M.Vijayavanan et al.

/ International Journal of Computer Science & Engineering Technology (IJCSET)

Automatic Classification of ECG Signal for

Heart Disease Diagnosis using
morphological features
M.Vijayavanan
PG Scholar: Dept of Computer Science & Engineering
Annamalai University
Annamalai Nagar, Tamilnadu, India
[email protected]

V.Rathikarani
Assistant Professor: Dept of Computer Science & Engineering
Annamalai University
Annamalai Nagar, Tamilnadu, India
[email protected]

Dr. P. Dhanalakshmi
Associate Professor: Dept of Computer Science & Engineering
Annamalai University
Annamalai Nagar, Tamilnadu, India
[email protected]
Abstract— An Electrocardiogram (ECG) is a test that records the electrical activity of the heart to locate
the abnormalities. Automatic ECG classification is an emerging tool for the cardiologists in medical
diagnosis for effective treatments. In this paper, we propose efficient techniques to automatically classify
the ECG signals into normal and arrhythmia affected (abnormal) category. For these categories
morphological features are extracted to exemplify the ECG signal. Probabilistic neural network (PNN) is
the modeling technique engaged to capture the distribution of the feature vectors for classification and
the performance is calculated. ECG time series signals in this work are collected from MIT-BIH
arrhythmia database. The proposed an accurately classify and discriminate the difference between
normal ECG signal and arrhythmia affected signal with 96.5% accuracy.
Keywords- Electrocardiogram (ECG), Cardiac Arrhythmia, Discrete Wavelet Transform (DWT),
Morphological features, Probabilistic Neural Network(PNN), Massachusetts Institute of Technology - Boston's
Beth Israel Hospital (MIT-BIH).
I. INTRODUCTION
Heart is the most imperative organ of human body. According to World Health Organization (WHO)
cardiovascular disease (CVD) is the major disease. which cause more people to die every year. Each year 9.4
million deaths occur each year, in this 45% deaths occur due to coronary heart disease. Traditional techniques of
Visual analysis of ECG for doctors are complex and time consuming task. Visual analysis needs experience to
identify the problems in ECG [1]. This paves a way for Computerized ECG. In computerized ECG the automatic
classification of heart disease into normal and abnormal is done in an automated manner.
Electrocardiogram is a test that is performed on the heart to detect the abnormalities in the cardiac cycle by
analyzing the electrical activities of the human heart by placing 12 lead electrodes over the human surface. Thus
each electrode specify some important morphological characteristics in the ECG signal. Thus for evaluating
arrhythmia leads such as II, III, AVR and V5 electrodes. Thus for arrhythmia analysis the abnormalities of the p
wave in the above listed electrode are to be analyzed. ECG pattern and heart rate variability may have to be
observed over several hours[2]. Since visual analysis of long-term recordings of the heart activity, with more
than 100,000 ECG beats in a single recording, automated computer analysis is of major significance [6]. It is a
challenge to extract the most common and most important theme from amorphous raw ECG data.
In this work Feature extraction in which the morphological features such as P, Q, R, S, T peak points, QRS
complex duration, PR interval, QT interval, ST interval, ST segment, RR interval and PR segment are extracted.
Probabilistic Neural Network is trained by using the extracted features from both Normal and Arrhythmia
categories. PNN is tested using the untrained records and the performance calculation is done.

ISSN : 2229-3345 Vol. 5 No. 04 Apr 2014 449

 M.Vijayavanan et al. / International Journal of Computer Science & Engineering Technology (IJCSET)

MIT-BIH Preprocessing Morphological Feature

Database Extraction

Normal
Probabilistic Neural Network
(PNN)
Arrhythmia

Figure 1 Block Diagram of ECG Classification

A. Electrocardiogram
ECG is a test that monitors the electrical activity of the heart for a certain period as detected by using
the electrodes attached to the outer surface of the skin and recorded by a device external to the body. The output
of each electrode is called a Lead. The 12- Lead ECG is normally taken for all the patients. It is used to calculate
the rate and regularity of the heart beats, as well as the dimension and location of the chambers and the
occurrence of any damage to the heart and the effects of drugs or devices used to regulate the heart. A normal
sample ECG is shown in Fig.2. An ECG wave consists of P, Q, R, S, T waves. P wave is the Atria
depolarization (Atrial Contraction), T wave is the depolarization of ventricles (Ventricular relaxation). QRS
complex is the depolarization of ventricles (Ventricular contraction).
The QRS complex, ST segment, PR interval, RR interval, PR segment, QT interval is the most
important regions in an ECG signal for the diagnosis of different cardiac diseases.

Figure 2 ECG sample

B. Cardiac Arrhythmia
Cardiac Arrhythmia is the irregular heartbeat or abnormal rhythm of the heart which may cause
permanent damage to the heart. fundamentally there are two types of Arrhythmia called Ventricular and
Supraventricular. Ventricular arrhythmia happens in the heart's two lower chambers called Ventricles and
Supraventricular arrhythmia happens in the heart's upper chambers call Atrium. Depend upon the heart rate
arrhythmia is classified into two types Bradycardia, Tachycardia and Fibrillated. An arrhythmia affected heart
beat can be very speedy, very time-consuming and uncoordinated. The normal range of heart beat is 60-100
beats per minute. Bradycardia is slow heart rate in which the heart beats less than 60 beats per minute.

ISSN : 2229-3345 Vol. 5 No. 04 Apr 2014 450

 M.Vijayavanan et al. / International Journal of Computer Science & Engineering Technology (IJCSET)

Tachycardia is high-speed heart rate in which the heart beats more than 100 beats per minute. Fibrillation is a
fast uncoordinated heart beat.
II. LITERATURE REVIEW
The paper [3] described an automatic classification system based on combination of diverse features for
the purpose of automatic heartbeat recognition. In all stages, the classifiers are based on the k-nearest neighbor
algorithm. The classification accuracy of the proposed system is measured by the following sensitivity,
selectivity, and specificity. The system was classified 16 heartbeat types. Their system provided a sensitivity of
85.59%, selectivity of 95.46%, and an overall average specificity of 99.56%.The paper [4] investigated the use
of CWT (Continues Wavelet Transform) with two powerful data transformation techniques DWT (Discrete
Wavelet Transform), and DCT (Discrete Cosine Transforms proposed and compared. The classifiers under
examination are MLP (Multi-Layered Perceptron a conventional neural network) and SVM (Support Vector
Machine). The training algorithms used in MLP and SVM are Back Propagation (BP), Kernel–Adatron (K–A),
respectively.)[5] The paper presented a new automated diagnostic system to classification of electrocardiogram
(ECG) arrhythmias. The diagnostic system was executed using type-2 fuzzy c-means clustering (T2FCM)
algorithm, wavelet transform (WT) and neural network. Method of combining T2FCM and WT was used to
improve performance of neural network.
The paper [7] described a hybrid intelligent system for classification of cardiac arrhythmias is
proposed. Three methods of classification were used. Finally, a Mamdani type fuzzy inference system was used
to combine the results of several individual classifiers outputs to achieved a classification rate of 98%. The
paper [8] proposed a system for cardiac arrhythmia detection in ECGs with adaptive feature selection and
modified support vector machines (SVMs). Wavelet based coefficients and signal parameters are enumerated as
candidates for the classification of each class pair. A k-means clustered SVM classifier and a majority based
voting mechanism, was proposed to further improve the recognition rate for extremely similar classes. Their
results showed that the proposed ECG analysis approach obtained a recognition rate of 98.92%.[9] This paper
estimated, a neural network models with adaptive activation function were implemented to classify ECG
arrhythmias. NNAAF models included three types named as NNAAF-1, NNAAF-2 and NNAAf-3. These three
NNAAF models were compared with the MLP model all models tested by the ECG signals of 92 patients. The
average accuracy rate of all models in the training processing was found as 99.92%.The paper [10] proposed a
fractal QRS-complexes pattern recognition for imperative cardiac Arrhythmias Fractal dimension
transformation (FDT) was employed to adjoin the QRS-complex of ECG signals, including the fractal features
of several cardiac diseases. In this work, probabilistic neural network (PNN) classifier with morphological
features are proposed for recognizing the states of cardiac physiologic function.
III. FEATURE EXTRACTION
Feature extraction plays a vital role in any classification system. In this work the morphological features such
as P, Q, R, S, T peak points, QRS complex duration, PR interval, QT interval, ST interval, ST segment, RR
interval and PR segment are extracted from the ECG signal.
A. Preprocessing
The ECG recording extracted from the database may consist of numerous noises such as muscle contraction,
electrode displacement, patient movements or other kind of noises. Hence the signal should be preprocessed to
remove these kinds of artifacts from the ECG signal. The Discrete Wavelet Transform (DWT) is used in this
work.
1. Discrete Wavelet Transform (DWT)
The discrete wavelet transform (DWT) is an active tool for Time-Frequency analysis of signals. Recently,
Wavelets have been applied to several problems in Electro cardiology, including data compression, analysis of
ventricular late potentials, and the detection of ECG characteristic points. The Wavelet transformation is a linear
operation that decomposes the signal into a number of scales related to frequency components and analyses each
scale with a certain resolution. The WT uses a short time interval for evaluating higher frequencies and a long
time interval for lower frequencies. Due to this property, high frequency components of short duration can be
observed successfully by Wavelet Transform. One of the advantages of the Wavelet Transform is that it is able
to decompose signals at various levels, to remove baseline wandering on the ECG signal for efficient feature
extraction. A family of analyzing wavelets in the time frequency domain is obtained by applying a scaling factor
and a translation factor to the basic mother wavelet. Wavelet Transform function ψ is given by
∞
𝑤(𝑎, 𝑏) = ∫−∞ 𝑓(𝑡), 𝛹𝑎, 𝑏(𝑡)𝑑𝑡 (1)

1 𝑡−𝑏
Ψa, b(t) = � � (2)
√𝑎 𝛹∗ 𝑎

ISSN : 2229-3345 Vol. 5 No. 04 Apr 2014 451

 M.Vijayavanan et al. / International Journal of Computer Science & Engineering Technology (IJCSET)

The DWT is sufficient for most practical applications and for the reconstruction of the signal. In this work the
noise is filter done in the following way. We decompose the ECG signal into 8 levels by using DWT and
reconstruct the approximation (A8) and detail (D8) signals at level 8 as shown below [3].

Figure 3 Eight Level Wavelet Decomposition

B. Morphological Feature Extraction

The morphological features extracted are P, Q, R, S, T peak points, QRS complex duration, PR interval,
QT interval, ST interval, ST segment, RR interval and PR segment.
The R-peak has the maximum amplitude in an ECG signal. Hence the R-peak is identified by using the
Daubechies 8th level reconstructed coefficients. The heart rate for each patient is calculated by finding the
distance between two peaks (R-R). The other peaks are recognized by traversing the windowing function on
either side of R peak. The Q and S peaks are found by traversing on the left and right side of the R peak within
the specified window and locating the negative peak values. By traversing the left side of the Q peak the
maximum value is located as the P peak. Likewise by traversing the right side of the S peak, the maximum value
is located as the T peak within the window.

Figure 4 R-Peak Detection

The onset and offset of all points are calculated. Depending upon these data points the morphological features
such as P, Q, R, S, T peak points, QRS complex duration, PR interval, QT interval, ST interval, ST segment, RR
interval and PR segment are extracted.

ISSN : 2229-3345 Vol. 5 No. 04 Apr 2014 452

 M.Vijayavanan et al. / International Journal of Computer Science & Engineering Technology (IJCSET)

IV. CLASSIFICATION USING PNN

A probabilistic neural network (PNN) is a feed forward probability based neural network. Figure 5 shows
the overall architecture of the PNN that recognizes two distinct classes, but it can be extended to any
number K of classes. The input layer (on the left) contains N nodes: one for each of the N input features
forming the feature vector. These are fan-out nodes such that each input feature node branch to all nodes in
the hidden (pattern) layer so that each hidden node receives the complete input feature vector for each data. The
hidden nodes are connected to one node called group nodes: one group for each of the K classes as shown in
the figure 5.
Each node in a single group of hidden layer is correspond to number of observations that are taken for
training. These nodes are corresponds to a Gaussian function centered on its associated feature vector in the
kth class. There are k number of groups available in the hidden layer. All of the Gaussians in a class group
feed their functional values to the same output layer node for that class.

Figure 5 Architecture of Probabilistic Neural Network

Working principle of PNN

At the output node for class two, all of the Gaussian values for the two classes are summed and the sum is
scaled so that the overall probability volume is unity thus the sum forms a probability density function. Let there
be P pattern representing feature vectors labelled as Class 1 and similarly there be Q feature vectors labelled as
Class 2.
PNN Algorithm:
The pattern representing feature vectors have been given to form the training set. For each data in the training
set, the target represent the class to which it belongs. The following steps are present in training the PNN.
Step 1. Read in data representing data’s feature vector and the target representing the no. of classes and the
category of each data observations.
Step 2. Project these into the k groups where each group contains one class of vectors
Step 3. For each k define a Gaussian function centered on each feature vector in set k define the summed
Gaussian output function
Once the PNN is trained, then the test unknown feature vectors can be fed into it and it is tested as follows
Step 1. Read input test vector and feed it to each Gaussian function in each class
Step 2. In each group of hidden layer, calculate all nodes Gaussian functional values
Step 3. In each group of hidden nodes, feed all its Gaussian values to the single output node for that group
Step 4. At each class node in the output layer, sum all the Gaussian values which are provided as the inputs to
output layer.
Step 5. Find maximum value of all summed Gaussian values at the output node.

ISSN : 2229-3345 Vol. 5 No. 04 Apr 2014 453

 M.Vijayavanan et al. / International Journal of Computer Science & Engineering Technology (IJCSET)

V. EXPERIMENTAL RESULTS
The ECG dataset of Normal and Arrhythmia categories of different age groups both male and female
patients are collected from MIT-BIH (Massachusetts Institute of Technology- Beth Israel Hospital) database in
the recording format and it is converted into mat file. The ECG recording of about 1 hour and 2 hours durations
are taken for processing.

100
performance of PNN in %

80

60
Normal
40
Arrhythmia
20

0
5 10 15 20
Duration of Training Data in minutes

Figure 6 Performance of PNN

Since the sampling rate is 360 Hertz, the duration of training data is 5, 10, 15 and 20 minutes
respectively. The morphological features are extracted from the ECG signal of both Normal and Arrhythmia
patients. About 150 records of each category are trained by using the Probabilistic Neural Network (PNN) and
50 untrained records are given for testing the PNN. The classification result is shown in Fig 6 The performance
of the system is evaluated, and the method achieves about 96.5% classification rate.
VI. CONCLUSIONS
In this work, a system is designed to automatically categorize the ECG signals based on morphological
features. Probabilistic Neural Network (PNN) employed for classifying the normal and arrhythmia categories.
The database employed in this work for arrhythmia classification is collected from the numerous patient’s ECGs
in the MIT-BIH resources. The system showed an accuracy of 96.5% classification rate. This study proposes a
effective and reliable classification method for analyzing the ECG signals. This approach provided an
computation less method to doctors for evaluating the ECG signals. The records of ECG in the MIT-BIH
arrhythmia database are experimented to demonstrate the efficacy of the proposed method. The observation
from the result shows that Probabilistic Neural Network (PNN) with morphological features provide the better
performance. The total classification accuracy (TCA) for the experiments was about 96.5%. In future this work
can be extended to diagnose various diseases using different classifiers and different feature extraction
techniques.
REFERENCES
[1] Serkan Kiranyaz, Turker Ince, Jenni Pulkkinen, Moncef Gabbouj, “Personalized long-term ECG classification” A systematic approach
in Expert Systems with Applications, 38 (2011) 3220–3226.
[2] Mehmet Korürek, Ali Nizam, “Clustering MIT–BIH arrhythmias with Ant Colony Optimization using time domain and PCA
compressed wavelet coefficients” in Digital Signal Processing, 20 (2010) 1050–1060.
[3] Yakup Kutlu , DamlaKuntalp, “A Multi-Stage Automatic Arrhythmia Recognition And Classification System”, ELSEVIER,
Computers in Biology and Medicine 41 (2011) 37–45.
[4] Hamid Khorrami , Majid Moavenian, “A Comparative Study Of Dwt, Cwt And Dct Transformations In Ecg Arrhythmias
Classification”, Elsevier, Expert Systems With Applications 37 (2010) 5751–5757.
[5] Yuksel Ozbay , Rahime Ceylan , Bekir Karlik, “Integration Of Type-2 Fuzzy Clustering And Wavelet Transform In A Neural
Network Based Ecg Classifier” ,Elsevier, Expert Systems With Applications 38 (2011) 1004–1010
[6] L. Lu, S. Z. Li, and H. J. Zhang, “Content based audio segmentation using support vector machines”, in IEEE Int’l Conf. Multimedia
and Expo, 2001,776–779.
[7] Oscar Castillo, Patricia Melin, Eduardo Ramírez, José Soria“Hybrid Intelligent System For Cardiac Arrhythmia Classification With
Fuzzy K-Nearest Neighbors And Neural Networks Combined With A Fuzzy System” Elsevier, Expert Systems With Applications 39
(2012) 2947–2955
[8] Chia-Ping Shen A, Wen-Chung Kao , Yueh-Yiing Yang , Ming-Chai Hsu , Yuan-Ting Wuc, Feipei Lai“Detection Of Cardiac
Arrhythmia In Electrocardiograms Using Adaptive Feature Extraction And Modified Support Vector Machines” Elsevier, Expert
Systems With Applications 39 (2012) 7845–7852
[9] Yuksel Ozbay A,, Gulay Tezelb“A New Method For Classification Of Ecg Arrhythmias Using Neural Network With Adaptive
Activation Function”, Elsevier, Digital Signal Processing 20 (2010) 1040–1049
[10] Chia-Hung Lin , Yi-Chun Dub“Fractal Qrs-Complexes Pattern Recognition For Imperative Cardiac Arrhythmias” Elsevier, Digital
Signal Processing 20 (2010) 1274–1285

ISSN : 2229-3345 Vol. 5 No. 04 Apr 2014 454

 M.Vijayavanan et al. / International Journal of Computer Science & Engineering Technology (IJCSET)

[11] Amit k. mishra , shantanu raghav“local fractal dimension based ecg arrhythmia classification” elsevier, biomedical signal processing
and control 5 (2010) 114–123
[12] Abdelhamid Daamouchea, Latifa Hamamib, Naif Alajlanc, Farid Melgania“A Wavelet Optimization Approach For Ecg Signal
Classification” Elsevier, Biomedical Signal Processing And Control 7 (2012) 342– 349
[13] P. Dhanalakshmi, S. Palanivel, V. Ramalingam, “Classification of Audio Signals Using SVM and RBFNN” in Elsevier, (2009), 6069-
6075
[14] Kemal Polat, Seral Sahan and SalihGunes, “A new method to medical diagnosis: Artificial immune recognition system AIRS with
fuzzy weighted preprocessing and application to ECG arrhythmia” in Expert Systems with Applications, 31(2006) 264-269, August
[15] Yakup Kutlun and DamlaKuntalp, “A multi-stage automatic arrhythmia recognition and classification” system in Computers in
Biology and Medicine, 41(2011), 37-45, January.
[16] Mehmet Korurek and Ali Nizam, “A new arrhythmia clustering technique based on Ant Colony Optimization” in Journal of
Biomedical Informatics, 41(2008), 874-881, February.
[17] Yüksel Özbay, Rahime Ceylan and Bekir Karlik, “A fuzzy clustering neural network architecture for classification of ECG
arrhythmias” in Computers I Biology and Medicine, 36(2005),376-388,April
[18] Chia-Hung Lin, Yi-Chun Du and Tainsong Chen, “Adaptive wavelet network for multiple cardiac arrhythmias recognition” in Expert
Systems with Applications, 34(2008),2601-2611, May.
[19] Majid Moavenian and Hamid Khorrami, “A qualitative comparison of Artificial Neural Networks and Support Vector Machines in
ECG arrhythmias classification” in Expert Systems with Applications, 37(2010),3088-3093, April
[20] Hamid Khorrami and Majid Moavenian, “A comparative study of DWT, CWT and DCT transformations in ECG arrhythmias
classification” in Expert Systems with Applications, 37(2010), 5751-5757, August.
[21] Mehmet Korürek and Ali Nizam, “Clustering MIT– BIH arrhythmias with Ant Colony Optimization using time domain and PCA
compressed wavelet coefficients” in Digital Signal Processing, 20(2010), 1050-1060, July.
[22] Yu, S. N., & Chen, Y. H. (2007). “Electrocardiogram beat classification based on wavelet transformation and probabilistic neural
network” Pattern Recognition Letters, 28, 1142– 1150.

ISSN : 2229-3345 Vol. 5 No. 04 Apr 2014 455