Analysis of ECG Signal Denoising Algorithms in DWT and ... · Analysis of ECG Signal Denoising...
Transcript of Analysis of ECG Signal Denoising Algorithms in DWT and ... · Analysis of ECG Signal Denoising...
Analysis of ECG Signal Denoising Algorithms in
DWT and EEMD Domains
Neelam Bhardwaj1, Sanjeev Nara
2, Sunita Malik
1, and Geeta Singh
2
1Department of Electronics and Communication, Deenbandhu Chhotu Ram University of Science & Technology,
Murthal, Haryana, India 2Department of Biomedical Engineering, Deenbandhu Chhotu Ram University of Science & Technology, Murthal,
Haryana, India
Email: {neelam.2149, sanjeevnara91}@gmail.com, {sunitamalik.ece, geetasingh.bme}@dcrustm.org
Abstract—Accurate analysis of ECG signals becomes
difficult when a lot of noise such as AC (Power line)
Interference, Electromyogram (EMG), Baseline wandering,
channel noise, electrode motion, motion artifact, Gaussian
noise & high frequency noise based on the frequency
variation are present in the ECG signal. Thus, for better
analysis and characterization of ECG, noise removal
becomes an essential part. Denoising of ECG signals plays a
very important role in diagnosis and detection of various
cardiovascular diseases. The various methods available for
denoising of ECG signals include linear filtering, Empirical
Mode Decomposition (EMD), Independent and Principal
Component Analysis, Neural networks, adaptive filtering etc.
In recent studies by several researchers compared to the
above mentioned denoising methods Discrete Wavelet
Transform (DWT) and Ensemble Empirical Mode
reducing noise from ECG signal. This paper presents the
performance analysis on ECG denoising algorithms in
EEMD and wavelet domains by evaluating Signal to Noise
Ratio (SNR) and Root Mean Square Error (RMSE) in order
to compare the effectiveness of these two methods in
reducing the noise.
Index Terms—EEMD, DWT, ECG denoising, SNR, RMSE
I. INTRODUCTION
Electrocardiography (ECG or EKG) is the electrical
activity of the heart externally recorded with the help of
skin surface electrodes. It is a non invasive recording
produced by an electrocardiographic device. Monitoring
of these signals has great clinical significance as they
characteristically precede the normal mechanical function
[1]. It provides valuable information about a wide range
of cardiac disorders such as presence of inactive part or
enlargement of heart muscle. Artifacts in ECG signal that
are extremely common include poor contact between skin
and electrode, electrical interference by outside sources,
electrical noise from the body and machine malfunction.
Removal of these artifacts is necessary to prevent false
interpretation of a heart's rhythm. Filtration of ECG
Signal is also important in order to extract important
features that are used for recognizing variability in heart’s
activity. Clean ECG signal provide detailed information
Manuscript received January 1, 2015; revised December 8, 2015.
about the electrophysiology of the heart diseases and
ischemic changes that may occur [2], [4]. It acts as
decision tool for clinical decision-making.
In order to extract efficient morphology of ECG
signals conventional filtering, Empirical Mode
Decomposition (EMD), Independent and Principal
Component Analysis, Neural networks, adaptive filtering
etc. are methods usually used for ECG denoising. Low
pass, high pass and band pass filtering are conventional
filtering techniques that are used to reduce undesired
artifacts only when noise spectrum is known. These
techniques are inefficient when the noise spectrum is
random in nature [2]. Linear filtering introduces the
ringing effect on the ECG signal analysis. In order to
rectify this limitation cubic spline filter is used for noise
removal from ECG signals [3]. Adaptive filtering well
known for its faster filter response is used for removal of
power line interferences but filtering it requires a
reference signal to be recorded together with the ECG [4].
Independent Component Analysis (ICA) is also employed
for removing the noises from physiological signals but it
does not allow the prior information about the signals for
efficient filtering [5]. Empirical Mode Decomposition
(EMD) is a versatile time–frequency data analysis
method for extracting signals from data synthesized in
noisy nonlinear and non-stationary processes. The
frequent appearance of mode mixing defined as a single
Intrinsic Mode Function (IMF) that either consists of
signals of widely disparate scales or a signal of a similar
scale residing in different IMF components is a major
drawback of this method [6].
Comparatively, the denoising methods based on
EEMD and that based on wavelet are considered to be
more effective and versatile in reducing noise from the
ECG signals. These methods are used to resolve the
limitations that are presented during efficient noise
removal from ECG signals using above mentioned
filtering methods. Ensemble Empirical mode
decomposition is a data analysis method which represents
substantial improvement over the EMD method. In it, by
using the statistical characteristics of white noise, the
ECG denoising is done. In wavelet transform, a signal is
analyzed as a linear combination of the sum of the
product of the wavelet coefficients and mother wavelet [7]
International Journal of Signal Processing Systems Vol. 4, No. 5, October 2016
©2016 Int. J. Sig. Process. Syst. 442doi: 10.18178/ijsps.4.5.442-445
Decomposition (EEMD) are found to be more effective in
and does not introduce any artificial information to the
original signal. However, selection of appropriate wavelet
function and thresholding method plays an important role
in signal denoising. In this paper, a comparison is carried out between
DWT and EEMD methods of ECG denoising in order to
determine their effectiveness reducing the noise. The
performance of methods under evaluation has been
verified by calculating the signal to noise ratio and root
mean square error.
II. METHODOLOGY
A. Participants
7 students were recruited as participants with 20-24
years of age. No exclusion criteria were used. Each
participant was provided written informed consent before
participating.
B. Instrumentation
For ECG data acquisition MP 100 system of BIOPAC
Inc. was used by utilizing AcqKnowledge 4.1.1 software
for data recording and storage. Disposable Electrodes
were placed on the subject's body to measure the
Electrocardiogram (ECG). Specifically Lead II
configuration was used in measuring ECG because it
closely resembles the normal pathway of current of flow
in the heart and therefore shows an upright complex with
an ideal signal [8]. Normal ECG measurements were then
obtained from each participant.
C. Processing of Signals
In order to carry out the comparison, processing of
signals was done through two methods - Discrete
Wavelet Transform and Ensemble Empirical mode
decomposition as described earlier:
1) Discrete wavelet transform
A wavelet transform is mathematical function used to
decompose a signal and provide us with time frequency
representation of signal. This time-frequency
representation is necessary so to be able to analyze
frequency components of signal as well as times at which
they occur. Discrete Wavelet Transform (DWT)
expresses a signal as a combination of sum of the product
of the wavelet coefficients and mother wavelet [7].
A family of mother wavelet is available having energy
spectrum as that of ECG signal. DWT decomposes a
signal into approximate and detail coefficients that helps
to scrutinize the signal at different frequency bands with
different resolutions. ECG signal denoising using DWT
consists of three steps as shown in Fig. 1.
Figure 1. Flowchart of DWT algorithm
The performance of ECG signal denoising using
Wavelet transform depends on the estimation of threshold
value, γ. This threshold value is proposed by Donoho and
Johnstone and is given as in (1).
= 2 logγ N (1)
where φ is standard deviation of detailed DWT
coefficients of wavelet level and N is the length of vector
of DWT coefficients [6]. After determination of γ,
thresholding can be divided into two categories-soft and
hard thresholding. In hard thresholding, signal values
smaller than γ are turned to zero while in soft
thresholding, signal values smaller than γ are turned to
zero and γ is subtracted from signal values greater than γ
that results in no discontinuity of the signal [3], [7]. The
implementation of DWT through MATLAB (2013a) is
shown in Fig. 2.
Figure 2. Implementation of DWT in MATLAB
2) Ensemble empirical mode decomposition
This method introduced by Huang et al. successfully
eliminates the mode mixing phenomenon and nature of
signals can be accurately reflected. EEMD is an adaptive
technique that is different from all previous methods of
data analysis as it is based on the decomposition of the
original data.
In this method, data are collected by separate
observations, each of which contains different noise in
order to improve the accuracy of measurements [9]. The
white noise added is treated as the possible random noise
that can be encountered in the measurement process and
helps in avoiding the mode mixing. The principle behind
this is very simple- the added white noise would populate
the whole time-frequency space uniformly with the
constituting components of different scales. When signal
is added to this uniformly distributed white background,
the bits of signal of different scales are automatically
projected onto proper scales of reference established by
the white noise in the background [10]. As a result, each
individual trial may produce very noisy outputs, for each
of the noise-added decompositions consists of the signal
and the added white noise. Since the noise in each trial is
different in separate trials, it is canceled out in the
ensemble mean [10], [11].
International Journal of Signal Processing Systems Vol. 4, No. 5, October 2016
©2016 Int. J. Sig. Process. Syst. 443
The EEMD algorithm used for enhancement of ECG
signals and reduction of noise is shown in Fig. 3.
Figure 3. Flow chart of EEMD algorithm
This algorithm was implemented with the help of code
developed in MATLAB and the series of output are
shown in Fig. 4.
0 500 1000 1500 2000 2500 3000 3500 4000 4500
-0.5
0
0.5
1Original Signal
0 500 1000 1500 2000 2500 3000 3500 4000 4500-0.4
-0.2
0
0.2
0.4Residual Signal
0 500 1000 1500 2000 2500 3000 3500 4000 4500-1
-0.5
0
0.5
1received Signal
0 500 1000 1500 2000 2500 3000 3500 4000 4500-0.5
0
0.5
1Original Signal
Figure 4. Implementation of EEMD in MATLAB
III. RESULT
The main aim of this paper was to evaluate the
effectiveness of DWT and EEMD in signal enhancement.
The performance of the above two methods is evaluated
based on the Signal to Noise Ratio (SNR) and Root Mean
Square Error (RMSE). The SNR can be defined as in (2).
(2)
where x(t) is the signal and n(t) is the noise. RMSE is
defined as in (3).
(3)
where the numerator part is the square error, x ̂(t) is the
reconstructed ECG signal and M is the length of signal.
Figure 5. Comparison of SNR values
Fig. 5 shows the difference between SNR values of
DWT and EEMD denoising methods of ECG of different
subjects. This figure implies that the higher SNR is
Original Signal
Successive IMFs
International Journal of Signal Processing Systems Vol. 4, No. 5, October 2016
©2016 Int. J. Sig. Process. Syst. 444
obtained by using DWT denoising method for all the
ECG signals.
Fig. 6 represents the comparison of the RMSE results
obtained for the same group of ECG signals as used for
comparison of SNR values.
Figure 6. Comparison of RMSE values
It is vivid from Fig. 5 and Fig. 6 that for a particular
ECG signal, wavelet method yields the smallest RMSE
and higher SNR value that confirms its ability to provide
enhanced ECG signal with better quality.
IV. CONCLUSION
Many researchers have already revealed that DWT and
EEMD and DWT based denoising methods are more
effective in reduction of noise from ECG signals as
compared to conventional signal enhancing algorithms on
basis of their higher SNR and lower RMSE values. This
paper analyzed both denoising schemes and verifies their
effectiveness in noise reduction. The comparison carried
out between these two algorithms using SNR and RMSE
values reveal that ECG signal enhancement using wavelet
domain is far more effective than that of EEMD.
ACKNOWLEDGMENT
We are extremely grateful to the Department of
Biomedical Engineering, DCRUST, Murthal, Haryana for
allowing us to acquire signals in their lab.
Acknowledgement would be incomplete without the
mention of our student volunteers who enthusiastically
supported us by giving their ECG signals.
REFERENCES
[1] P. Karthikeyan, M. Murugappan, and S. Yaacob, “ECG signal
denoising using wavelet thresholding techniques in human stress
assessment,” International Journal on Electrical Engineering and Informatics, vol. 4, no. 2, pp. 306-319, July 2012.
[2] B. Weng and M. Blanco-Velasco, “ECG denoising based on the
empirical mode decomposition,” in Proc. 28th Annual International Conference of the IEEE Engineering in Medicine
and Biology Society, 2006, pp. 1-4.
[3] E. S. El-Dahshan, “Genetic algorithm and wavelet hybrid scheme for ECG signal denoising,” Telecommunication Systems, vol. 46,
pp. 209-215, 2010.
[4] Md. A. Kabir and C. Shahnaz, “An ECG signal denoising method based on enhancement algorithms in EMD and wavelet domains,”
IJRRAS, vol. 11, no. 3, pp. 500-516, June 2012.
[5] T. He, G. Clifford, and L. Tarassenko, “Application of ICA in removing artefacts from the ECG,” Neural Processing Letters, pp.
105-116, 2006.
[6] Z. Zhao, Y. Luo, and Q. Lu, “Adaptive noise removal of ECG signal based on ensemble empirical mode decomposition,” in
Adaptive Filtering Applications, L. Garcia, Ed., InTech, June 2011,
pp. 123-140. [7] A.-R. Al-Quawasmi and K. Daqrouo, “ECG signal enhancement
using wavelet transform,” WSEAS Transactions on Biology and
Biomedicine, vol. 7, no. 2, pp. 62-72, April 2010. [8] Task Force of the European Society of Cardiology and the North
American Society of Pacing and Electrophysiology, “Heart rate
variability: Standards of measurement, physiological interpretation, and clinical use,” European Heart Journal, vol. 17, pp. 354-381,
1996.
[9] N. E. Huang, et al., “The empirical mode composition and the Hilbert spectrum for nonlinear and non-stationary time series
analysis,” Proceeding of R. Soc. Lond. A, vol. 454, pp. 903-995,
1998. [10] Z. Wu and N. E. Huang, “Ensemble empirical mode
decomposition: A noise-assisted data analysis method,” Advances
in Adaptive Data Analysis, vol. 1, no. 1, pp. 1-41, 2009. [11] K. M. Chang, “Ensemble empirical mode decomposition for high
frequency ECG noise reduction,” Biomedical Engineering, vol. 55,
pp. 193-201, August 2010.
Neelam Bhardwaj is a post graduate student in Electronics &
Communication at Deenbandhu Chhotu Ram University of Science and
Technology, Murthal, Sonepat, Haryana. She completed her Bachelors in Biomedical Engineering from Maharshi Dayanand University,
Rohtak in 2009. Her areas of interest are Biomedical Instrumentation
and Signal processing Email: [email protected]
Sanjeev Nara is a post graduate student in Biomedical Engineering at
Deenbandhu Chhotu Ram University of Science and Technology, Murthal, Sonepat, Haryana. He completed his Bachelors in Biomedical
Engineering from Maharshi Dayanand University, Rohtak in 2012. His
areas of interest are Medical Image processing, Biosignal processing and Medical Imaging.
Email: [email protected]
Ms. Sunita Malik is employed as Assistant Professor in Electronics &
Communication Engineering Department, Deenbandhu Chhotu Ram University of Science and Technology, Murthal (Sonepat), Haryana,
India. Her research area includes signal processing, VLSI, embedded
systems, and microcontrollers. She has contributed research papers in refereed national and international journals.
Email: [email protected]
Dr. Geeta Singh is employed as Assistant Professor in Biomedical Engineering Department, Deenbandhu Chhotu Ram University of
Science and Technology, Murthal (Sonepat), Haryana, India. Her
research area includes Biomaterials and Polymers for drug delivery systems. She has contributed research paper in refereed national and
international journals.
Email: [email protected]
International Journal of Signal Processing Systems Vol. 4, No. 5, October 2016
©2016 Int. J. Sig. Process. Syst. 445