chapter 4 wavelet transform-genetic algorithm...
TRANSCRIPT
102
CHAPTER 4
WAVELET TRANSFORM-GENETIC ALGORITHM
DENOISING TECHNIQUE
4.1 INTRODUCTION
This chapter introduces an effective combination of genetic
algorithm and wavelet transform scheme for the denoising of
electrocardiogram (ECG) signals, corrupted by non-stationary noises, using
genetic algorithm (GA) and wavelet transform (WT). The wavelet theory
denoising has been widely exploited in the noisy ECG filtering. Several
wavelet denoising ECG signal algorithms were developed, each exploring a
particular parameter; the wavelet function, threshold calculus and level
decomposition.
Xiao-Ping Zhang et al (1998) proposed a new adaptive denoising
method based on stein’s unbiased risk estimate (SURE) and on a new class of
thresholding functions. Unlike the standard soft-thresholding function, these
functions have continuous derivatives. The new thresholding functions do the
similar manipulations as the standard soft thresholding function and they
make it possible to search for optimal thresholds using gradient based
adaptive algorithms. This method is very effective in adaptively finding the
optimal solution in mean square error (MSE) than that of conventional
wavelet shrinkage methods.
An effective technique for the denoising of electrocardiogram
signals corrupted by non stationary noises (Ercelebi 2004) is based on a
103
second generation wavelet transform and level-dependent threshold estimator.
Here, the wavelet coefficients of ECG signals were obtained with lifting-
based wavelet filters. A lifting scheme is used to construct second-generation
wavelets and is an alternative and faster algorithm for a classical wavelet
transform. Numerical results comparing the performance of this method with that
of the nonlinear filtering techniques (median filter) demonstrate consistently
superior denoising performance of this method over median filtering.
Kania et al (2007) were investigated the application of wavelet
denoising in noise reduction of multichannel high resolution ECG signals. In
particular, the influences of the selection of wavelet function and the choice of
decomposition level on efficiency of denoising process were considered and
whole procedures of noise reduction were implemented in the Matlab
environment. The Fast Wavelet Transform was used. The advantage of used
denoising method is that the noise level decreasing in ECG signals, in which
noise reduction occurs by averaging and has limited application.
Manikandan and Dandapat (2007) proposed a novel Wavelet
Energy based diagnostic distortion (WEDD) measure to assess the
reconstructed signal quality for ECG compression algorithms. WEDD is
evaluated from the Wavelet coefficients of the original and the reconstructed
ECG signals. For each ECG segment, a Wavelet energy weight vector is
computed via five-level biorthogonal discrete wavelet transform (DWT).
WEDD provides a better prediction accuracy and exhibits a statistically better
monotonic relationship with the Mean Opinion Score(MOS) ratings than
wavelet based weighted percentage root mean square difference measure
(WWPRD), PRD and other objective measures.
Prasad et al (2008) proposed a shrinkage method based on a New
Thresholding filter for denoising of biological signals The efficacy of this
filter is evaluated by applying this filter for denoising of ECG signals
104
contaminated with additive Gaussian noise. The performance of this filter is
compared with that of hard and soft thresholding filters using Mean Square
Error and Signal to Noise ratio (SNR). The New Thresholding filter is
significantly more efficient than Hard and Soft filters in denoising the signals.
It embodies the features of both Hard and Soft filters.
Alfaouri and Daqrouq (2008) proposed a new approach based on
the threshold value of ECG signal determination using Wavelet Transform
coefficients. Electrocardiography has had a profound influence on the practice
of medicine. The ECG signal allows for the analysis of anatomic and
physiologic aspects of the whole cardiac muscle. This method is compared
with Donoho's method for signal denoising where in better results are
obtained for ECG signals by this algorithm.
Sumithra and Thanuskodi (2009) proposed a new thresholding
algorithm called trimmed thresholding algorithm. However, the soft
thresholding is best in reducing noise but worst in preserving edges and hard
thresholding is best in preserving edges but worst in de-noising. Motivated by
finding a more general case that incorporates the soft and hard thresholding to
achieve a compromise between the two methods, the trimmed thresholding
method is proposed to enhance the speech from background noise.
Umamaheswara Reddy et al (2009) proposed a new thresholding
technique for denoising of ECG signal. This new de-noising method called as
improved thresholding de-noising method could be regarded as a
compromising between hard- and soft-thresholding de-noising methods. The
advantage of the improved thresholding de-noising method is that it retains
both the geometrical characteristics of the original ECG signal and variations
in the amplitudes of various ECG waveforms effectively.
Mahesh et al (2010) proposed a wavelet denoising algorithm. This
method implemented Haar and Daubechies wavelets are on speech signals
105
and performance is evaluated. Haar wavelet is not suitable for speech
denoising application. As Haar is not smooth when compared to other
wavelets, it has limitations when applied to non stationary signal such as
speech. Higher order Daubechies can be used and are found to be suitable for
the work done. Also soft thresholding is better than hard thresholding.
Sayed and Ei-Dahshan (2010) proposed an effective hybrid scheme
for the denoising of electrocardiogram signals corrupted by non-stationary
noises using genetic algorithm (GA) and wavelet transform. Selection of a
suitable wavelet denoising parameter is critical for the success of ECG signal
filtration in wavelet domain. Therefore, in this noise elimination method, the
genetic algorithm has been used to select the optimal wavelet denoising
parameters which lead to maximize the filtration performance.
Efficient selection of wavelet denoising parameters, such as
wavelet function, threshold function (method), and threshold selection rules
are critical to the success of signal denoising. Usually, these parameters are
selected empirically; which leads to low noise elimination performance. So
the contribution of this work is to introduce an evolutionary optimization
method based on the Genetic Algorithm to search the wavelet denoising
parameters in order to obtain the optimal ECG signal filtration efficiency. The
efficiency performance of our scheme is evaluated using percentage root
mean square difference and signal to noise ratio.
4.2 APPLICATIONS OF WAVELET TRANSFORM IN ECG
SIGNAL
The wavelet transform is a powerful and promising method for time
and frequency signal analysis. A signal is decomposed into building blocks
that are well represented in time and frequency. In the search for significant
features of the ECG signal, it is filtered using wavelet filtering based on the
wavelet transform.
106
While the set of decomposition functions of the Fourier transform
are the functions of sin(k 0t) and cos(k 0t) only, the set of decomposition
functions of the wavelet transform are wider and different sets of
decomposition functions are used. Virtually all wavelet systems have these
very general characteristics. Where the Fourier transform maps a one-
dimensional signal to a one-dimensional sequence of coefficients, the wavelet
transform maps it into two dimensional arrays of coefficients. This allows
localizing the signal in both time and frequency. The concept of the wavelet
transform is usually introduced by the resolution concept to define the effect
of changing scale.
The application of wavelet noise suppression requires the selection
of different parameters. The wavelet noise reduction performance of the ECG
signal is conditioned by three processing parameters named “wavelet
denoising parameters”, namely
Type of wavelet basis function ,
Thresholding function ,
Threshold selection rules ,
4.3 SELECTION OF THE WAVELET
This is the most interesting question for most of the users. The
wavelet has one or two parameters. Because wavelets have so many
constraints, that are not associated with the signal, but more with math and
calculation limitations, it is virtually impossible to blindly select a wavelet.
The most general-purpose usable wavelet is Daubechies. The Haar wavelet is
actually a differential operator. The Daubechies1 equals Haar.
107
As mentioned, the wavelets have one primary parameter. This
parameter defines two things: region of support and the number of vanishing
moments. The region of support means, how long the wavelet is. This will
affect the localization capabilities. The longer the wavelet, the larger the part
of the time series that will be taken into account for calculating the amplitude
at any time position. And more averaging will occur, similar to that in DFT.
The number of vanishing moments is always the same as the region of
support level. The number of vanishing moments defines the order of the
polynomial that will be ignored if present in the time series.
The attention of researchers has gradually turned from frequency-
based analysis using Fourier transforms to scale-based analysis using wavelet
transforms when it started to become clear that an approach measuring
average fluctuations at different scales might prove less sensitive to noise.
Based on experimental results, any one kind of wavelet has to be chosen for
usage. The mother wavelet DB1 (Daubechies One) may be used because its
detail coefficients indicate sharp changes in a signal indicating transition state
(acceleration or deceleration) and implement it. To segment a signal
automatically using wavelets, an algorithm may be developed and
implemented. The different wavelet families make different trade-offs
between how compactly the basis functions are localized in space and how
smooth they are. Within each family of wavelets are wavelet subclasses
distinguished by the number of coefficients and by the level of iteration.
Wavelets are classified within a family most often by the number of vanishing
moments.
4.4 DIFFERENT FAMI LIES OF WAVELETS FUNCTION
Several families of wavelets have proven to be useful. Some
wavelet families are Meye (meyr), Mexican hat (mexh), Morlet (morl),
108
Gaussian (gaus1-gaus8), Symlet (sym1-sym45), Coiflet (coif1-coif5),
Daubechies (db1-db45), and Biorthogonal (bior1.1-bior1.5 and bior2.2-
bior2.8 and bior3.1-bior3.9). In this proposed method the following four
wavelet transforms namely Harr, Daubechies (db1-db45), Symlet (sym1-
sym45) and Biorthogonal (bior1.1-bior1.5 and bior2.2-bior2.8 and bior3.1-
bior3.9) are chosen.
4.5 THRESHOLD SELECTION RULES
The choice of the thresholding functions and threshold values plays
an important role in the global performance of a wavelet processor for noise
reduction. Threshold selection Rules are based on the underlying model.
There are mainly four threshold selection rules.
1. Rigrsure
Threshold is selected using the principle of Stein’s Unbiased Risk
Estimate (quadrature loss function). One gets an estimate of the risk for a
particular threshold value t. Minimizing the risks in t gives a selection of the
threshold value.
2. Sqtwolog
Fixed form threshold yielding minimax performance multiplied by
a small factor proportional to log (length(s)). It is usually equal to sqrt (2* log
(length (s)))
3. Heursure
Threshold is selected using a mixture of first two methods. . As a
result, if the signal-to-noise ratio is very small, the SURE estimate is very
noisy. Hence, if such a situation is detected, the fixed form threshold is used.
109
4. Minimaxi
This method uses a fixed threshold, chosen to yield minimax
performance for mean square error against an ideal procedure. The minimax
principle is used in statistics in order to design estimators. Since the de-noised
signal can be assimilated to the estimator of the unknown regression function,
the minimax estimator is the one that realizes the minimum of the maximum
mean square error obtained for the worst function in a given set.
4.6 GENETIC ALGORITHM APPROACH IN FITNESS
FUNCTION
GA works with a set of candidate solutions called a population.
Based on the principle of ‘survival of the fittest’, the GA obtains the optimal
solution after a series of iterative computations on its operators: the
reproduction, the crossover, and the mutation. The size of the population and
the probability rates for crossover and mutation are called the control
parameters of the genetic algorithm. GA generates successive populations of
alternate solutions that are represented by a chromosome, i.e. a solution to the
problem, until acceptable results are obtained based on the fitness function.
The fitness function has to provide some measures of the GA’s
performance in a particular environment and assess the quality of a solution in
the evaluation step. The objectives of denoising are to suppress effectively the
noise and restore the original ECG signal. A common goal of optimization in
ECG noise suppression is to minimize the mean square error between the
original ECG signal and the denoisy version of this ECG signal, and so the
MSE has been chosen as the fitness function. Given an original signal x (n),
consisting of N samples, and a reconstructed approximation to this signal
)(ˆ nx , the MSE is given by
110
2
1
1ˆMSE ( ) ( )
N
n
x n x nN
(4.1)
4.7 ECG GA-WAVELET BASED DENOISING
Consider an ECG signal corrupted by standard white Gaussian
noise. The GA was used to search for the optimum wavelet denoising
parameters for ECG signal noise elimination problems. The proposed GA-
wavelet based denoising is shown in Figure 4.1 and can be explained in the
following steps.
Step 1. The inputs for the proposed technique are noisy ECG signal and
wavelet denoising parameters
Step 2. Set the proper wavelet thresholding denoising parameter ranges for
ECG signal and construct the objective functions, including the
mean square error.
Step 3. Optimize the wavelet denoising parameters using GA, by means of
selection, crossover and mutation a satisfied termination criteria is
reached (according to the noise suppression performance) and
select the optimal denoising parameters.
Step 4. Perform a 1-D discrete wavelet transform for the noisy ECG signal
to get all the wavelet coefficients.
Step 5. Threshold the noisy coefficients in ECG signal with the optimal
thresholds and get the modified new ECG components.
Step 6. Reconstruct the denoising ECG signal.
111
Figure 4.1 The GA-wavelet denoising technique
Table 4.1 shows the denoising results of ECG signal obtained using
the GA-wavelet denoising technique for Input SNR: 0–45 dB. Here the
decomposition level is chosen as 3 for all the methods. Among the chosen
wavelet functions (Harr, Daubechies (db1-db45), Symlet (sym1-sym45) and
Biorthogonal (bior1.1-bior1.5 and bior2.2-bior2.8 and bior3.1-bior3.9)) and
the selection rules (Rigrsure, Sqtwolog, Heursure and Minimaxi), one which
provides the best performance in terms of SNR and PRD for the soft threshold
method and proposed method are tabulated. For instance, as seen the
Table 4.1, for input SNR of 10 dB, GA selects the wavelet function
Daubechies6 and Rigrsure for soft threshold method that gives the output
SNR of 23.23 dB and Biorthogonal 3.9 and Rigrsure for proposed method to
give output SNR of 16.75 dB.
11
2
Table 4.1 The performance of denoising the ECG signals in terms of SNR and PRD
Output SNR (dB) Improvement SNR (dB) PRD % Wavelet Function Threshold Selection
Rule Input
SNR
(dB)Soft
Threshold
Method
Proposed
Method
Soft
Threshold
Method
Proposed
Method
Soft
Threshold
Method
Proposed
Method
Soft Threshold
Method
Proposed
Method
Soft
Threshold
Method
Proposed
Method
0 16.67 9.87 16.67 9.87 14.65 38.77 Harr Biorthogonal 3.9 Sqtwolog Minimaxi
5 20.82 11.8 15.82 6.8 9.09 32.22 Biorthogonal 3.9 Biorthogonal 3.9 Sqtwolog Rigrsure
10 23.23 16.75 13.23 6.75 6.89 20.04 Daubechies6 Biorthogonal 3.9 Rigrsure Rigrsure
15 28.18 25.2 13.18 10.2 3.89 10.24 Biorthogonal 3.9 Biorthogonal 3.9 Minimaxi Minimaxi
20 31.77 34.6 11.77 14.6 2.57 3.59 Biorthogonal 3.9 Biorthogonal 3.9 Rigrsure Heursure
25 37.01 42.2 12.01 17.2 1.40 1.74 Biorthogonal 3.9 Biorthogonal 3.9 Heursure Rigrsure
30 41.11 48.4 11.11 18.4 0.87 0.96 Biorthogonal 3.9 Biorthogonal 3.9 Rigrsure Rigrsure
35 45.41 55.2 10.41 20.2 0.53 0.49 Biorthogonal 3.9 Biorthogonal 3.9 Rigrsure Sqtwolog
40 49.52 61.5 9.52 21.5 0.33 0.27 Biorthogonal 3.9 Biorthogonal 3.9 Rigrsure Sqtwolog
45 53 67.1 8 22.1 0.22 0.16 Biorthogonal 3.9 Biorthogonal 3.9 Heursure Sqtwolog
113
It is observed from the Table 4.1, improvement in SNR obtained increases for
the proposed method than that of soft threshold method as the input SNR
increases that is from the input SNR value of 20 dB in the Table 4.1.
4.8 RESULTS AND DISCUSSION
This section presents the simulation performed to verify the
effectiveness of the proposed method. The performance of the proposed
modified method on the basis of two performance measures; 1) Percent root
mean square difference and 2) Signal to noise ratio.
Case 1
A denoising technique for ECG signals is proposed based on
genetic algorithm and wavelet transform. The noise reduction of a signal
depends on the optimum value of the level of decomposition, the suitable
forms of wavelet family and the thresholding techniques. The original ECG
signal and the corrupted ECG with noise is shown in Figure 4.2. The noisy
signal is decomposed using DWT into wavelet coefficients. Thresholding
technique is applied and reconstructed using IDWT to obtain denoised signal.
Figure 4.3 and Figure 4.4 show the signal obtained using soft threshold and
proposed method (modified soft threshold method) respectively.
The percent root mean-square difference and the signal-to-noise
ratio SNR are used as measures of noise reduction performance. The PRD and
SNR (in dB) are calculated as follows by using the equations (2.49) and
(2.50) respectively,
N
n
N
n
nxnxnxPRD1 1
22)(/)(ˆ)(100 (2.49)
114
N
n
N
n
nxnxnxSNR1 1
22
10 )(ˆ)(/)(log10 (2.50)
Case 2
This section represents the simulation of various parameters
performed to verify the effectiveness of the proposed method (modified soft
threshold) when compared with the soft threshold method. The improved
simulated result in signal to noise ratio (SNR) for proposed method is
represented in Figure 4.5. The output signal to noise ratio for proposed
method is initially low when compared with soft threshold method. When the
input SNR is above 20 dB, signal to noise ratio in the proposed method
increased when compared with the soft threshold method. Similarly the PRD
of proposed method initially increases and then decreases as input SNR
increases when compared with the soft threshold method as shown in
Figure 4.6. Finally the bar chart is plotted for signal to noise ratio for different
types of wavelet functions and threshold selection rules are plotted in
Figures 4.7 and 4.8.
Figure 4.2 (a) Original signal (b) The corrupted ECG with noise at input
SNR 20 dB
115
Figure 4.3 The denoised ECG signal resulting from the soft threshold
technique ( = soft, = Rigrsure, = Biorthogonal 3.9)
Figure 4.4 The denoised ECG signal resulting from the modified soft
threshold (proposed method) technique ( = soft, = Heursure,
= Biorthogonal 3.9)
Figure 4.5 Comparision of signal to noise ratio for soft threshold and
proposed method (modified soft threshold method)
116
Figure 4.6 Percent root mean square difference (PRD) for various methods
Figure 4.7 Signal to noise ratio for different types of wavelet function
Figure 4.8 Signal to noise ratio for different types of threshold selection
rule
117
4.9 CONCLUSION
A denoising technique for ECG signals is proposed based on
genetic algorithm and wavelet transform. Selection of wavelet denoising
parameters is critical to the success of noise elimination process for the ECG
signal. For efficient selection of wavelet denoising parameters, besides
experience, GA is proposed to optimize the entire range set of wavelet
denoising parameters leading to an efficient ECG signal filtration. The noise
reduction of a signal depends on suitable forms of wavelet family and the
thresholding techniques. This varies for different kinds of input signals. In
spite of hard thresholding being the simplest method, soft thresholding can
produce better results than hard thresholding. This is because hard
thresholding may cause discontinuities in the signals. In this work, soft
thresholding is compared with a new thresholding algorithm called modified
soft thresholding in terms of SNR and PRD. Modified soft thresholding gives
better results than the soft thresholding. Taken into consideration that GA is a
powerful tool for parameters selection and optimization, therefore the
combination between the GA and wavelet transform makes this denoising
technique more powerful than the available systems.