ECG SIGNAL DENOISING USING EMPIRICAL MODE DECOMPOSITION
0 likes663 views
The document presents a method for denoising ECG signals corrupted with power line interference using empirical mode decomposition and thresholding. It provides background on sources of power line interference in ECG signals and existing approaches to remove it. The proposed approach decomposes noisy ECG signals into intrinsic mode functions using EMD, then applies various thresholding techniques to the IMFs to remove noise before reconstructing the signal. It tests the method on signals from the MIT-BIH Arrhythmia Database corrupted with 10-50% noise and evaluates performance based on correlation coefficient and SNR improvement. Results show Donoho’s thresholding and hard thresholding achieved the best denoising based on these metrics.
![International Journal of Advances in Management, Technology & Engineering Sciences ISSN : 2249 – 7455
Vol. III, Issue 3 (I), December 2013 1
ECG SIGNAL DENOISING USING EMPIRICAL MODE DECOMPOSITION
Sarang L. Joshi
Dept. of Electronics & Telecommunication
Vishwakarma Institute of Technology ,Pune
Rambabu A. Vatti
Dept. of Electronics & Telecommunication
Vishwakarma Institute of Technology, Pune
Introduction
Electrocardiogram (ECG) records electrical activity of heart. ECG is an important biomedical signal which is used
extensively in diagnosis of heart diseases.ECG is usually corrupted by one or more types of noises which include power
line interference, motion artifact, baseline wander, muscle contraction , electrode contact noise.[1] For accurate
diagnosis a clean (noise free) ECG signal is always required.
Power line interference consists of 60/50 Hz pickup and harmonics and the amplitude is upto 50% of peak-to-peak ECG
amplitude [2]. Some of the common causes of the 50 Hz interferences are [3]:
• stray effect of the alternating current fields due to loops in the cables
• improper grounding of ECG machine or the patient
• disconnected electrode
• Electromagnetic interference from the power lines
• Electrical equipments such as air conditioner, elevators and X-ray units draw heavy power line current, which
induce 60/50 Hz signals in the input circuits of the ECG machine.
Various approaches have been proposed in the past based on Wavelet Theory , Empirical mode decomposition,FIR and
IIR filtering for the removal of PLI.
Amit Nimunkar et. al.[4] proposed an EMD based approach to filter 60 Hz noise from ECG signal. The 60 Hz noise is
removed as the first IMF when the SNR is fairly low whereas a pseudo noise is added at higher frequency when the SNR
is high. The approach is successful in removing power line noise in a single step however some distortion in waveform
is observed at the terminal portion of QRS complex. Also there is slight decrease in the overall magnitude of the QRS
complex. Anil Chacko et. al [5] proposed a denoising technique for ECG signals based on EMD. Spectral Flatness is used
to determine the noisy IMFs which are then filtered using butterworth filters. Manuel B. V. et. al. in [6] proposed a
method based on Empirical Mode Decomposition to remove high frequency noise and baseline wander. The noise
components lie in the first several Intrinsic Mode Functions(IMF). Different IMFs are chosen and processed to
successfully to achieve the denoising. Binwei et. al. [7] proposed an ECG denoising method based on the EMD using an
information preserving partial reconstruction. The denoising is carried out in four steps: delineation and separation of
the QRS complex , preservation of QRS complex, determining noisy IMFs, partial reconstruction . Md. Ashfanoor Kabir
et. al. in [8] proposed a windowing method in the Empirical Mode Decomposition domain. The method preserves the
QRS complex information in the first three high frequency intrinsic mode functions. The noisy signal is enhanced in the
Empirical Mode Decomposition domain and then transformed into the wavelet domain in which an adaptive
thresholding scheme is applied to the wavelet coefficients to preserve the QRS information.
Empirical Mode Decomposition :
Empirical Mode Decomposition (EMD) is a data-driven technique introduced by N.E.Huang et. al.[9] for processing non-
linear and non-stationary data. Traditional data analysis methods, like Fourier and wavelet-based methods, require
some predefined basis functions to represent a signal. The EMD relies on a fully data-driven mechanism that does not
require any priori known basis. It is especially well suited for nonlinear and non-stationary signals, such as biomedical
signals. The EMD decomposes the signal into a sum of Intrinsic Mode Functions (IMFs) using Sifting process.
Intrinsic Mode Function (IMF) satisfies following properties : 1] In the whole data set, the number of extrema and the
number of zero crossings must either equal or differ at most by one; and 2] The IMF is symmetric about the local
mean.
Sifting process:
The first step involves the identification of all the local maxima and minima of input signal x(t). All the local maxima are
connected by a cubic spline curve to form the upper envelope eu(t). In the similar manner, all the local minima are](https://image.slidesharecdn.com/ecgemd-150825062759-lva1-app6892/85/ECG-SIGNAL-DENOISING-USING-EMPIRICAL-MODE-DECOMPOSITION-1-320.jpg)
![ISSN : 2249 – 7455 International Journal of Advances in Management, Technology & Engineering Sciences
2 Vol. III, Issue 3 (I), December 2013
connected by a cubic spline curve to form the lower envelope el(t). The mean of the two envelopes m(t) is calculated
as m(t) = [eu(t) + el(t)]/2 and is subtracted from the signal.
Thus, the first proto IMF h1(t) is obtained as h1(t) = x(t) − m1(t). The said procedure to extract the IMF is known as the
sifting process. Since h1(t) still contains multiple extrema in between zero crossings, the sifting process is performed
again on h1(t). This process is applied repetitively to the proto-IMF hk(t) until the first IMF c1(t), which satisfies the IMF
properties, is obtained. Some stopping criteria are used to terminate the sifting process [9].
The stopping criteria of the sifting process:
The stopping criterion determines the number of sifting steps to produce an IMF. Two different stopping criteria exist :
Sum of the Difference, SD,is defined as :
)()()(
1
0
tRtCtX N
N
n
n
When SD is smaller than a predefined value the sifting process is stopped . SD is generally assigned a value between 0.2
and 0.3. When the SD is smaller than a threshold, the first IMF c1(t) is obtained, and r1(t) = x(t) − c1(t). As the residue
r1(t) still contains some useful information it is treated as a new signal and the above procedure is applied on it to
obtain r1(t) − c2(t) = r2(t), rN−1(t) − cN(t) = rN(t). The whole procedure terminates when the residue rN(t) is either a
constant, a monotonic slope, or a function with only one extremum. Combining the equations yields the EMD of the
original signal :
A second criterion is based on the S-number, which is the number of consecutive siftings when the numbers of
zero-crossings and extrema are equal or at most differ by one. The sifting process will stop only if for S consecutive
times the number of zero-crossings and extrema are equal or at most differ by one. In most cases S is found to be
between 4 and 8 [10].
The result of the EMD produces N IMFs and a residue signal. Lower-order IMFs capture fast oscillation modes while
higher-order IMFs represent slow oscillation modes.
Thresholding
1. Hard Thresholding : The hard thresholding operation is defined as , D(U, λ) = U for all |U|> λ .and D(U, λ) = 0
for all |U|< λ. Hard thresholding is a “keep or kill” procedure .
2. Soft Thresholding :The soft thresholding operation is defined as , D(U, λ) = sgn(U)max(0, |U| - λ) .Soft
thresholding shrinks coefficients above the threshold in absolute value.
Proposed approach:
The dataset used in this study is obtained from physio-Bank entitled “MIT-BIH Arrhythmia Database”[11] available on-
line. The source of the ECGs included in the MIT-BIH Arrhythmia Database is a set of over 4000 long-term Holter
recordings that were obtained by the Beth Israel Hospital Arrhythmia Laboratory. The database contains 48 records
sampled at 360 Hz each of which is slightly over 30 minutes long.In most records, the upper signal is a modified limb
lead II (MLII), obtained by placing the electrodes on the chest. The lower signal is usually a modified lead V1.We select
signal no. 209.dat,213.dat,221.dat,223.dat,231.dat (ML II).Each of the signal is corrupted with 60Hz noisel , the
amplitude of which is varied from 10 percent to 50 percent of the peak to peak ECG amplitude. The corrupted signal is
decomposed into intrinsic mode functions using empirical mode decomposition. Each IMF is then subjected to
donoho’s threshold , global threshold and minimax threshold . Both hard and soft thresholding are performed on each](https://image.slidesharecdn.com/ecgemd-150825062759-lva1-app6892/85/ECG-SIGNAL-DENOISING-USING-EMPIRICAL-MODE-DECOMPOSITION-2-320.jpg)
ECG SIGNAL DENOISING USING EMPIRICAL MODE DECOMPOSITION
- 1. International Journal of Advances in Management, Technology & Engineering Sciences ISSN : 2249 – 7455 Vol. III, Issue 3 (I), December 2013 1 ECG SIGNAL DENOISING USING EMPIRICAL MODE DECOMPOSITION Sarang L. Joshi Dept. of Electronics & Telecommunication Vishwakarma Institute of Technology ,Pune Rambabu A. Vatti Dept. of Electronics & Telecommunication Vishwakarma Institute of Technology, Pune Introduction Electrocardiogram (ECG) records electrical activity of heart. ECG is an important biomedical signal which is used extensively in diagnosis of heart diseases.ECG is usually corrupted by one or more types of noises which include power line interference, motion artifact, baseline wander, muscle contraction , electrode contact noise.[1] For accurate diagnosis a clean (noise free) ECG signal is always required. Power line interference consists of 60/50 Hz pickup and harmonics and the amplitude is upto 50% of peak-to-peak ECG amplitude [2]. Some of the common causes of the 50 Hz interferences are [3]: • stray effect of the alternating current fields due to loops in the cables • improper grounding of ECG machine or the patient • disconnected electrode • Electromagnetic interference from the power lines • Electrical equipments such as air conditioner, elevators and X-ray units draw heavy power line current, which induce 60/50 Hz signals in the input circuits of the ECG machine. Various approaches have been proposed in the past based on Wavelet Theory , Empirical mode decomposition,FIR and IIR filtering for the removal of PLI. Amit Nimunkar et. al.[4] proposed an EMD based approach to filter 60 Hz noise from ECG signal. The 60 Hz noise is removed as the first IMF when the SNR is fairly low whereas a pseudo noise is added at higher frequency when the SNR is high. The approach is successful in removing power line noise in a single step however some distortion in waveform is observed at the terminal portion of QRS complex. Also there is slight decrease in the overall magnitude of the QRS complex. Anil Chacko et. al [5] proposed a denoising technique for ECG signals based on EMD. Spectral Flatness is used to determine the noisy IMFs which are then filtered using butterworth filters. Manuel B. V. et. al. in [6] proposed a method based on Empirical Mode Decomposition to remove high frequency noise and baseline wander. The noise components lie in the first several Intrinsic Mode Functions(IMF). Different IMFs are chosen and processed to successfully to achieve the denoising. Binwei et. al. [7] proposed an ECG denoising method based on the EMD using an information preserving partial reconstruction. The denoising is carried out in four steps: delineation and separation of the QRS complex , preservation of QRS complex, determining noisy IMFs, partial reconstruction . Md. Ashfanoor Kabir et. al. in [8] proposed a windowing method in the Empirical Mode Decomposition domain. The method preserves the QRS complex information in the first three high frequency intrinsic mode functions. The noisy signal is enhanced in the Empirical Mode Decomposition domain and then transformed into the wavelet domain in which an adaptive thresholding scheme is applied to the wavelet coefficients to preserve the QRS information. Empirical Mode Decomposition : Empirical Mode Decomposition (EMD) is a data-driven technique introduced by N.E.Huang et. al.[9] for processing non- linear and non-stationary data. Traditional data analysis methods, like Fourier and wavelet-based methods, require some predefined basis functions to represent a signal. The EMD relies on a fully data-driven mechanism that does not require any priori known basis. It is especially well suited for nonlinear and non-stationary signals, such as biomedical signals. The EMD decomposes the signal into a sum of Intrinsic Mode Functions (IMFs) using Sifting process. Intrinsic Mode Function (IMF) satisfies following properties : 1] In the whole data set, the number of extrema and the number of zero crossings must either equal or differ at most by one; and 2] The IMF is symmetric about the local mean. Sifting process: The first step involves the identification of all the local maxima and minima of input signal x(t). All the local maxima are connected by a cubic spline curve to form the upper envelope eu(t). In the similar manner, all the local minima are
- 2. ISSN : 2249 – 7455 International Journal of Advances in Management, Technology & Engineering Sciences 2 Vol. III, Issue 3 (I), December 2013 connected by a cubic spline curve to form the lower envelope el(t). The mean of the two envelopes m(t) is calculated as m(t) = [eu(t) + el(t)]/2 and is subtracted from the signal. Thus, the first proto IMF h1(t) is obtained as h1(t) = x(t) − m1(t). The said procedure to extract the IMF is known as the sifting process. Since h1(t) still contains multiple extrema in between zero crossings, the sifting process is performed again on h1(t). This process is applied repetitively to the proto-IMF hk(t) until the first IMF c1(t), which satisfies the IMF properties, is obtained. Some stopping criteria are used to terminate the sifting process [9]. The stopping criteria of the sifting process: The stopping criterion determines the number of sifting steps to produce an IMF. Two different stopping criteria exist : Sum of the Difference, SD,is defined as : )()()( 1 0 tRtCtX N N n n When SD is smaller than a predefined value the sifting process is stopped . SD is generally assigned a value between 0.2 and 0.3. When the SD is smaller than a threshold, the first IMF c1(t) is obtained, and r1(t) = x(t) − c1(t). As the residue r1(t) still contains some useful information it is treated as a new signal and the above procedure is applied on it to obtain r1(t) − c2(t) = r2(t), rN−1(t) − cN(t) = rN(t). The whole procedure terminates when the residue rN(t) is either a constant, a monotonic slope, or a function with only one extremum. Combining the equations yields the EMD of the original signal : A second criterion is based on the S-number, which is the number of consecutive siftings when the numbers of zero-crossings and extrema are equal or at most differ by one. The sifting process will stop only if for S consecutive times the number of zero-crossings and extrema are equal or at most differ by one. In most cases S is found to be between 4 and 8 [10]. The result of the EMD produces N IMFs and a residue signal. Lower-order IMFs capture fast oscillation modes while higher-order IMFs represent slow oscillation modes. Thresholding 1. Hard Thresholding : The hard thresholding operation is defined as , D(U, λ) = U for all |U|> λ .and D(U, λ) = 0 for all |U|< λ. Hard thresholding is a “keep or kill” procedure . 2. Soft Thresholding :The soft thresholding operation is defined as , D(U, λ) = sgn(U)max(0, |U| - λ) .Soft thresholding shrinks coefficients above the threshold in absolute value. Proposed approach: The dataset used in this study is obtained from physio-Bank entitled “MIT-BIH Arrhythmia Database”[11] available on- line. The source of the ECGs included in the MIT-BIH Arrhythmia Database is a set of over 4000 long-term Holter recordings that were obtained by the Beth Israel Hospital Arrhythmia Laboratory. The database contains 48 records sampled at 360 Hz each of which is slightly over 30 minutes long.In most records, the upper signal is a modified limb lead II (MLII), obtained by placing the electrodes on the chest. The lower signal is usually a modified lead V1.We select signal no. 209.dat,213.dat,221.dat,223.dat,231.dat (ML II).Each of the signal is corrupted with 60Hz noisel , the amplitude of which is varied from 10 percent to 50 percent of the peak to peak ECG amplitude. The corrupted signal is decomposed into intrinsic mode functions using empirical mode decomposition. Each IMF is then subjected to donoho’s threshold , global threshold and minimax threshold . Both hard and soft thresholding are performed on each
- 3. International Journal of Advances in Management, Technology & Engineering Sciences ISSN : 2249 – 7455 Vol. III, Issue 3 (I), December 2013 3 of the IMFs. The signal is reconstructed excluding the first IMF. The performance is compared using correlation coefficient, SNR improvement and Root Mean Square Error (RMSE). The simulation is performed in MATLAB 7.10.0. Fig. 1 : Flowchart of proposed approach Results Fig.2: Denoising of Signal 213.dat Table 1: Results showing SNR improvement of output signals obtained for Signal 213.dat Noise amplitude(%) 10 20 30 40 50 Noise power 2.7004x10^3 1.0802x10^4 2.4304x10^4 4.3207x10^4 6.7511x10^4 Corr coef of corrupted signal 0.9985 0.9940 0.9867 0.9767 0.9644 Correlation coef of denoised signal 1 1 1 1 1 Donoho hard thresholding 28.9954 34.7059 38.2298 40.6980 42.6704 Donoho soft thresholding 28.5835 28.2636 31.7858 34.2787 36.2234 Minimax hard thresholding 27.6839 33.2165 37.0399 39.5155 41.4796
- 4. ISSN : 2249 – 7455 International Journal of Advances in Management, Technology & Engineering Sciences 4 Vol. III, Issue 3 (I), December 2013 Minimax soft thresholding 18.8264 25.3704 28.8926 31.3865 33.3301 Global hard thresh 27.4654 33.2771 36.8005 39.2772 41.2400 Global soft thresh 18.3046 24.8595 28.3817 30.8761 32.8192 Table 2 : Results obtained for Signal 213.dat Noise amplitude(%) 10 20 30 40 50 RMSE RMSE RMSE RMSE RMSE Donoho hard thresholding 1.8448 1.9119 1.9114 1.9181 1.9106 Donoho soft thresholding 1.9344 4.0139 1.0137 4.0164 4.0134 Minimax hard thresholding 2.1455 2.1924 2.1920 2.1979 2.1913 Minimax soft thresholding 5.9484 5.6005 5.6003 5.6034 5.5999 Global hard thresholding 2.1167 2.2537 2.2553 2.2590 2.2526 Global soft thresholding 6.3167 5.9398 5.9396 5.9426 5.9392 Conclusion In this paper, a hybrid approach based on EMD and thresholding to remove power line interference from corrupted ECG signal is presented. The results confirms the success of the proposed method. Though both hard and soft thresholding are able to remove the noise successfully we observe that better results are obtained with Donoho’s Threshold and hard thresholding. References: 1. Sarang L. Joshi ,Rambabu A. vatti , Rupali V. Tornekar.” A Survey on ECG Signal Denoising Techniques. “ Proceedings of 3rd International Conference on Communication Systems and Network Technology,April 2013.Vol. no. pp 60,64. 2. B. Pradeep Kumar,S. Balambigai, Dr. R. Aokan. “ECG denoising based on hybrid technique”.ICAESM-2012 3. Garg, Girisha, Shorya Gupta ,Vijander Singh , J.R.P. Gupta and A.P.Mittal. "Identification of optimal wavelet- based algorithm for removal of power line interferences in ECG signals." Power Electronics (IICPE), 2010 India International Conference on. IEEE, 2011. 4. Amit J.Nimunkar,Willis J. Tompkins.” EMD-based 60-Hz noise filtering of the ECG ”.Proceedings of the 29th Annual International Conference of the IEEE EMBS Cite Internationale, Lyon, France. 5. Anil Chacko , Samit Ari. International Conference On Advances In Engineering, Science And Management (lCAESM -2012) 6. Manuel Blanco-Velasco, Binwei Weng, Kenneth E. Barner.” ECG signal denoising and baseline wander correction based on the empirical mode decomposition”. Computers in Biology and Medicine 38 (2008) 1 – 13 7. Binwei Weng,Manuel Blanco-Valasco and Kenneth E. Barner. ”ECG Denoising Based on the Empirical Mode Decomposition “. Proceedings of the 28th IEEE EMBS Annual International Conference ,New York 2006. 8. Md. Ashfanoor Kabir , Celia Shahnaz.” ECG Signal Denoising Method Based on Enhancement Algorithms in EMD and Wavelet Domains.” 978-1-4577-0255-6/11.IEEE-2011. 9. N.E. Huang et al. "The empirical mode decomposition and the Hilbert spectrum for nonlinear and non- stationary time series analysis." Proceedings of the Royal Society of London. Series A: Mathematical, Physical and Engineering Sciences 454.1971 (1998): 903-995. 10. Ayenu-Prah, Albert, and Nii Attoh-Okine. "A criterion for selecting relevant intrinsic mode functions in empirical mode decomposition." Advances in Adaptive Data Analysis 2.01 (2010): 1-24. 11. MITBIH Arrhythmia database www.physionet.org/physiobank/database/mitb