Disease Classification using ECG Signal Based on PCA Feature along with GA & ANN Classifier

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 08 | Aug -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 739
DISEASE CLASSIFICATION USING ECG SIGNAL BASED ON PCA FEATURE ALONG
WITH GA & ANN CLASSIFIER
,
(Scholar,Ece),Bcet,Gurdaspur,Punjab,India,
(AssistantProfessor,Ece)Bcet,Gurdaspur,Punjab,India,
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract- Electrocardiogram (ECG), a non-invasivetechnique
is utilized as a primary diagnostic tool for cardiovascular
disease. Cleared ECG signals offer essential information about
the electrophysiology of the heart diseases and ischemic
modifications that may occur. Itprovidesvaluableinformation
about the functional aspects of the heart and cardiovascular
system. The purpose of this research work is to classify the
disease dataset using Genetic Algorithm and train by artificial
neural network on the basis of the features extracted and also
to test the image on the basis of the features at the database
and the features extracted of the waveform, to be tested. The
advantage of proposed method is to minimize the error rate of
the classification which occurs due to insignificant count of R-
peaks. Database from physionet.org has been used for
performance analysis. Several experiments are performed on
the test dataset and it is observed that Artificial neural
network classifies ECG beats better as compared to K-nearest
neighbor (K-NN). Precision, Recall, F-measure and accuracy
parameters are used for detecting the ECG disease. All the
simulation process will be measured in MATLAB environment.
Keywords- ECG signals, PCA (Principal component
analysis), (GA) Genetic algorithm ,(ANN)Artificialneural
network, MATLAB.
1. INTRODUCTION
1.1 ECG (Electro-Cardiogram)
Electro-Cardiogram is used to access the electrical activityof
a human heart. The diagnosis of the heart ailments by the
doctors is done by following a standard changes. In this
project our aim is to automate the above procedure so that it
leads to correct diagnosis [1].
Heart disease contains any disorder that influences the
heart’s ability to function normally. Over the last decades,
many physicists have been an increasing effort to develop
computer-based mechanical diagnostics of the ECG. TheECG
signal is created by electrical currentoftheheart.Theshapes
of the ECG waveform depend on the anatomic features ofthe
human body and heart, and thus are distinctive from one
person to another. Figure 1 illustrates the human heart and
the signals responsible for generating a normal ECG signal.
Each portion of a heartbeat produces a different deflection
on the ECG. ECG signals are recorded as a series of positive
and negative waves. The ﬁrst peak (P wave) of the normal
heartbeat is a small upward wave which indicates atrial
depolarization. Approximately 160msaftertheonsetoftheP
wave, the QRS wave is caused by ventricle depolarization.
Finally, one observes the ventricular T wave in the
electrocardiogram which represents the stage of
repolarization of the ventricles.
Fig-1: Human heart and the signals
1.2 PCA (Principle component algorithm)
PCA [2] method is just the effective feature extraction
method based on face global feature. PCA is considered as
one of the most successful linearity analysis algorithm.Itcan
reduce the dimension effectively and hold the primary
information at the same time.
1.2.1 TRADITIONAL PCA METHOD
Traditional PCA method was also named Eigen-face. The
primary principle of PCA can be simply stated as followed:

Suppose that there are a set of N training face images
{ … , } taking values in a n-dimensional image
space. The purpose of the PCA is to find a n×m linear
transformation matrix P, transforming the n-dimensional
vector YK (1 ≦ K ≦ N) to a m-dimensional, more
representative (enlarging the difference between each two
images) vector XK, where = (K=1,2…N).
Let Y’ be the average feature vector before transformation,
then X’ is the average value of all training samples. The new
average vector after transformation is as followed:
Y= (1)
Employ covariance matrix to represent the scatter degree of
all feature vectors relate to the average vector. Subtract all
the image samples to the average vector X’, then the
covariance matrix before transformation is:
(2)
In order to maximize the scatter degree of ZK , the
transformation matrix must just be the matrix which
compose by the eigen-vectors of . Because every
eigenvector has the face-like shape, they are called as eigen-
faces. Normalize all eigenvectors can eliminate the
correlation of them, and form a set of orthogonal projection
coordinates bases. The training images are seemed to be the
weighted combination of the set of orthogonal bases.
Compute the eigen values of Matrix Sx: let the eigen values
queue from large to small, then the m largest eigenvalues u1,
u2…um are obtained. Let the transformation matrix P as: P=
[u1, u2…um]T, then it can be used to transform the n-
dimensional face images to m-dimensional weightedvectors,
but also remain main information of the original images.
Through these, it can play well performance in dimension
reduction.
1.3 Genetic Algorithm(GA)
In pattern classification problem, the undesirable or
redundant features increase the complexity of the feature
space or may decrease the classification accuracy [3].
Therefore, GA is used in this work to select the optimum
number of features by eliminating the undesirable features
which in turn improves the classification performance.Here,
binary coding system as shown in Fig. 2 is used to represent
the chromosome. If bit of the chromosome is 1 then
corresponding feature is selected and if bit of the
chromosome is 0 then corresponding feature is ignored.The
description of the genetic algorithm (GA) based optimum
feature selection method is mentioned below [15]:
Step 1: Initialize the population: Initialize the number of
chromosome , where each chromosome representsbinary
bit pattern of N bits. Keep in mind is populationsizeand N
is chromosome length.
Step 2: Feature subset selection: A feature subset is
selected according to Fig. 2.
Fig-2: Chromosome bit pattern of the selected feature set.
Step 3: Fitness calculation: For each corresponding
chromosome, training data set is usedtotraintheN classifier
and fitness function is evaluated basedonmeansquare error
(MSE) values. Fitness value uniquely separates one
chromosome from others. The fitness value is calculated by
the following equation:
Fitness=0.8*(nt )-1+0.2*bct(M- )
Where, nt is a training of nte M is the selected feature
and (M−Ms)is the number of reduced features. High fitness
value is obtained from the individual chromosome with low
training nte and high number of reduced features.
Step 4: Termination criteria: Whentheterminationcriteria
(reaching of maximum generation) is satisfied, the process
ends and find the optimized solution which gives the
maximum fitness value; otherwise, it proceed with the next
generation.
Step 5: Genetic operation: In this step, the system searches
for better solutions by genetic operations which includes
selection, crossover, mutation, and replacement. In this
context, Roulette wheel selection is used to select the parent
chromosomes in the current generation based on their
fitness value. Uniform crossover operator is applied to
generate the offspring (child chromosome) and replaces the
parent chromosome which has participated in the crossover

operation depending on the value of the crossover
probability . After crossover, mutation operationisapplied
to flip the bit of the parent chromosomes based on the
mutation probability . Finally, all the chromosomes of
previous population (except best) are replaced by the
chromosomes of current generation.
1.4 Artificial Neural Network (ANN)
Our ANN model is a multilayerfeed-forwardnetwork trained
to perform classification from the neural network toolbox in
Matlab [4]. The ANN model has3distinctlayersnamelyinput
layer, hidden layer, and output layer.Theinputlayerconsists
of 800 input neurons (for a single electrode) where each
input equals to one data points from EEG data shown in
Figure.
The hidden layer consists of 10 neurons where the weights
are initialized randomly and tansig was used as the
activation function for the neurons. The number of neurons
in hidden layer was tested using a range of value ranging
from 5 to 15 neurons. The network performed marginally
better for 10 neurons in the hidden layer compared to other
tested values. There are 2 output neurons which represent
normal and depressive state. The network was trainedusing
scaled conjugate gradient back-propagation learning
algorithm which is referred to as train’scgfunctioninMatlab
.
Fig-3: Multilayer feed-forward network
Trains cg learning algorithm was used because of the
memory requirements are relatively small and yet much
faster than other standard learning algorithms available in
the toolbox. The training of the network stops when training
exceeds the maximum number of epochs (1000) or exceeds
the maximum amount of time (3600s),orwhen performance
meets the goal, or when theperformancegradientfallsbelow
minimal gradient.
2. LITERATURE REVIEW
C.R et.al, 1977, proposed the work in which noise removal
algorithm has been proposed using aseline of ECG signal. In
this way low frequency noise has been removed without
affection ST segments.
C.S.E.W, 1985, proposed an evaluation method of wave
recognition computer programs. In the end an evaluation
method for testing of ECG signal has been recommended.
And these are based on amplitude and interval details.
Islam, M. K., 2012 deals with learns and examinationof ECG
signals processing by means of MATLAB tool successfully.
Study of ECG signal involve making & simulation of ECG
signal, attainment of real time ECG data, ECG signal filtering
& giving out, feature extraction, assessment between
different ECG signal analysis algorithms & technique,
recognition of any abnormalities in ECG and calculating beat
rate and so on utilizing the majority familiar and versatile
MATLAB software beside with lab view. An utilization of
toolbox, MATLAB functions and Simulink can guide us to
work with ECG signals for allowanceandanalysisbothin real
time and by simulation with great accuracy and ease.
Das Manab Kumar, et.al, 2013 designed a classifier model
so as to categorize the beat from ECG signal of the MIT-BIH
ECG database. The classifier model comprised of three
important stages: feature extraction, selection of qualitative
features; and determination of heartbeat classes. In first
stage, features were extracted using S-transform where-as
second stage utilizes the genetic algorithm to optimize the
extracted features which represent the major information of
the ECG signal. The final-stage classifies the ECG arrhythmia.
In this study, authors have classified six types of arrhythmia
such as normal, premature ventricular contraction, atrial
premature contraction, right bundle branch block,
ventricular fusion and fusion. The experimental results
indicate that proposed methodgavebetter resultthanearlier
reported techniques.
Table-1: Comparison of existing techniques
Authors Simulat
or
Techniqu
es used
Descriptio
ns
ECG data
set
Y. Jewajinda
and P.
Chongstitvata
na (2010)
MATLAB Genetic
algorithm
and
neural
Network
Used for
online ECG
heart beat
recognition
using
feature
extraction
and
The MIT-
BIH
arrhythm
ia

classificatio
n.
J. A. Nasiri et
al. (2009)
MATLAB Genetic
algorithm
and
Support
vector
machine
Genetic
algorithm
will find the
best value
by
searching
and thus
optimize th
classificatio
n fitness
function.
The MIT-
BIH
arrhythm
ia. The
database
has 48
records
with
every
record
being an
ECG
signal for
the
period of
30
minutes.
R.
Poli et al.
(1995)
MATLAB Genetic
algorithm
GA was
used to
enhance the
QRS
complex
detection.
The MIT-
BIH
arrhythm
ia.
Acharya et al.
(2003)
MATLAB ANN and
Fuzzy
logic
Both are the
classifiers
that were
used to
diagnose
ECG signal.
Accuracy up
to 95% has
been
obtained.
The MIT-
BIH
arrhythm
ia
Ceylan,
Rahime, and
Yüksel Özbay
(2007)
MATLAB PCA and
Wavelet
transform
were used
for feature
extraction.
Fuzzy and
neural
network
were
used.
Authors
taken the
record of 92
patient in
which 40
are male
and 52 are
females of
average age
39.75 19.0
6. It was
concluded
that Fuzzy
c-mean
PCA-NN
system
perform
better than
PCA_NN
MIT-BIH
ECG
database
S. M. Jadhav
et al. (2010)
Neuro
Solution
s
(version
ANN Collected
data from
452
patientsand
UCI ECG
arrhythm
ia data
set.
5.0) calculated
different
parameters
like mean
squared
error(MSE),
receiver
operating
characterist
ics (ROC)
and area
under curve
(AUC).
Güler, İnan,
and Elif
Derya
Übeylı(2005)
MATLAB
version
6.0 with
neural
network
s
toolbox)
Neural
network
ECG signal
features
were
extracted
using
discrete
wavelet
transform
technique.
The
accuracy of
the
combined
neural
network
model was
higher than
the normal
neural
model.
MIT-BIH
ECG
database
Rai, Hari
Mohan et al.
(2013)
MATLAB
software
package
7.13.
ANNalong
with
Descrete
wavelet
transform
for
extracting
features.
Authors
took total
48 files out
of which 25
files were
from
normal
class and 20
file. Back
propagation
, Feed
forward
network
and
multilayred
preceptors
were used
as a
classifier
MIT–BIH
arrhythm
ia
database

3. Methodology
Step 1: To analyze Bradacardia and Tachycardia heart
disease, the whole process is divided into two steps named
as:
 Training phase
 Testing Phase
Flow chart-1: Proposed work Flowchart
1. Training Phase
i. Upload ECG signal dataset for Bradycardia and
Tachycardia heart disease.
ii. Extract feature from the uploaded ECG signal based
on the threshold value according to the QRS peaks.
iii. Develop a code for theGeneticalgorithmtooptimize
the features according to the objective function of
GA. s
iv. Store data in Data base.
2. Testing phase
i. Upload test waveform for testing the dieses.
ii. Analyze R peak of the QRS complex and then
optimize that by using GA.
iii. Initialize Artificial neural network (ANN) for
classification purpose.
iv. Classify the Diseases according to categories which
are generated during the training phase. And then
calculate the performance parameters.
4. Result and Discussion
This section explains the results obtained after the
implementation of the proposed work.Inthisresearchwork,
a system is developed for ECG disease that is Bradycardia
and Tachycardia recognition. For optimizing the extracted
features Genetic is used whereas, for classification ANN is
used.
Above figure shows the main window of the proposed ECG
Disease Detection System. The system is mainly categorized
into three steps: Feature Extraction & optimization, training
and testing. In the given figure there are two panels named
as Training and testing panel. After uploading the samples
for ECG signal, next step is to extract features. When we click
on QRS complex, the waveform gets displayed on the screen.
These signals are finding on the basis of thresholdtechnique.
Database (Train
and Save)
Upload ECG signal
for Bradacardia
&Techacardia
Start
Evaluate
disease &
Calculate
parameters
Training
Section
Testing
Section
Analysis of R
peak using
PCA
Reduce ‘R’
peak using
GA
optimizatio
n
Upload test
Sample
Analysis of R
peak using PCA
and Reduce R
peak using GA
optimization
Matching with ANN
train dataset

Fig-4: Main window of the proposed ECG Disease
Detection System
After extracting the QRS complex from the ECG signal. The
extracted features are improved by using optimization
algorithm known as Genetic algorithm.
Fig-5: Classification of test data using ANN
In the proposed work, the cardiac disorder was classified
into two parts named as: (i)Bradycardia (ii)Techycardia.For
effective training, it is desirable that the training data set be
uniformly spread throughout the class domains. The
available data can be used iteratively, until theerrorfunction
is reduced to a minimum.
Table-2: Performance parameters for cardiac disorder
S
No.
Precision Recall F-measure Accuracy
1 0.814 0.844 0.828 96.73
2 0.736 0.836 0.782 95.81
3 0.937 0.735 0.823 98.79
4 0.983 0.839 0.905 97.84
5 0.749 0.913 0.823 96.78
6 0.983 0.833 0.902 97.88
7 0.749 0.737 0.743 97.74
8 0.846 0.830 0.837 98.36
9 0.887 0.874 0.881 99.63
10 0.846 0.739 0.788 97.69
Chart-1: Accuracy of the ECG signal
The graph obtained for accuracy has been displayed above.
The graph has number of sample along x-axis and accuracy
value along y-axis. The average value of accuracy obtained
for the proposed work is 97.72.
5. CONCLUSION
ECG signals are used for detecting the cardiac diseases and
the improvement in ECG feature extraction has become
important for diagnosing the long recording.AnECGsignal is
a graphical representation of the cardiac movement for
computing the cardiac diseases and to ensure the
abnormalities in the heart. The objective of this research
work is to categorize the disease dataset using Genetic
algorithm and to train the Neural Network on the source of
the features extracted and moreover to test the image on the
origin of the features at the database and thefeaturesextract
of the image to be tested. This research was based on
studying the executed approaches in the ECG diseases and
then to recommend a novel practice /algorithm for

classification of two cardiac disordersnamedasBradycardia,
and Tachycardia reliant on Artificial neural network and the
Genetic algorithm. Clinical databases have accrued large
quantities of knowledge regardingpatientsandtheirmedical
condition. This dissertation presents the ECG Disease
Detection System based on GA, and ANN, in which detection
is based on performance parameters likeprecision,Recall,F-
measure and Accuracy. Simulation results have shown that
the obtained value of precision, recall, f-measure and
accuracy in favor of proposed tested waveform are 0.853,
0.818, 0.8312 and 97.72 % for accuracy.
6. Future Scope
Future scope lies in the use of former classifiers like SVM
with the aim of having multidimensional data and making
use of feature reduction algorithms, sothataccuracyratecan
be enhanced. SVMs bring a unique solution, since the
optimality problem is rounded. This is an advantage to
Artificial neural network (ANN) which has several solutions
related with local minima and for this reason may not be
tough over different samples. For optimization algorithms
similar to artificial bee colony (ABC) and (PSO) Particle
swarm optimization would be used.
ACKNOWLEDGEMENT
I would like to get this opportunity to convey my thoughtful
gratitude and regard tomy GuideMrs.ANITASUMAN,for her
commendable guidance, precious feedback and constant
support throughout the period of the project. Her valuable
ideas were of immense helpful throughout my development
work. Her perceptive condemnation kept me working to
make this task in a much better way. Working under herwas
an tremendously knowledgeable experience for me.
References
[1] Karimian, Nima, et al. "Highly reliable key generation
from electrocardiogram (ECG)." IEEE Transactions on
Biomedical Engineering 64.6 (2017): 1400-1411.
[2] Khandelwal, Chhaya Sunil, et al. "Review Paper on
Applications of Principal Component Analysis inMultimodal
Biometrics System." Procedia Computer Science 92 (2016):
481-486.
[3] Desale, Ketan, et al.. "Comparative Study of Pre-
processing Techniques for Classifying Streaming Data."
(2015).
[4] Saravanan, K., et al. "Review on classification based on
artificial neural networks." International Journal of Ambient
Systems and Applications (IJASA) Vol 2.4 (2014): 11-18.

Disease Classification using ECG Signal Based on PCA Feature along with GA & ANN Classifier

More Related Content

What's hot (15)

Similar to Disease Classification using ECG Signal Based on PCA Feature along with GA & ANN Classifier (20)

More from IRJET Journal (20)

Recently uploaded (20)

Disease Classification using ECG Signal Based on PCA Feature along with GA & ANN Classifier