`
`Integration of HRV, WT and neural networks for ECG arrhythmias
`classification
`
`READS
`137
`
`Abdennaceur Kachouri
`University of Sfax
`
`329 PUBLICATIONS(cid:100)(cid:100)(cid:100)1,968 CITATIONS(cid:100)(cid:100)(cid:100)
`
`SEE PROFILE
`
`Article · May 2011
`
`CITATIONS
`5
`
`3 authors:
`
`Adel Dallali
`Faculty of Sciences of Gafsa.
`
`15 PUBLICATIONS(cid:100)(cid:100)(cid:100)89 CITATIONS(cid:100)(cid:100)(cid:100)
`
`SEE PROFILE
`
`Mounir Samet
`University of Sfax
`
`205 PUBLICATIONS(cid:100)(cid:100)(cid:100)1,496 CITATIONS(cid:100)(cid:100)(cid:100)
`
`SEE PROFILE
`
`Some of the authors of this publication are also working on these related projects:
`
`Study and design of ultra wide band antenna View project
`
`Optical image encryption View project
`
`All content following this page was uploaded by Adel Dallali on 07 April 2020.
`
`The user has requested enhancement of the downloaded file.
`
`APPLE 1041
`
`1
`
`
`
` VOL. 6, NO. 5, MAY 2011
`ARPN Journal of Engineering and Applied Sciences
`
`©2006-2011 Asian Research Publishing Network (ARPN). All rights reserved.
`
` ISSN 1819-6608
`
`www.arpnjournals.com
`
`
`
`INTEGRATION OF HRV, WT AND NEURAL NETWORKS FOR
`ECG ARRHYTHMIAS CLASSIFICATION
`
`
`
`
`
`A. Dallali, A. Kachouri and M. Samet
`Laboratory of Electronic and Technology of Information (LETI)
`National School of Engineers of Sfax, BP W 3038, Sfax, Tunisia
`Email: dallali_a@voila.fr
`
`
`ABSTRACT
`The classification of the electrocardiogram registration (ECG) into different pathologies disease devises is a
`complex pattern recognition task. The registered signal can be decomposed into three components, QRS complex, P and T
`waves. The QRS complex represent the reference for the other ECG parameters; the width and amplitude QRS have more
`important to identify the ECG pathologies. The statistical analysis of the ECG indicate that they differ significantly
`between normal and abnormal heart rhythm, then, it can be useful in detection of ECG arrhythmia. The traditional methods
`of diagnosis and classification present some inconvenient; seen that the precision of credit note one diagnosis exact
`depends on the cardiologist experience and the rate of concentration. Due to the high mortality rate of heart diseases, early
`detection and precise discrimination of ECG arrhythmia is essential for the treatment of patients. During the recording of
`ECG signal, different form of noises can be superimposed in the useful signal. The pre-treatment of ECG imposes the
`suppression of these perturbation signals, three methods for the noisily of signals are used; temporal, frequency, and time
`frequency method filter. The features are extracted from wavelet decomposition of ECG signal intensity. The inclusion of
`Artificial Neural Network (ANN) based on feed forward back propagation with momentum, in the diagnostic and
`classification of ECG pathologies have very important yield [1, 2]. The four parameters considered for ECG arrhythmia
`classification are the interval RR, the QRS width, the QRS amplitude, and the frequency of appears QRS. Due to the large
`amount of input data, needed to the classifier, the parameters are grouped in batches introduced to artificial neural network.
`The classification accuracy of the ANNs introduced classifier up to 90.5% was achieved, and a 99.5% of sensitivity.
`
`Keywords: cardiac pathologies, ECG, heart rate variability, wavelet transform, ANNs, classification.
`
`INTRODUCTION
`In
`recent years, computer assisted ECG
`interpretation has played an important role in automatic
`diagnosis of heart anomalies [1, 3]. The wave forms of
`ECG; width reflects the physical condition of human heart,
`is the most biological signal to study and diagnosis cardiac
`dysfunctions. So, it is important to record the patient’s
`ECG for a long period of time for clinical diagnosis. The
`clinical significance diagnosis depends on different
`parameters of ECG; complex QRS, wave P, frequency,
`Heart Rate Variability R-R. In these applications, it is
`more important to develop signal processing methods that
`permit real time feature extraction and de - noising of the
`ECG characteristic. The extracted parameters are used for
`the classification of the cardiac pathologies and make an
`automatic tool of diagnosis in the services of doctors
`before the arrival of a quantified patient. Many techniques
`were used for the diagnosis of ECG signal; temporal
`methods [4, 5], frequency method [4] and time frequency
`methods [5, 6].
`The real time records of ECGs are accompanied
`by a high frequency signals that superposed with the
`
`
`
`
`
`
`
`
`
`useful ECG. The suppression of these perturbation signals
`is necessary to a performance classifier system. The ECG
`data must be filtered in order to attenuate undesired
`electrical components of ECG. Over recent years, wavelets
`transforms play an increasing role in the pre-processing
`medical signal. The ECG signals are filtered by band pass
`filters based and discrete wavelet transform.
`In the recent years, various algorithms are
`developed for classification and identification of the ECG
`anomalies. These algorithms are most based in fuzzy logic
`and Neural Network techniques. The remaining of the
`paper is organized as follows: The first stage, point out to
`the materials and methods used. In this stage, we present
`the ECG signal and their significant parameters for
`diagnostic. In the second stage, time and frequency
`domain are applied to de-noising ECG signal and extract
`the corresponding features. The extracted features are used
`to train an ANNs for classification of different anomalies
`is will be treated in third stage. The simulation results of
`the neural network classifier will be discussed at the end
`of the paper.
`
`
`74
`
`2
`
`
`
` VOL. 6, NO. 5, MAY 2011
`ARPN Journal of Engineering and Applied Sciences
`
`©2006-2011 Asian Research Publishing Network (ARPN). All rights reserved.
`
` ISSN 1819-6608
`
`www.arpnjournals.com
`
`
`
`f
`
`Codage
`
`Archivage o
`ECG data
`
`COMPRESSION
`
`Transmission with
`the aim of remote
`consultation
`
`ECG
`
`The electrocardiography
`is the fundamental tool
`for the diagnosis and
`the detection of
`cardiac anomalies
`
`Transformation
`
`Parameters
`
`Automatic
`Diagnostic
`
`ANOMALIE
`
`
`
`
`
`Quantification
`
`
`
`
`
`
`
`Automatic
`
`Diagnostic
`
`
`
`Figure-1. Overview of fully implantable neural recording system using a thermoelectric power.
`
`
`
`
`
`
`
`
`Characteristics of the ECG
`the wave’s electrical
`The ECG
`represents
`propagation through the respective regions of the heart
`(SA. node, Arial Muscle, AV node, Atria ventricular
`Bundle, Left and Right Bundle Branches). These waves
`are the major evident observable of the human heart and
`
`
`have been used to intensive diagnosis since of their
`significance in the context of pathologies [8]. Usually, the
`listing of the electrical wave’s variations on the papers
`constitutes the ECG signal. Figure-2 shows the temporal
`characteristics of normal ECG.
`
`Associated
`wave
`P wave
`
`Mechanical
`actions
`Auricular
`depolarization
`Depolarization
`of the ventricle
`Repolarization
`of the ventricles
`Repolarization of the auricles
`
`QRS
`Complex
`
`T wave
`
`Table-1. ECG properties.
`Amplitude (mV) Wave frequency
`Duration
`(Hz)
`(sec)
`<0.12
`10
`
`(cid:148) 0.3
`
`0.08 à 0.12
`
`Q<0 - S>0 R (0.5-2)
`DI + DII+ DIII > 15
`
`0.2
`
`0.2
`
`20 - 50
`
`5
`
`Hidden wave
`
`Axe
`
`20° à 80°
`-30° à +110°
`< -30° axe gauche
`> 110° axe droit
`
`
`
`
`The analysis of the ECG morphologic (P wave,
`QRS wave T wave…) is essential in diagnosis. The Table-
`1 summarizes the electric properties of a normal ECG. It is
`know that electrocardiogram signals ECGs are used
`extensively
`in different monitoring and diagnostic
`cardiology applications [9, 10]. So, a Holter monitor
`produces a large amount of non-stationary and quasi
`periodic data; example of noise that can be superposed on
`the useful signal, which are difficult to classify directly the
`ECG frames. Many methods have been proposed to solve
`the problem.
`
`REVIEW OF LITERATURE
`Sokolow et al., (1990), indicate that the state of
`cardiac health is generally reflected in the shape of the
`ECG waveform and heart rate. Cuiwei Li et al., (1995)
`showed that it is easy with multi scale information /
`decomposition in wavelets transformation to characterize
`the ECG waves. Khadra et al., (1997) proposed a
`classification of life threatening cardiac arrhythmias using
`
`wavelet transforms. MG Tsipouras et al., (2004) used time
`frequency
`analysis
`for
`classification
`of
`atrial
`tachyarrhythmias. Later, Al-Fahoum and Howit (1999)
`joint
`radial basis neural networks
`to wavelet
`transformation to classify cardiac arrhythmias. Weissan et
`al., (1990); Akselord et al., (1981); Pomeranz et al.,
`(1985) showed that the spectral analysis is the essential
`linear techniques used for the HRV signals analysis. Silipo
`et al., (1998) has shown that the Ann’s approach is shown
`to be capable of dealing with the ambiguous nature of the
`ECG signal when tested and compared with the most
`common traditional ECG analysis on appropriate data
`bases [11]. Ali Shahidi Zandi et al., (2005) used a method
`based on the continuous Wavelet transform and Artificial
`Neural Network for detection of ventricular late potentials
`in High-Resolution ECG signals. Mei Jiang Kong et al.,
`(2005) used block-based neural networks to classify ECG
`Signals. Fira et al., (2008) proposed an ECG compressed
`technique and its validation using NN’s. The choice of the
`wavelet family as well as the selection of the analyzing
`
`
`75
`
`3
`
`
`
` VOL. 6, NO. 5, MAY 2011
`ARPN Journal of Engineering and Applied Sciences
`
`©2006-2011 Asian Research Publishing Network (ARPN). All rights reserved.
`
` ISSN 1819-6608
`
`www.arpnjournals.com
`
`
`
`
`function and level decomposition into these families have
`been discussed
`to
`the Daubechies decompositions
`provided by the Daubechies wavelet (level 3), the coiflet
`wavelet (level 6) and the symetric wavelet (level 6) [12].
`
`
`
`
`Figure-2. Temporal characteristics of normal ECG.
`
`
`MATERIALS AND METHODS
`
`ECG database
`The ECG recording from MIT_BIH arrhythmia
`database was studied. Each recording has duration of 30
`minutes and includes two leads. The sampling frequency is
`360 Hz and the resolution is 1200 samples per 1 mV.
`
`Processing
`
`ECG signals can be contaminated with several
`kinds of noise, such as power line interference (A/C),
`baseline wandering (BW), and electromyographic noise
`(EMG), which can affect the extraction of parameters. The
`processing of the ECG recorded signal was consistent the
`suppression of
`these perturbation signals;
`the high
`frequency noise and the low frequency drift.
`
`
`
`In the present work, heart rate variability is used as the
`base signal for classification of cardiac abnormalities into
`three classes. Four parameters extracted from the cardiac
`signals are used for the proposed classification.
`
`
`
`Low and high pass filter for drift, high frequency and
`line base suppression.
`Time frequency methods, based on the Discrete
`Wavelet Transform
`(DWT)
`and
`thresholding
`coefficients [13, 14] are applied to de-noising ECG
`signals; the algorithm for de-noising ECG by DWT is
`to decompose the signal in approximation and details
`coefficients.
`
`(cid:131)
`
`(cid:131)
`
`
`
`with
`
`
`
`
`
`(1)
`
`
`(2)
`
`
`
`
`
`
`
`
`ECG
`
`
`
`h(n)
`
` 2
`
`g(n)
`
` 2
`
`h(n)
`
`g(n)
`
` 2
`
` 2
`
`h(n)
`
`g(n)
`
` 2
`
` 2
`
`
`
`
`
`
`
`
`
`
`
`
`
`Figure-3. Multi-resolution analysis: decomposition of ECG signal.
`
`
`Parameters detection
`The first feature extracted from the ECG recorded
`is the R wave. Initially, a point in the QRS complex is
`detected (max of QRS). Then, the wave of the QRS
`
`complex (R wave) is identified in the window [R_wave -
`280 ms, R wave + 120 ms]. A wave P detection, using the
`algorithm proposed by Pan and Tompkins [15]. The RR-
`interval signal is constructed by measuring the time
`
`
`76
`
`4
`
`
`
` VOL. 6, NO. 5, MAY 2011
`ARPN Journal of Engineering and Applied Sciences
`
`©2006-2011 Asian Research Publishing Network (ARPN). All rights reserved.
`
` ISSN 1819-6608
`
`
`interval between successive R waves. The frequency of the
`ECG signal of a normal subject is approximately 60 Hz
`and can go up 130 Hz for an abnormal patient. The block
`diagram of
`the proposed method
`for ECG beat
`
`
`www.arpnjournals.com
`
`
`
`classification is depicted in Figure-4. The method is
`divided into three steps: (1) prepossessing (2) extraction of
`parameters and (3) classification by ANNs.
`
`P wave
`detected
`
`Frequency
`
`Extraction
`
`ANNs
`
`Classification
`
`
`
`ECG
`
`
`
`
`Preprocessing
`
`
`Pre-processed ECG
`
`di
`
`
`R wave detection
`
`
`(RR_interval)
`
`RR_interval duration
`
`
`QRS complex
`
`
`detection
`
`
`
`Parameters for
`
`
`classification
`
`
`
`
`
`
`Figure-4. Block diagram of the proposed scheme for ECG parameters
`extraction and classification.
`
`
`
`
`
`
`
`
`FILTERING OF THE ECGs SIGNALS
`
`Monitoring of
`the electrocardiogram signal
`during normal activity using Holter devices has become
`standard of cardiac arrhythmias. The most important
`problems in real _ time ECG recording are [12, 13]:
`(cid:131) Muscle noise
`Power line interference (50 or 60 Hz noise induced by
`(cid:131)
`lines)
`(cid:131) Base line wander ( a very low frequency change of
`iso-electric level of ECG)
`
`(cid:131) Artifacts due to electric motion
`Physiological variability of QRS complex
`(cid:131)
`
`
`The pre-treatment of ECG signals imposes the
`extraction of the useful ECG signal from noisily ECG
`signal.
`
`Temporal filtration
`The temporal methods of filtration are based on
`low and high pass filters in cascade.
`
`ECG
`
`Low pass filter
`
`Xi
`
`
`
`
`
`High pass filter
`
`Xfiltre
`
`Figure-5. Cascade temporal filter.
`
`
`
`
`
`There are many power spectrum features were
`extracted from the ECG signal at frequency interval (4 to
`30 Hz), shown in Figure-6. The term power spectrum
`
`
`means the amount of power per unit of frequency as a
`function of the frequency.
`
`
`77
`
`5
`
`
`
` VOL. 6, NO. 5, MAY 2011
`ARPN Journal of Engineering and Applied Sciences
`
`©2006-2011 Asian Research Publishing Network (ARPN). All rights reserved.
`
` ISSN 1819-6608
`
`www.arpnjournals.com
`
`
`Figure-6. Power spectral density of signal MIT_203 (a) original signal (b) filtered signal.
`
`
`
`
`
`Original signal
`WT filter signal
`
`500
`
`1000
`
`1500
`
`2000
`
`2500
`
`3000
`
`3500
`
`4000
`
`4500
`
`5000
`
`WT filter error
`
`012
`
`-1
`
`-2
`
`0
`
`0.2
`
`0.1
`
`0
`
`-0.1
`
`-0.2
`
`
`
`0
`100
`200
`300
`400
`500
`600
`700
`800
`900
`1000
`Figure-7. Signal ECG of MIT 203 presenting a deviated
`line base (a) original signal filtered signal by classic filter
`(b) original signal filtered by wavelet transform filter.
`
`
`NON LINEAR ANALYSIS
`The cardiovascular system is too complex to be
`linear, and treating it as a non-linear system can lead to
`better understanding of the system dynamics. To study this
`system, we utilized nonlinear parameters as follows:
`
`SD1/SD2
`Poincare plot is a graphical representation of the
`correlation between successive RR intervals. The Poincare
`plot may be analyzed quantitatively by calculating the
`standard deviation of the distances of the RR (j) to the line
` and
`. The standard
`deviation are referred respectively to; the fast beat-to-beat
`variability of RR (j), and the description of the longer-term
`variability of (RR (j). The ratio SD1/SD2 is to describe the
`relation between these components.
`
`
`
`78
`
`
`
`
`
`012
`
`
`Time frequency filtered
`For many years now, wavelets have proved their
`efficiency in various applications. Signal processing is
`performed using the wavelet equivalent filters. The multi-
`scale feature of wavelet transform is applied to distingue
`the noise, baseline drift and artifacts. The choice of
`wavelet family as well as the selection of the function
`analyzed. Daubechies wavelet family presents the similar
`shape of the QRS complex and their energy spectrums are
`focalized around
`low
`frequencies.
`In
`this work,
`Daubechies wavelet at level three‘db3’ is used. We set the
`, this value is fixed for all the earlier
`threshold
`analysis with the wavelet function. The wavelet transform
`algorithm adopted for de-noising the ECG recorded is
`based to decompose the signal ECG in approximations and
`details coefficients [15]. Figure-7 illustrates the filtration
`of the ECG record MIT_203 using wavelet transform and
`temporal classic filter. It is noticed that the line base is
`obtained and the high frequency noise is eliminated.
`
`
`Original signal
`Filtred signal
`
`-1
`
`-2
`
`0
`
`500
`
`1000
`
`1500
`
`2000
`
`2500
`
`3000
`
`3500
`
`4000
`
`4500
`
`5000
`
`Classic filter error
`
`0
`
`100
`
`200
`
`300
`
`400
`
`500
`
`600
`
`700
`
`800
`
`900
`
`1000
`
`
`
`0.2
`
`0.1
`
`0
`
`-0.1
`
`-0.2
`
`
`
`6
`
`
`
` VOL. 6, NO. 5, MAY 2011
`ARPN Journal of Engineering and Applied Sciences
`
`©2006-2011 Asian Research Publishing Network (ARPN). All rights reserved.
`
` ISSN 1819-6608
`
`www.arpnjournals.com
`
`
`
`
`DFA
`
`The Detrended Fluctuation Analysis (DFA) is
`used to quantify the fractal scaling properties of short
`interval R-R
`interval signals. This
`technique
`is a
`modification of the root-mean-square analysis of random
`walks applied to non stationary signals [16]. The root-
`mean-square fluctuation of an integrated and detrended
`time series is measured at different observation windows
`and plotted against the size of the observation window on
`a log-log scale. The root-mean-square fluctuation of this
`integrated and detrended series is calculated using the
`equation:
`
`
`
`(3)
`
`
`
`Spectral entropy
`spectral
`the
`quantifies
`entropy
`Spectral
`complexity of the time series. Application of Shannon’s
`channel entropy gives an estimate of the spectral entropy
`of the process, where entropy is given by:
` (4)
`
`
`
`
`
`Where pf is the PDF value at frequency f. The spectral
`entropy H (0 <H <1) describes the complexity of the HRV
`
`EC
`
`Filtration
`
`Parameters extraction
`
`NEURAL
`
`NETWORK
`
`(ANN)
`
`signal. This spectral entropy H was computed for the
`various types of cardiac signal.
`
`ARTIFICIAL
`CLASSIFIER
`The Artificial Neural Network (ANN) is has to be
`the structure of discreet propagation “feed
`applied,
`forward” is used in the training stage of the ANNs. In this
`paper, we are interested in the classification of the
`arrhythmias presenting some anomalies. All the ECG data,
`used from the MIT-BIH Arrhythmia Database which was
`digitized at a sampling rate of 360 Hz [17].
`Artificial Neural Network is biologically inspired
`network that are suitable for classification of biomedical
`data. A combination of wavelets transform and ANNs is
`proposed to classify cardiac arrhythmias [18-24]. The
`precision of classification results of the anomalies depends
`on the number of parameters selected; the number of
`neurons of input layer is equals to the numbers of
`parameters used
`for classification. The parameters
`extracted are used to train the ANNs. Typically, for
`classification,
`the configuration usually used are
`multilayer feed forward neural networks with Log-sigmoid
`activation function that using the generalized back
`propagation for training which minimize the squared error
`between the desired outputs and the actual outputs of the
`ANNs. The desired output is being a real number in the
`interval (0 – 1).
`
`ANNs structure
`
`Decision
`
`ECG
`classifie
`
`
`
`
`
`The classification steps of the signals are depicted
`in Figure-8. One distinguishes the stage of filtering, the
`stage of extraction of the parameters and the stage of
`neuronal classification (training and validation of the
`method). The architecture of the ANN contains: four
`inputs neurons, two hidden layer with eight neurons and
`one output neurons Figure-10. The training of the artificial
`neural network ends if the sum of the square errors for all
`segments is less than 0.01. The number of data set used for
`training and testing of the ANNs classifier and the results
`obtained are
`tabulated
`in Table-2. The parameters
`extracted (P-wave, HRV, QRS and, Frequency), are used
`as inputs vector to ANNs classification. The output of the
`classifier is a graphical representation Figure-11.
`
`
`Pre-processing
`
`Wavelet Transform
`ECG Classifier
`Figure-8. Classification of ECG by artificial neural networks.
`
`A dynamical analysis of heart rate behaviour
`derived from non-linear mathematics can reveal abnormal
`patterns of RR interval dynamics which cannot be detected
`by commonly employed moment statistics of heart rate
`variability. The HRV signal can be used as a reliable
`indicator of heart diseases (Figure-9).
`
`
`
`
`
`
`
`
`
`0.35
`
`0.3
`
`0.25
`
`0.2
`
`0.15
`
`0.1
`
`0.05
`
`
`
`0
`
`0
`10
`20
`30
`40
`50
`60
`70
`80
`Figure-9. R peak detection of ECG records used
`for classification.
`
`90
`
`
`
`Figure-10. Three layer feed forward neural network.
`
`
`
`
`79
`
`7
`
`
`
` VOL. 6, NO. 5, MAY 2011
`ARPN Journal of Engineering and Applied Sciences
`
`©2006-2011 Asian Research Publishing Network (ARPN). All rights reserved.
`
` ISSN 1819-6608
`
`
`SIMULATION AND RESULTS
`To evaluate the performance of classifier; three
`criteria are used that defined below:
`
`
`
`www.arpnjournals.com
`
`
`
`The performance of the combination between wavelet
`transform and neural network for classification are
`considered to be acceptable comparatively with the
`manual diagnostic. These approaches can constituent a
`tool for the diagnostic and the classification of the normal
`and abnormal cardiac signal. The three types of ECG
`signals are normal signal ECG (Norm), Left bundle branch
`block beat (Lbbb) and, Right bundle branch block beat
`(Rbbb). For
`training
`the neural network a back-
`propagation algorithm with gradient descent and cross-
`validation was used.
`The patterns used in the three sets above were
`distinct. The records no. 100, 101,104, 105, 106, 107, 108,
`109, 111, 112, 113, 203, 207, 212, 214, 231 were used for
`training, testing and validation. The error function used
`was the mean square error and the stop criterion was the
`least error for the validation set. The ANN algorithm has
`the advantage of not requiring information on the class
`statistics; it can be easily implemented and has a small
`probability of error.
`
`
`
`
`
`
`
` (5)
`
` (6)
`
` (7)
`
` (8)
`
`
`
` (9)
`The results of classification using the partition
`neuronal classifier are summarized in Table-2. In this case,
`the accuracy and sensitivity are higher and the rate of false
`classification is weaker than total signal classified (6.2%).
`
`
`
`
`Class
`
`1 Norm
`2 Lbbb
`3 Rbbb
`Total
`
`Data sets
`testing
`60
`35
`20
`129
`
`Sets
`misclassified
`3
`2
`2
`93.79
`
`Accuracy
`(%)
`
`Sensitivity
`(%)
`
`95.00
`94.28
`90.00
`93.03
`
`100
`100
`95.24
`96.94
`
`Entropy
`(%)
`
`92.80
`87.49
`76.14
`71.09
`
`Table-2. Classification of cardiac arrhythmia using ANNs.
`Sets
`correctly
`classified
`57
`33
`18
`121
`
`
`
`the previously, we extracted
`As well as
`characteristic parameters of the ECG signals for various
`pathologies. These decomposition parameters constitute a
`data base for the learning ANNs. The back-propagation
`neural network (BPNN) used in this study is a three-layer
`feed-forward structure. In order to overcome the difficulty
`of intensive computational time taken using ANNs
`classifier, attempt has been made to reduce the numbers of
`input data parameters using WT which is beneficial for
`ECG decomposition. The objective of using WT is to
`increase the numbers of points of ECG and extract the
`significant parameters for automatic and fast ECG beat
`classification
`
`DISCUSSION AND CONCLUSIONS
`The HRV signal, frequency, P and QRS waves
`can be used as reliable indicators of heart diseases. In this
`paper, both Wavelet Transform and Neural Network
`classifier are presented as diagnostic tools to aid the
`physician in the analysis of heart diseases. Three types of
`ECG samples were selected from MIT - BIH arrhythmia
`database for experiments. The aim of using Artificial
`Neural Networks (ANN) is to decrease the error by
`
`
`grouping similar parameters training data. The features of
`obtained training parameters are extracted using wavelet
`transform (WT). The ANN - WT has been presented and
`developed to classify electrocardiography signals. Further
`it is observed that the % error is also less (Figure-11). The
`technique used is obtained by incorporating the ANNs,
`Wavelet transform and significant parameters; extracted
`from the ECG records such as: Heart Rate Variability
`(HRV, P-wave, QRS, Frequency), combining
`their
`advantages. The analysis of the results listed in Table-2
`show it is evident that the classifiers presented are
`effective for classification of cardiac arrhythmia with an
`overall accuracy of 93.03%. The accuracy of the tools
`depends on several factors, such as the size of database
`and the quality of the training set and, the parameters
`chosen to represent the input vector of the classifier. The
`single output neuron, allowing to easily classification
`according the abnormalities ECG signal, is used. In
`bottom, the error curve illustrates the smallest values
`obtained (less than 2%). The results conclude that it is
`possible to classify the cardiac arrhythmia with the help of
`neural networks. The advantage of the ANNs classifier is
`its simplicity and ease of implementation.
`
`
`80
`
`8
`
`
`
` VOL. 6, NO. 5, MAY 2011
`ARPN Journal of Engineering and Applied Sciences
`
`©2006-2011 Asian Research Publishing Network (ARPN). All rights reserved.
`
` ISSN 1819-6608
`
`www.arpnjournals.com
`
`
`
`
`
`Figure-11. The output of ANNs classifiers.
`
`
`
`
`
`
`
`
`
`
`REFERENCES
`
`[1] M.G. Tsipouras, V.P. Oikonomou, D.I. Fotiadis, L.K.
`Michlis, D. Sideris. 2004. Classification of Atrial
`Tachyrrhythmias in Electrocardiograms Using Time
`Frequency Analysis. Computers in Cardiology. 31:
`245-248. 2004 IEEE.
`
`[2] B. Anuradha, V. Reddy. 2008. Cardiac Arhythmia
`Classification Using Fuzzy Classifiers. Journal of
`Theoretical and Applied Information Technology.
`
`[3] M.G. Tsipouras, C. Voglis, I. E Lagaris and D. I.
`Fotiadisohet. 2005. Cardiac Arrhythmia Classification
`Using Support Vector Machines.
`
`[4] Fokapu O. and Girard J.P. 1994. Evolution temporelle
`et fréquentielle de l'ECG: analyse battement par
`battement, ITBM. 14(1): 102-112.
`
`[5] B. Anuradha, K. Suresh Kumar and V. C. Veera
`Reddy. 2008. Classification Of Cardiac Signals Using
`Time Domain Methods. ARPN Journal of Engineering
`and Applied Sciences, Tirupati, India.
`
`[6] Chouakri S. A., Bereksi-Reguig F., Ahmaidi S. and
`Fokapu O. 2005. Wavelet Denoising of The ECG
`Signal Based on Noise Estimation. 5th International
`ISAAC Congress, July 25-30, Catania, Sicily, Italy.
`
`[7] Chouakri S. A., Bereksi-Reguig F., Ahmaïdi S. and
`Fokapu O. 2005. Wavelet Denoising of
`the
`Electrocardiogram Signal Based on the Corrupted
`Noise Estimation. Computers in Cardiology. 32:
`1021-1024.
`
`[8] Cuiwei Li, Chongxum Zheng and Changfeng Tai.
`1995. Detection of ECG characteristic points using
`wavelets
`transforms.
`IEEE Transactions
`on
`Biomedical Engineering. 42(1): 21-28.
`
`[9] Kaoua M. Trighidet J. P. Herbeuval E. Yvrovd et Y.
`Flamant: La trasformée en Ondelette Appliquée a
`l’analyse des Signaux Cardiaques ;
`JTE96,8-9
`Novembre 1996, Hammamet, Nabeul, Tunisia.
`
`[10] Gley Kheder, Abdennaceur Kachouri and Mounir
`Samet. 2008. HRV analysis using wavelet package
`transform and Least Square Support Vector Machine;
`International Journal Of Circuits, Systems And Signal
`Processing. 2(1).
`
`
`[11] Senhadi L, Carrault G, Bellanger J. J. and Passariello
`G. 1995. Comparing Wavelet Transforms
`for
`Recognizing Cardiac Patterns. IEEE Engineering in
`Medicine and Biology Magazine. 14(2): 167-173.
`
`[12] A. Kachouri, M. Ben Messaoud and A. Dallali. 2009.
`Wavelet based on Electrocardiogram Signal Analysis
`for Classification and Diabnosis By Neural Networks.
`SSD’03. March 26-28. Sousse, Tunisia.
`
`
`
`
`81
`
`9
`
`
`
` VOL. 6, NO. 5, MAY 2011
`ARPN Journal of Engineering and Applied Sciences
`
`©2006-2011 Asian Research Publishing Network (ARPN). All rights reserved.
`
` ISSN 1819-6608
`
`www.arpnjournals.com
`
`
`
`
`[13] Khadra L., Al Fahoum A. S. and AL-Nashash H.
`1997. Detection
`of
`life-threatening
`cardiac
`arrhythmias using wavelet transformation. Med. Biol.
`Eng. Comput. 35: 626-632.
`
`[14]Afonso V. X., Tompkins W. J., Nguyen T. Q. and Luo
`S. 1999. ECG beat detection using filter banks. IEEE
`Transactions on Biomedical Engineering. 46(2): 192-
`202.
`
`[15]Pan J, Thompkins W. J. 1985. A real-time QRS
`detection algorithm. IEEE Trans. Biom. Eng. 32: 230-
`236.
`
`[26]Liviu Goras, Monica Fira. 2009. Preprocessing
`Method For Improving Ecg Signal Classification And
`Compression Validation. Physicon, Catania, Italy,
`September, 4.
`
`
`[27]Anupam Das, J. Maiti and R.N. Banerjee. 2009.
`Process control strategies for a steel making furnace
`using ANN with bayesian regularization and ANFIS.
`Expert Systems with Applications.
`
`
`[28]http://www.physionet.org/physiobank/database/mitdb/
`
`[29] http://www.physionet.org/physiobank/database/slpdb/
`
`
`
`
`[16]B. Anuradha and V.C. Veera Reddy. 2008. ANN for
`Classification of Cardiac Arrhythmias. ARPN Journal
`of Engineering and Applied Sciences. 3(3): 1-6.
`
`[17]Acharya R., Bhat P. S., Iyengar S. S., Roo A. and Dua
`S. 2002. Classification of heart rate data using neural
`network and fuzzy equivalence relation. The Journal
`of the Pattern Recognition Society.
`
`Introduction
`[18]Margarita Sordo. 2002.
`Networks in Healthcare-A review.
`
`[19]Silipo R. and Marchesi C. 1998. Artificial Neural
`Networks
`for automatic ECG Analysis.
`IEEE
`Transactions on signal processing. 46(5): 1417-1425.
`
`[20]AL-Fahoum A. S. and Howitt I. (199). Combined
`wavelet
`transformation and
`radial basis neural
`networks
`for classifying
`life-threatening cardiac
`arrhythmias. Med. Biol. Eng. Comput. 37: 566-573.
`
`[21]Wei Jiang Kong S. G., Peterson G. D. 2005. ECG
`signal
`classification using block-based neural
`networks. Proc. IJCNN. p. 326.
`
`[22]Nyongesa H. 2004. Classification of ECG by Auto-
`Regressive Modelling and Neural Networks. IEEE
`AFRICON. p. 841.
`
`[23]Fira M., Goras L. 2008. An ECG signals compression
`method and its validation using NN’s. IEEE Trans.
`Biomed. Eng. 45: 1319-1326.
`
`[24]R. Acharya U, A. Kumar1 P. S. Bhat, C.M. Lim, S. S.
`Iyengar, N. Kannathal, S. M. Krishnan. 2004.
`Classification of cardiac abnormalities using heart rate
`signals. Med. Biol. Eng. Comput. 42: 288-293.
`
`[25]Ali Shahidi zandi, Mohammad Hasan Moradi. 2005.
`Detection of Ventricular Late Potentials in High-
`Resolution ECG Signals by a Method Based on the
`Continuous Wavelet Transform and Artificial Neural
`Networks.
`
`to Neural
`
`
`
`
`
`
`View publication statsView publication stats
`
`
`82
`
`10
`
`