Docstoc

ANALYSIS ELECTROCARDIOGRAM SIGNAL USING ENSEMBLE EMPIRICAL MODE DECOMPOSITION AND TIME

Document Sample
ANALYSIS ELECTROCARDIOGRAM SIGNAL USING ENSEMBLE EMPIRICAL MODE DECOMPOSITION AND TIME Powered By Docstoc
					  International Journal of JOURNAL OF COMPUTER (IJCET), ISSN 0976-
 INTERNATIONALComputer Engineering and Technology ENGINEERING
  6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME
                           & TECHNOLOGY (IJCET)

ISSN 0976 – 6367(Print)
ISSN 0976 – 6375(Online)                                                         IJCET
Volume 4, Issue 2, March – April (2013), pp. 275-289
© IAEME: www.iaeme.com/ijcet.asp
Journal Impact Factor (2013): 6.1302 (Calculated by GISI)
                                                                             ©IAEME
www.jifactor.com



    ANALYSIS ELECTROCARDIOGRAM SIGNAL USING ENSEMBLE
    EMPIRICAL MODE DECOMPOSITION AND TIME-FREQUENCY
                       TECHNIQUES

       Samir Elouaham1, Rachid Latif1, Boujemaa Nassiri1, Azzedine Dliou1, Mostafa
                            Laaboubi2, Fadel Maoulainine3
      1
       (ESSI, National School of Applied Sciences, Ibn Zohr University Agadir, Morocco)
             2
               (High School of technology, Ibn Zohr University Guelmim, Morocco)
       3
         (Team of Child, Health and Development, CHU, Faculty of Medicine, Cadi Ayyad
                                University, Marrakech, Morocco)


  ABSTRACT

          Electrocardiogram signals (ECG) are among the most important sources of diagnostic
  information in healthcare. During ECG measurement, there may be various noises which interfere
  with the ECG information identification that cause a misinterpretation of the ECG signal. In this
  paper, the Empirical Mode Decomposition (EMD), the Ensemble Empirical Mode
  Decomposition (EEMD) and the Discrete Wavelet Transform (DWT) were used to overcome
  these problems. These techniques are applied to a noisy electrocardiogram abnormal signal
  obtained by adding white noise. A comparative performance study of these three techniques in
  terms of several standard metrics was used. The EEMD was chosen for its better localization of
  the components of the ECG signal. The non-stationary and non-linear nature of the ECG signals
  makes the use of time-frequency techniques inevitable. The parametric and non-parametric time–
  frequency techniques allow giving simultaneous interpretation of the non-stationary signal in both
  time and frequency which allows local, transient or intermittent components to be elucidated. In
  this paper, the parametric techniques used are periodogram (PE), capon (CA) and time-varying
  autoregressive (TVAR) and non-parametric techniques used are Smoothed Pseudo Affine Wigner
  Distributions (SPAWD) and S-transform (ST). The abnormal signal used is obtained from the
  patient with an atrial fibrillation. The PE technique shows its superior performance, in terms of
  resolution and interference-terms suppressing, as compared to other time-frequency techniques
  used in this paper. From the obtained results, the EEMD technique is a more powerful tool for
  elimination and restoration of the original signal than the other techniques used in this paper. This
  study shows that the combination of the EEMD and the Periodogram techniques are a good issue
  in the biomedical field.

                                                  275
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

Keywords: EEMD, Time-frequency, PE, Non-stationary, ECG signals.

INTRODUCTION

        Electrocardiography is a commonly used, noninvasive procedure for recording
electrical changes in the heart. The record, which is called an electrocardiogram (ECG),
shows the series of waves that relate to the electrical impulses which occur during each beat
of the heart. The waves in a normal record are named P, QRS complex, and T and follow in
alphabetical order. The figure 1 shows the normal ECG signal [1-3]:




                           Fig.1: A waves of the normal ECG signal

        The ECG signal is often corrupted by various noises such as                   prolonged
repolarization, changes of electrode position, muscle contraction (electromyography) and
power line interference (electrode) [3]. Among the objectives of this study is to separate the
signal component from the undesired artifacts. These artifacts can’t facilitate a good, easy
and accurate interpretation of the ECG signals that implies bad diagnostic given by the
expert. Therefore the elimination of noises is inevitable. To overcome this problem, several
techniques are presented whose goal is the cancellation of the noises existing during the
recordings of the biomedical signals. The techniques such as elliptic filter, median filter,
Wiener filter, Discrete Wavelet Transform (DWT), Empirical Mode Decomposition (EMD)
and Ensemble Empirical Mode Decomposition (EEMD) are the solutions proposed [4-12].
The DWT, EMD and EEMD techniques are used in several fields as acoustic, climatic and
biomedical; these techniques give good results. The disadvantage of elliptic filter, median
filter, Wiener filter is that they eliminate the high frequency components of ECG signals.
The drawback of DWT is its non-adaptive basis due to the selection process of the basis
function that is controlled by the signal components that are relatively large in a frequency
band [10, 12]. And the disadvantage of Empirical Mode Decomposition (EMD) is the
appearance of mode-mixing effect in signal restoration [4-6]. To overcome these problems
the new technique called Ensemble Empirical Mode Decomposition (EEMD) is proposed by
Wu and Huang [7]. The choice of a powerful technique among of these techniques is related
to the result obtained after denoising ECG signals; these results are the original characteristic
waveforms such as QRS complexes, the P and T waves and also Q, R and S waves. A mean
square error (MSE) and percent root mean square difference (PRD) between filter ECG
output and clean ECG were used for giving the performance of the denoising techniques
used. The results obtained by the EEMD show high resolution, noise cancellation and
preservation of true waveforms of ECG signals more than the other techniques as EMD and
DWT. Among the objectives presented in this paper is the choice of the useful technique for
any application that needs the denoising of non-stationary and non-linear biomedical signals
such as ECG, EEG, EOG and EMG in the pretreatment stage.

                                              276
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

        The biomedical signals such as electromyography (EMG), electroencephalogram
(EEG) and electrocardiogram (ECG) are non linear and non stationary. Due to multi-
component signals, expert can’t give a good diagnostic. There are several techniques to
analyze the ECG signals. Traditionally time-domain is based on the measurement of the
surface QRS and the amplitude of the P, the QRS complexes and the T waves. The
disadvantage of this technique is that it cannot give the useful information of frequency
components in the time domain [13]. To overcome this problem the Fast Fourier Transforms
(FFT) is presented for giving a frequency-domain representation of a signal where the
analysis can identify the change of the frequency components of the abnormal ECG signals
[14]. This technique also has a disadvantage; the frequency component is not revealed vary
with time. The FFT technique is unsuitable. To tackle these limitations, the time-frequency
techniques are useful [15- 30]. In recent years the time-frequency methods constitute an
important amelioration in signals analysis specifically in the biomedical signals. These
techniques are Wavelets, Wigner-Ville, Choi-Williams and Born-Jordan [28]. In this paper
we present the parametric techniques such as periodogram (PE) capon (CA) and time varying
autoregressive (TVAR) and non parametric as S-transform (ST) and smoothed pseudo affine
Wigner distributions (SPAWD) [15-20]. The drawback of the SPAWD is the appearance of
the cross-terms. These cross-terms hide useful information of interest in the signal that can
help the expert to extract real features. The features obtained from ST are not completely
distinctive and don’t give clear information about the component of the signal. These
disadvantages are resolved by the PE time-frequency technique. The performance of these
techniques is given by the calculation of the variance obtained by adding the noise of the
modulated signal. The results show that the periodogram technique is more powerful than the
other techniques. The electrocardiogram signals used are normal and abnormal. The
abnormal cardiac signal was taken from a patient with atrial fibrillation [31].
This paper is organized as follows: Section II talks about the denoising techniques,
parametric and non parametric time-frequency techniques. Section III provides normal and
abnormal electrocardiogram signals. The obtained results of the denoising techniques are
given in section IV. The section V gives the results of time-frequency techniques used.
Finally, conclusion is provided in section VI.

1.     TECHNIQUES USED

1.1 Denoising techniques
1.1.1 EMD

        The EMD was proposed by Huang and al. as a tool to adaptively decompose a signal
into a collection of AM–FM components [4-6]. The EMD method has no mathematical
foundations and analytical expressions for the theoretical study. The various works have
successfully used the EMD to real data in several fields such as biomedicine, study of climate
phenomena, seismology or acoustics [4-6]. These studies show satisfaction and matching
condition used in non-stationary signal processing. The EMD decomposes adaptively a non-
stationary signal into a sum of functions oscillatory band-limited d(t) called Intrinsic Mode
Functions IMFJ(t). These functions IMFJ(t) oscillate around zero and can express the signal
x(t) by the expression:



                                             277
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME
                                                 k
                                      x (t ) =   ∑d       j   (t ) + r (t )                   (1)
                                                 j =1



Where r(t) is the residue of low frequency.
Each IMFJ(t) must satisfy two conditions:
- The number of zero crossings and the number of extreme signal must be equal throughout
the analyzed signal,
- At any point, the average of the envelopes defined by local extreme of the signal must be 0.
The higher order IMFJ(t) corresponds to low oscillation components, while lower-order
IMFJ(t) represents fast oscillations. For different decomposed signals the number of IMFJ(t)
is variable. It also depends on the spectral content of the signal. The Rilling study presents the
technical aspects of the EMD implementation and makes the five-step algorithm given by the
following [4-6]:
a) Extract the extreme of the signal x(t),
b) Deduce an upper envelope emax (t) (resp. lower emin (t)) by interpolation of the maxima
(resp. minima),
c) Define a local average m(t) as the sum of the half-envelopes by the expression:

                                     m (t )=(emax (t )+emin (t ))/2                           (2)

d) Deduce a local detail dJ(t)=IMFJ(t) by the expression:

                                     d ( t ) = x ( t )− m ( t )                               (3)

e) The iteration is given by the expression (1).
The first IMF contains the terms of higher frequencies and contains the following terms of
decreasing frequency up to forward only a residue of low frequency.

1.1.2 EEMD

        The ensemble EMD method has been proposed to overcome mode mixing problem
existing in EMD technique [7]. The EMD technique allows giving all solutions that give the
true IMF by repeating the decomposition processes. The procedure of the EEMD method is
given as follow:
Step 1: Add white noise with predefined noise amplitude to the signal to be analyzed.
Step 2: Use the EMD method to decompose the newly generated signal.
Step 3: Repeat the above signal decomposition with different white noise, in which the
amplitude of the added white noise is fixed.
Step 4: Calculate the ensemble means of the decomposition results as final results.
The signal x(k) is decomposed into a finite number of intrinsic mode functions (IMFs) and a
residue.
                                                  n )     )
                                         x ( k )= ∑ c i + r                                   (4)
                                                        i =1
                                              )
Where n represents the number of the IMFs, ci is the ith IMF that is the ensemble mean of the
                                                                             )
corresponding IMF obtained from all of the decomposition processes and r is the mean of
the residues from all of the decomposition processes.
                                                        278
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

1.1.3       Discrete Wavelet Transform
        Wavelet theory appeared in early 1990 [10-12]. It affects many areas of mathematics,
particularly signal processing and image. The multi-resolution analysis provides a set of
approximation signals and details of a start signal. In discrete wavelet transform (DWT), for
analyzing both the low and high frequency components in x(n), it is passed through a series of
low-pass and high-pass filters with different cut-off frequencies. The DWT is one of the
operations that provide a multi-resolution signals. The DWT satisfies the energy conservation.
The original signal can be properly reconstructed via employing that technique, which gained
popularity in ECG denoising [10]. In the wavelet domain, this term means the noise rejection by
thresholding adequate. The contamination of the noise signal is concentrated in the details [10-
12]. We have used the orthogonal DWT for signal decomposition on the time-scale plan, which
represents the signal x(t) by:
                                                               k    ∞                    j
                                                      x (t ) = ∑ ∑ w j ( k )ψ (2 t − k )                             (5)
                                                              j =1 k =−∞


Where the function ψ (t) represents a discrete analysis wavelet and the coefficients wj (k) represent
the signal at level j. The Performance of DWT depends on the choice of the wavelet and its
similarity to analyzed signal.
1.2 Time-frequency techniques
The time-frequency technique is a tool to treatment non-stationary signal, which used time and
frequency simultaneously to represent the non-stationary signal.
1.2.1       Parametric techniques
The parametric time-frequency techniques used in this work are the Capon (CA), the
Periodogram (PE) and the Time Varying autoregressive (TVAR).

      1.2.1.1 Capon technique
The estimator of minimum variance called Capon estimator (CA) does not impose a model on the
signal. At each frequency f, this method seeks a matched filter whose response is 1 for the
frequency f and 0 everywhere else [15-16].
                                                                                         1
                                     CA( n, f )= a (n, f ) H Rx a ( n, f ) =                                       (6)
                                                                                 Z H .Rx  n  −1.Z f
                                                                                   f      

Where
 - CA ( n , f ) is the output power of the Capon filter, excited by the discrete signal x(n) sampled at
the period te,
- a ( n , f ) = ( a 0 ,..., a p ) is the impulse response of the filter at frequency n,
-              {    T
    Rx n = E x n x n
                             }      is the autocorrelation matrix of crossed x(n) of dimension ( p + 1) *( p + 1) ,

- x n = ( x( n− p ),..., x( n) ) is the signal at time n,
     
      H
-           (
    Z f = 1,e2iπ fte ,...,e2 iπ fte p   ) is the steering vector,
- ( p + 1) is the number of filter coefficient, the exponent H is conjugate transpose and the
superscript T for transpose.

                                                                           279
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

   1.2.1.2 Periodogram technique

The Periodogram (PE) is the derivate of the Capon (CA) technique. The spectral estimator of
this method is defined by the following equation [16-19]:

                                       PE (n, f )= Z H .Rx .Z f /((p+1) 2 )
                                                     f                                              (7)

The two previous techniques defined by the equations 6 and 7 can be applied sliding
windows. There is no theoretical criterion for choosing the filter order and duration of the
window [17-20]. The parametric techniques depend on the signal so that the frequency
response has a different shape and then different properties according to the signal
characteristics. The choice of the window is more crucial to the time-frequency resolution.
The CA and PE estimators usually have a better frequency resolution. Both techniques are
well suited to signals containing some strong spectral components such as ECG and EMG
biomedical signals.

   1.2.1.3 Time-Varying autoregressive

The time-varying frequency can be extracted from its parameters ai (t) . Since the non-
stationary signal is modeled as the output of the TVAR process, with a zero-mean white noise
input w(t), the power spectral density of the stationary signal is given by [21]:
                                                                           2
                                                                        σw
                                 TVAR (t , f ) =                                                2
                                                                                                    (8)
                                                         i= p           − j 2πυ i
                                                   1 + ∑ i =1 ai (t ) e
         2
Where   σ w is   the variance of the white noise w(t).

   1.2.2 Non-parametric techniques
   1.2.2.1 S-transform

The S-transform (ST) is a time-frequency representation known for its local spectral phase
properties. A key feature of the ST is that it uniquely combines a frequency dependent
resolution of the time-frequency space with absolutely referenced local phase information.
This allows to define the meaning of phase in a local spectrum setting and results in many
advantageous characteristics. It also exhibits a frequency invariant amplitude response, in
contrast to the wavelet transform. The ST technique is given by [21]:
                                                                     (τ − t )2 f 2
                                            +∞          f        −
                                                                                   − i 2π ft
                               ST (t , f ) = ∫ h (t )        e              2
                                                                                 e             dτ   (9)
                                            −∞          2π
Where h(t) is analysis windows.

   1.2.2.2 Smoothed Pseudo Affine Wigner Distributions

The affine Wigner distributions show great potential as flexible tools for time-varying
spectral analysis. However, for some distributions of the Cohen’s class, they present two
major practical limitations: first the entire signal enters into the calculation of these

                                                            280
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

distributions at every point (t, f), and second, due to their nonlinearity, interference
components arise between each pair of signal components [18-19]. To overcome these
limitations, the time-windowing function h introduced to attenuate interference components
that oscillate in the frequency direction and for suppressing interference terms oscillating in
the time direction. We must smooth in that direction with a low-pass function G. The
smoothed pseudo affine Wigner distribution (SPAWD) is given by equation:

                                       +∞           µk (u)                             *
                                 k
                       SPAWDx (t , f ) = ∫ G(u)                   Tx (t ,λk (u) f ; Ψ)Tx (t , λk (−u) f ; Ψ)du   (10)
                                       −∞         λk (u)λk (−u)

Where          Tx (t , f ; Ψ )is the continuous wavelet transform, µ k (u ) is a real positive function,
               k (e −u  − 1) k 1 1 and Ψ (τ ) = h (t ) ei 2πτ is a band pass wavelet function.
λ k ( u )= (                ) −
                 e − ku − 1

2.             BIOMEDICAL SIGNALS

       The biomedical signals such as EMG, EEG and ECG are non-stationary and
nonlinear. The figure 2 presents the normal electrocardiogram signal; this signal presents the
P, T waves and the QRS complex.




                                                   Fig.2: Normal ECG signal

The abnormal signal used in this work is obtained by the patient who has anomaly named
atrial fibrillation. The figure 3 shows this abnormal signal [31]. The atrial rate exceeds 350
beats per minute in this type of arrhythmias. This arrhythmia occurs because of
uncoordinated activation and contraction of different parts of the atrial. The higher atria rate
and uncoordinated contraction leads to ineffective pumping of blood into the ventricles.




                                                  Fig.3: Abnormal ECG signal




                                                                     281
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

3.      RESULTS OF DENOISING TECHNIQUES

        The ECG signals are often interfered by noises such as power interference noise and
the electromyography (EMG) noise caused by the muscle activity during recording. These
artifacts strongly influence the utility of recorded ECG signals. In this work the EMD, EEMD
and the DWT are presented for suppressing the noise that interferes with information; the
white noise is used. These techniques are applied to the abnormal electrocardiogram signal.
We compare the performance of the EEMD technique quantitatively with respect to the other
techniques based on two metrics: Mean Square Error (MSE) and Percent Root Mean Square
Difference (PRD). These metrics are computed as follows:

                                               1                                  2
                                       MSE =         N
                                                   ∑ n =1 ( x ( n )− x ( n ))                             (11)
                                               N

                                                  N                           2
                                                ∑ n =1 ( x ( n ) − x ( n ))
                                       PRD =            N      2
                                                                                  *100                    (12)
                                                      ∑ n =1 x ( n )


Where x(n) is the original ECG signal, x(n) denotes the reconstruction of the ECG signal and
N is the number of ECG samples used.
The table 1 shows the MSE and PRD at different input SNR levels, the range of input SNR
levels is from -20 dB to 10 dB. The obtained results of the EEMD technique give the
smallest MSE and PRD which attests its capability to yield improved ECG signal with better
quality than the other techniques used at different inputs of the SNRs.
The table 2 gives the obtained results of the MSE and PRD for the denoising techniques using
different abnormal ECG signals under consideration at a particular input SNR level of -5 dB.
The EEMD technique outperforms other denoising techniques; the MSE and PRD of the
EEMD are relatively lower for all abnormal signals than the other techniques. The obtained
results show that EEMD technique is more effective than the other techniques at the level of
noise suppression and recovery of a form of original signal.

                                         MSE                                                    PRD
 Signal          SNR
  used           (db)      EEMD        EMD                DWT                         EEMD       EMD       DWT
                  -20      0.7006      0.9124             1.0919                      647.827   739.256   808.730
                  -15      0.2805      0.2884             0.3560                      409.901   415.661   461.761
  Atrial          -10      0.0749      0.0928             0.1046                      211.744   235.730   250.267
fibrillati         -5      0.0274      0.0291             0.0389                      128.120   132.112   152.662
on                 0       0.0080      0.0086             0.0138                      69.9574   71.1765   90.7660
(040126)           5       0.0042      0.0044             0.0071                      49.9532   51.5151   65.1999
                   7       0.0032      0.0036             0.0055                      43.8844   46.6015   57.5039
                   10      0.0030      0.0033             0.0038                      42.1739   44.2632   47.6864


             Table 1: Comparison of the MSE and PRD obtained by using different techniques




                                                     282
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME


                             MSE                                   PRD
 Abnormal   EEMD             EMD         DWT       EEMD         EMD              DWT
ECG Signals
   used

   04126        0.0274      0.0291       0.0389    128.120     132.112          152.662

   04746        0.0928      0.0943       0.1040    149.5683   150.7659         158.3118

   04015        0.4104      0.4564       0.4725    129.7979   137.1329         139.5305
   04043        0.1163      0.1396       0.1862    123.3458   126.4191         144.0233


   04908        0.3313      0.3829       0.4211    130.3802   140.1509         146.9852

   04936        0.1857      0.1980       0.2422    129.7550   133.9816         148.1775

   05091        0.0660      0.0736       0.0872    131.7030   139.1122         151.3531

    Table 2: Comparison of the MSE and PRD obtained by using different techniques of the
                          different abnormal signals adding -5 dB

        Among the goals of this study is to eliminate the noise that corrupts the original ECG
signal. The figure 4 shows the reconstruction of the original abnormal signal without noises
by the denoising techniques used. The noise added is 8 db. This noise is an artifact frequency
component, which will cause misinterpretation of the physiological phenomena. After the
cancellation of artifacts by the EMD, EEMD and DWT techniques, the figure 4c given by
EEMD technique presents the true shape of the T wave and the true area of the Q and the
QRS complex. This artifact can be caused by breathing or movement of patients or by
instruments. It can be observed that the abnormal ECG signal largely restore the original
shape and clearly eliminates noises by EEMD technique (Fig. 4c). The figure 4a shows that
the DWT technique doesn’t restore the true area of QRS complex, the true area of T waves
and also true area of the Q waves. One of the major drawbacks of the EMD technique is the
frequent appearance of mode mixing effect in signal restoration. This effect does is not
revealed by EEMD technique (Fig. 4c). It is also evident that the EEMD is more suitable for
specific abnormal ECG signal feature enhancement. The obtained results show the
effectiveness of the EEMD technique and its capability of extracting useful information from
ECG signals affected by noises as compared to the other techniques used in this work such as
EMD and DWT.




                                             283
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME




        Fig.4: Denoising abnormal ECG signal using DWT (a), EMD (b) and EEMD (c)


4.      RESULTS OF TIME-FREQUENCY TECHNIQUES
     4.1 Performance techniques

        The time-frequency techniques used in this study are applied to a monocomponent
signal to find the most performance technique. The monocomponent signal used is given by
the following equation:
                                         x ( t ) = a e jφ ( t )                (13)

The instantaneous frequency (IF) is given by the following equation:
                                                     1
                                              f =      dφ / dt= f 0 +β t                    (14)
                                                    2π
Where a=1, fo=0.05fs, β = 0.4fs, φ(t) is the analytic signal phase and fs = 1/T is the sampling
frequency.
The bias (B) and the variance (VAR) of the estimate present the most important factors that
decide the quality of estimation. These two notions can be defined by the following
expressions:
                                               ˆ               ˆ
                                            B( fi (t )) = ε  ∆fi (t )                     (15)
                                                                      
                                                                           2
                                              ˆ               ˆ
                                         VAR( fi (t )) = ε (∆fi (t ))                      (16)
                                                                      

Where ∆ fˆ i (t ) = fi (t) - fˆi (t ) , fi (t) and fˆi are the instantaneous frequency and instantaneous
frequency estimate respectively. The signal length used in the time-frequency techniques is
N=256 samples and the total signal duration is 1 s. The sampling frequency was fs=2 NHz.


                                                     284
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

Using different Signal-to-Noise Ratio (SNR), gaussian white noise samples are added to the
signal. The figure 5 shows the performance of the PE, CA, TVAR, ST and SPAWD time-
frequency techniques applied to a linear FM signal with 256 points.




 Fig.5: Performance of variance of various techniques of a linear FM signal with length N=256
                                            samples
       According to the results of the figure 5, the PE time-frequency technique has a
minimal variance for all SNR's. The low minimum variance can indicate the performance of
the time-frequency technique. The PE technique surpasses the other time-frequency
techniques in robustness where it gives the minimum of the variance at low SNR.

   4.2 Time-frequency images

In this section, we applied the parametric technique PE which has a minimal variance in
parametric technique and the non parametric technique SPAWD that has a minimal variance
in non-parametric techniques on ECG signals. The figures 6 and 7 present the time-
frequency images of the normal and abnormal ECG signals (figure 2 and 3). These time-
frequency images are obtained by using the calculation of the equation 7 and 10 of the PE
and the SAPWD techniques respectively. We converted the normal and abnormal ECG
signals into their analytical forms by using Hilbert transform first, then we apply the non-
parametric technique.
The figure 6 presents the time-frequency images of the normal ECG signal in 2D (a and b)
and 3D (a’ and b’). The PE and SPAWD time-frequency techniques are capable to identify
the frequency components over time in the normal signal with big difference between the
both. The PE technique (Fig 6a and 6a’) can follow and identify the different T waves and
QRS complexes existing in the normal signal. The SPAWD technique (Fig 6b and 6b’) shows
the presence of the interference-term which doesn't allow finely the identification of the QRS
complexes and also it cannot show the T waves in this normal signal. We can conclude that
the PE technique provides a good localization and visualization of the QRS complexes and T
waves over time.




                                             285
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

                                                                         300
                                             a                                                                                                     b
                                                                         250




                                              F r e q u e n c y (H z )
                                                                         200

                                                                         150

                                                                         100

                                                                         50

                                                                               100   200   300   400       500      600   700   800   900   1000
                                                                                                       Time (Samples)



                                             a’                                                                                                    b’




Fig.6: Periodogram (a and a’) and SPAWD (b and b’) time-frequency images of a normal ECG
                                        Signal

         The figure 7 presents the time-frequency images of the abnormal ECG signal in 2D (a
and b) and 3D (a’ and b’). This abnormal signal is obtained from a patient with artial
fibrillation. The parametric PE and non-parametric SPAWD techniques are able to give all
QRS complexes of the abnormal signal in time-frequency images (2D and 3D) with big
difference between the both. The non-parametric time-frequency image (figure 2D) presents
the interference-terms that hide the good visualization of the QRS complex. The SPAWD
technique can’t show the T wave in figure 7 (b and b’). The PE technique allows tracking the
change of the frequency components of each T wave and QRS complex. The obtained results
of the parametric technique in figure 7 (2D and 3D) show the morphology of QRS complexes
and T waves with clear and with good resolution and also we can note the overlapping
between the QRS complexes and T waves (QRS2 and T2), (QRS3 and T3) and (QRS5 and
T5). These overlaps indicate the abnormalities of this signal. The obtained results of the
parametric time-frequency technique allow giving all frequency components of the normal
and abnormal signals, with high time-frequency resolution and the interference-terms
suppressing. The PE technique is expected to be more efficient in analyzing the ECG signal
than the other time-frequency techniques used in this work.




                                            286
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

                                                                                                a                                                                                                     b
                                                                                                                                200
                              140
                              120
                                                                                                                                150
    F re q u e n c y (H z )




                                                                                                           F re q u e n c y (H z )
                              100
                               80                                                                                               100
                               60
                               40                                                                                                    50
                               20

                                    100   200   300   400       500     600   700   800   900   1000                                      100   200   300   400       500     600   700   800   900   1000
                                                            Time (Samples)                                                                                        Time (Samples)
                                                                                                                                                                                                       b’
                                                                                                a’




 Fig.7: Periodogram (a and a’) and SPAWD (b and b’) time-frequency images of an abnormal
                              ECG signal with atrial fibrillation


5                              CONCLUSION

        In this paper, we present a comparative performance study of the denoising
techniques and the parametric and non-parametric ones. The techniques used in this work are
the denoising methods: EEMD, EMD and DWT and also the time-frequency methods: PE,
CA, TVAR, SPAWD and ST. In the case of the time-frequency methods, the parametric
technique is able to detect the different QRS complexes and T waves of the atrial fibrillation
and normal signals with high resolution and cross-terms suppressing. The PE time-frequency
technique is expected to be more efficient in analyzing the abnormal ECG than the other
techniques used. In this study, the EMD, EEMD and DWT techniques were applied to the
abnormal ECG signal with atrial fibrillation, in order to eliminate the effects of noise which
hide the useful information. The main advantages of the EMD and EEMD techniques are that
they do not make any prior assumption about the data being analyzed. The EEMD technique
shows the cancellation of artifacts of abnormal signal which are due to different noises. The
DWT technique can’t restore the useful information of the different components of the
electrocardiogram signal as an area of QRS complex and T wave. The obtained result of the
analysis of ECG signal shows that the EEMD technique could be successfully applied for the
attenuation noise. The combination of the EEMD and PE techniques can be a good issue in
analyzing the ECG signals.




                                                                                                     287
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

REFERENCES

[1]     P. de Chazal, Maraia O’ Dwyer, Richard B. Reilly, Automatic classification of
heartbeats using ECG morphology and heartbeat interval features: IEEE Transactions on
Biomedical Engineering, 51 (7), 2004, 1196–1206.
[2]     G. Bortolan, Christian Brohet, Sergio Fusaro. Possibilities of using neural networks
for ECG classification, Journal of Electrocardiology, 29, 1996, 10-16.
[3]     N. E. Huang, Z. Shen, S. R. Long, M. C. Wu, E. H. Shih, Q. Zheng, C. C. Tung and
H. H. Liu, The empirical mode decomposition method and the Hilbert spectrum for non-
stationary time series analysis, Proc. Roy. Soc. London 454A, 1998, 903–995.
[4]     P. Flandrin, G. Rilling and P. Goncalves, Empirical mode decomposition as a filter
bank: IEEE Signal Process. Lett. 11, 2004, 112–114.
[5]     S. Elouaham, R. Latif, A. Dliou, F. M. R. Maoulainine, M. Laaboubi, analysis of
biomedical signals by the empirical mode decomposition and parametric time-frequency
techniques: International Symposium on security and safety of Complex Systems, May 2012,
Agadir, Morocco.
[6]     G. Rilling, P. Flandrin, P. Goncalves, On Empirical Mode Decomposition and its
Algorithms: IEEE-EURASIP Workshop on Nonlinear Signal and Image Processing,
June, 2003, Grado-Trieste, Italy.
 [7]    JR. Yen, JS. Shieh, NE. Huang. Complementary ensemble empirical mode
decomposition: a novel noise enhanced data analysis method. Adv Adapt Data Anal, 2010,
2(2), 135–56.
[8]     PH. Tsui, CC. Chang, NE. Huang. Noise-modulated empirical mode decomposition.
Adv Adapt Data Anal 2010, 2(1), 25–37.
[9]     Z. Wu, NE. Huang, Ensemble empirical mode decomposition: a noise-assisted data
analysis method. Adv Adapt Data Anal 2009, 1(1), 1–41.
[10] R. Larson, David. Unitary systems and wavelet sets Wavelet Analysis and
Applications. Appl. Numer. Harmon. Anal. Birkhäuser. 2007, 143–171.
[11] A.N. Akansu, W.A. Serdijn, and I.W. Selesnick, Wavelet Transforms in Signal
Processing: A Review of Emerging Applications, Physical Communication, Elsevier, 3, issue
1, 2010, 1-18.
[12]     L. Sharma, S. Dandapat, A. Mahanta, Multiscale wavelet energies and relative energy
based denoising of ecg signal, in: 2010 IEEE International Conference on Communication
Control and Computing Technologies, 2010, 491–495.
[13] C. Bigan, M.S. Woolfson, Time-frequency analysis of short segments of biomedical
data: IEEE Proc.-Sci. Meas. Technol. 147 (6), 2000, 368–373,
[14] R.H. Clayton, A. Murray. Estimation of the ECG signal spectrum during ventricular
fibrillation using the fast Fourier transform and maximum entropy methods: Proceedings of
the Computers in Cardiology, 1993, 867 – 870.
[15]     M.T. Özgen. Extension of the Capon’s spectral estimator to time–frequency analysis
and to the analysis of polynomial-phase signals, Signal Process, 83, n.3, 2003, 575–592.
[16]     S. Elouaham, R. Latif, A. Dliou, M. Laaboubi, F. M. R. Maoulainine. Parametric and
non-parametric time-frequency analysis of biomedical signals, International Journal of
Advanced Computer Science and Applications, 4(1), 2013, 74-79.
[17] Castanié F. Spectral Analysis Parametric and Non-Parametric Digital Methods, (ISTE
Ltd, USA, 2006) 175-211
[18] P. Goncalves, F. Auger, P. Flandrin. 1995. Time-frequency toolbox.
                                            288
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

[19] P. Flandrin, N. Martin, M. Basseville. Méthodes temps-fréquence (Trait. Signal 9,
1992).
[20] M. Jachan, G. Matz, and F. Hlawatsch. Time-frequency-autoregressive random
processes: Modeling and fast parameter estimation: IEEE ICASSP, VI, 2003, 125–128
[21] R. Stockwell, L. Mansinha, R. Lowe, Localization of the Complex Spectrum: The S.
Transform, IEEE Transactions on Signal Processing, 44 (4), 1996, 998–1001.
[22] R. Latif, E. Aassif, G. Maze, A. Moudden, B. Faiz, Determination of the group and
phase velocities from time-frequency representation of Wigner-Ville. Journal of Non
Destructive Testing & Evaluation International, 32, n.7, 1999. 415- 422.
[23]     R. Latif, E. Aassif, G. Maze, D. Decultot, A. Moudden, B. Faiz. Analysis of the
circumferential acoustic waves backscattered by a tube using the time-frequency
representation of wigner-ville, Journal of Measurement Science and Technology, 11, n. 1,
2000, 83-88.
[24] R. Latif, E. Aassif, A. Moudden, B. Faiz, G. Maze. The experimental signal of a
multilayer structure analysis by the time-frequency and spectral methods, NDT&E
International, 39, n. 5, 2006, 349-355.
[25] R. Latif, E. Aassif, A. Moudden, B. Faiz. High resolution time-frequency analysis of
an acoustic signal backscattered by a cylindrical shell using a Modified Wigner-Ville
representation, Meas. Sci. Technol.14, 2003, 1063-1067.
[26]     R. Latif, M. Laaboubi, E. Aassif, G. Maze. Détermination de l’épaisseur d’un tube
élastique à partir de l’analyse temps-fréquence de Wigner-Ville, Journal Acta-Acustica, 95, n.
5, 2009, 843-848.
[27] R.B Pachori, P. Sircar. A new technique to reduce cross-terms in the Wigner
distribution, Digital Signal Processing, 17, 2007, 466–474.
[28] A. Dliou, R. Latif, M. Laaboubi, F.M.R. Maoulainine. Arrhythmia ECG Signal
Analysis using Non Parametric Time-Frequency Techniques, International Journal of
Computer Applications, 41 (4), 2012, 25-30.
[29] A. Dliou, R. Latif, M. Laaboubi, F. M. R. Maoulainie, S. Elouaham, Noised abnormal
ECG signal analysis by combining EMD and Choi-Williams techniques: IEEE conference
publications; International Conference on Complex Systems (ICCS), 5-6 Novembre, 2012
Agadir, Morocco.
[30] H. Choi., W. Williams, Improved time-frequency representation of multicomponent
signals using exponential kernels: IEEE Trans. on Acoustics, Speech and Signal Processing,
37, 1989, 862-871.
[31] Physiobank, Physionet viewed August 2005 Physiologic signal archives for
biomedical research, http://www. physionet.org/physiobank/.
[32] Vivek S. chaudhari and Dr. P.B.Patil, “Electrocardiographic Detection of Left
Ventricular Hypertrophy by the New “Sai” Criteria” International Journal of Electrical
Engineering & Technology (IJEET), Volume 3, Issue 2, 2012, pp. 51 - 57, ISSN Print : 0976-
6545, ISSN Online: 0976-6553.
[33] Kavita L.Awade, “Design and Development of Microstrip Array Antenna for Wide
Dual Band Operation”, International journal of Electronics and Communication Engineering
&Technology (IJECET), Volume 1, Issue 1, 2010, pp. 33 - 43, ISSN Print: 0976-6464, ISSN
Online: 0976-6472




                                             289

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:1
posted:4/26/2013
language:Latin
pages:15