the widow function w n with hd(n) is equivalent to convolution of Hd w with W w , where W w the frequency domain representation of the window function is

   Table I Comparison of filters

   i=0

   W =

   w(n)ejwn

   (20)

   Thus the convolution of Hd w with W() yields the frequency response of the truncated FIR filter H()

3. Experimental results

   Simulation results are carried from the MIT BIH arrhythmia database to remove the noise in the ECG signal. MIT BIH ECG signals are illustrated by three files they are (.hea) denoted as header file, (.dat) denoted as binary file and (.atr) denoted as binary annotation file. Comprehensive information of ECG signal was present in this header file, that what type of lead used for the patient and number of lead used for diagnosing and their sampling frequency and patient history. In binary file what type of format that the signal is stored. ECG beat information is stored in binary annotation file. For this experimental result consider the ECG signals numbered as 100, 103, 104,

   105, 106, 115 likewise totally 29 patients ECG signals

   Input signal	Filters	MSE	PSNR
   M107ECG	Median	0.0637	72.0461
   	FIR	0.3498	57.2546
   	Gaussian	0.1271	66.0498
   	Butterworth	0.0895	69.0940
   M111ECG	Median	0.0345	77.3629
   	FIR	0.1385	65.3016
   	Gaussian	0.0782	70.2647
   	Butterworth	0.0477	74.5604
   M209ECG	Median	0.0857	69.4673
   	FIR	0.1822	62.9203
   	Gaussian	0.0530	73.6532
   	Butterworth	0.1103	67.2815
   M217ECG	Median	0.0601	72.5511
   	FIR	0.2570	59.9324
   	Gaussian	0.0986	68.2558
   	Butterworth	0.0703	71.1978

   The above table gives the comparison of MSE, PSNR for four types of filters for M107ECG, M111ECG, M209ECG, M217ECG signal from MIT-

   BIH database. From the comparison table it is clearly

   PSNR value (%)

   observed that the median filter and Gaussian filter gives better result compared to other filters in this paper. The better results given by median and Gaussian filters are shown in table V as a block letter.

   100

   80

   60

   40

   20

   0

   Comparsion of PSNR value

   M107ECG M105ECG

   M124ECG

   Various filter

   Figure 1 Comparison of PSNR value

   The comparison of PSNR value for various filters for the M100ECG, M105ECG and M124ECG signal from MIT BIH database is shown in figure 3. From the above graph it is noted that the median filters provides better result compared to other filter such as FIR filter, Gaussian filter, and Butterworth filter.

   1. Input Signal (M100ECGsignal)

   2. Noise reduction by Median filter

   3. Noise Reduction by FIR filter

   4. Noise Reduction by Gaussian filter

   5. Noise Reduction by Butterworth filter Figure 2 Results of removal of noise in M100ECG

   signal using different filters

   Comparison of signal M100ECG have taken and it is shown in above figure. The simulation results of graph have shown in figure 1 shows that the different type of filters is able to reduce the noise in ECG signal. To removal of noise in ECG signal of M100ECG is shown in figure 1, that the input signal is shown in figure 1(a), noise reduction in M100ECG by median filter is shown in figure 1(b), noise reduction by finite impulse response filter is shown in figure 1(c), noise reduction by Gaussian filter is shown in figure 1(d) and noise reduction by butterwort filter is shown in figure 1(e).

4. Conclusion

   EC preprocessing is most important one in their process. In this paper conducted to remove the noise in the ECG signal. First the removal of noise is very helpful to identify the correct ECG signal in medical field. Because in clinical field, ECG signal is mostly used to identify the patient heart diseases. That the noises are obtained by various problems, due to instrument error, electrode contacts, motion artifacts, power line interference. In this paper, ECG signals are collected from MIT-BIH database. Totally 30 ECG signals are obtained from this database. To remove the noise in the ECG signals are done four type of filter they are Median filter, FIR filter, Gaussian filter, Butterworth filter. Then the performances are evaluated by PSNR, MSE. Table I gives comparison between these four filters. Simulation results of median filter approach can efficiently remove the noise in the ECG signal and give best result compared to other approaches. The performance of PSNR for all filters is shown in figure 2. Therefore overall performance of Median filter gives the best performance followed by Gaussian filter.

5. References

1. G. Friesen, T. Jannett, M. Jadallah, S. Yates, S. Quint and H. Nagle, A comparison of the noise sensitivity of nine QRS detection algorithms, IEEE Trans. Biomed. Eng, 37(1) 8598, 1990.

2. N. Kannathal, U. Acharya, P. Joseph, L. Min, J. Suri, Analysis of electrocardiograms, Advances in Cardiac Signal Processing, Springer, pp.5582, 2007.

3. The MIT-BIH Arrhythmia Database. [Online]. Available from: http://physionet.org/physiobank/database/mitdb/ .

4. T. Slonim, M. Slonim and E. Ovsyscher, The use of simple FIR filters for filtering of ECG signals and a new method for post-filter signal reconstruction, Proc

   .Comput. Cardiol, pp. 871873, 1993.

5. V. Afonso, W. Tompkins, T. Nguyen, S. Trautmann, S. Luo, Filter bank-based processing of the stress ECG, in: Proceedings of IEEE 17th Annual Conference on Engineering in Medicine and Biology Society, pp.887 888, 1995.

6. Y. Wu, R. Rangayyan, Y. Zhou, S. Ng, Filtering electrocardiographic signals using an unbiased and normalized adaptive noise reduction system, Med. Eng. Phys.3, pp.1726, 2009.

7. O. Sayadi and M. Shamsollahi, ECG denoising and compression using a modified extended Kalman filter structure, IEEE Trans. Biomed. Eng, pp.22402248, 2008.

8. J. Yan, Y. Lu, J. Liu, X. Wu, Y. Xu, Self-adaptive model-based ECG denoising using features extracted by mean shift algorithm. Biomed. Signal Process Control, pp. 103113, 2010.

9. M. Blanco Velascoa, B. Weng, K. Barner, ECG signal denoising and baseline wander correction based on the empirical mode decomposition, Comput. Biol. Med, pp.113, 2008.

10. H. Liang, Z. Lin and F. Yin, Removal of ECG contamination from diaphragmatic EMG by non linear analysis, Non linear Anal, pp. 745753, 2005.

11. H. Liang, Q. Lin and J. Chen, Application of the empirical mode decomposition to the analysis of esophageal manometric data ingastro esophageal reflux disease, IEEE Trans. Biomed. Eng, pp.16921701. 2005.

12. G. Moody and R. Mark R, QRS morphology representation and noise estimation using the Karhunen Loeve transform, Comput. Cardiol, pp. 269272, 1989.

13. G.D. Clifford and L. Tarassenko, One-pass training of optimal architecture auto-associative neural network for detecting ectopic beats, Electron. Lett, pp.11261127, 2001.

14. P.E. Tikkanen, Non linear wavelet and wavelet packet denoising of electro-cardiogram signal, Biol. Cybern, pp. 259267, 1999.

15. Lisheng Xu, David Zhang, KuanquanWang, Naimin Li, Xiaoyun Wang, Baseline wander correction in pulse waveforms using wavelet-based cascaded adaptive filter, Computers in Biology and Medicine, pp. 716 731, 2007.

16. H. Yan and Y. Li, Electrocardiogram analysis based on the KarhunenLoeve transform, in: Proceedings of the 3rd International Conference on Biomedical Engineering and Informatics (BMEI), vol. 2, 2010, pp. 887890.

17. M. Popescu, P. Cristea, A. Bezerianos, High resolution ECG filtering using adaptive Bayesian wavelet shrinkage, in: Computers in Cardiology, Cleveland, OH,USA, vol. 2, 1998, pp. 401404.

18. V. Almenar and A. Albiol, A new adaptive scheme for ECG enhancement, Signal Processing 75 (1999) 253 263.

19. G. Tang and A. Qin, ECG de-noising based on empirical mode decomposition, in: 9thInternational Conference for Young Computer Scientists, 2008, pp. 903906.

20. M. Alfaouri and K. Daqrouq, ECG signal denoising by wavelet transform thresholding, American Journal of Applied Sciences 5 (3) (2008) 276281.

21. Y. Kopsinis, S. McLaughlin, Development of EMD- based denoising methods inspired by wavelet thresholding, IEEE Transactions on Signal Processing 5 (2009) 13511362.

22. C.S. Lee, Y.H. Kuo, P.T. Yu, Weighted fuzzy mean filters for image processing, Fuzzy Sets Syst. 89 (1997) 157180