Biotechnology & Biotechnological Equipment

ISSN: 1310-2818 (Print) 1314-3530 (Online) Journal homepage: https://www.tandfonline.com/loi/tbeq20

Detection and classification of cardiovascular

abnormalities using FFT based multi-objective
genetic algorithm

B. V. P Prasad & Velusamy Parthasarathy

To cite this article: B. V. P Prasad & Velusamy Parthasarathy (2018) Detection and classification
of cardiovascular abnormalities using FFT based multi-objective genetic algorithm, Biotechnology &
Biotechnological Equipment, 32:1, 183-193, DOI: 10.1080/13102818.2017.1389303

To link to this article: https://doi.org/10.1080/13102818.2017.1389303

© 2017 The Author(s). Published by Informa Published online: 23 Oct 2017.

UK Limited, trading as Taylor & Francis
Group.

Submit your article to this journal Article views: 7463

View related articles View Crossmark data

Citing articles: 5 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tbeq20
BIOTECHNOLOGY & BIOTECHNOLOGICAL EQUIPMENT, 2018
VOL. 32, NO. 1, 183–193
https://doi.org/10.1080/13102818.2017.1389303

ARTICLE; BIOINFORMATICS

Detection and classification of cardiovascular abnormalities using FFT based

multi-objective genetic algorithm
B. V. P Prasad and Velusamy Parthasarathy
Department of Electronics and Communication Engineering, Vel Tech Multi Tech Dr. Rangarajan Dr. Sakunthala Engineering College, Chennai,
India

ABSTRACT ARTICLE HISTORY

Signal processing and data analysis are widely used methods in a biomedical research. In recent Received 24 February 2017
years, detection of cardiovascular abnormalities in patients can be achieved by using Accepted 4 October 2017
electrocardiogram (ECG) recording. In this paper, a fuzzy-based multi-objective algorithm using KEYWORDS
Fast Fourier Transform (FFT) is proposed. Initially, an effective FFT is used to extract the feature Fast Fourier Transform (FFT);
points in ECG signals, such as PQRST wave’s amplitude and wave function and then the proposed cardiovascular abnormalities;
multi-objective genetic algorithm is used to classify the abnormality of heart patient. Basically, the electrocardiogram (ECG);
ECG behaviour depends on various factors such as age, physical condition of patients and the multi-objective genetic
surrounding environment. The efficient detection of abnormalities (e.g. arrhythmia and myocardial algorithm; abnormalities
abstraction) can be achieved by initializing the above-mentioned factors and maintaining a
database containing previously attributed signals, such MIT-BIH arrhythmia. The present study
provides efficiency of around 98.7% in detection of abnormalities in patients.

Introduction and peak value of ECG signals varies largely for different
patients with different conditions [3]. Therefore, an effi-
Background
cient algorithm for removing the noise in ECG signals is
The accurate detection of cardiovascular abnormalities required. The noise in ECG signals is due to factors such
of patients has attracted greater interest in the medical as baseline wandering, motion artefacts, supply-line
field in recent years. The main reason for cardiac mal- interference within signal, electrode contact noise and
function is due to irregular function of the sinoatrial some attenuation losses [4]. Baseline wandering noise is
node (SA node). The SA node controls the contraction produced by factors such as respiration, electrode
and relaxation of the heart by electrical impulses. These impedance variation and excessive body movements.
electrical pulses are recorded as electrocardiogram (ECG) Motion artefacts can be affected by the body motion of
signals. Therefore, monitoring the electrocardiogram sig- patients during recording. The supply-line interferences
nal will provide a better way to represent the bioelectri- are due to stray effect that is caused by power cables
cal activity of the heart and the type of data acquisition from the leads. The frequency of noise in an ECG signal
technique for ECG signal can be illustrated as single lead is affected by baseline wandering and it is in the range
to multi-lead systems [1]. Accurate detection of ECG sig- of 0.3–1.5 Hz. Several hardware design techniques are
nals is necessary to determine any cardiovascular abnor- available for the reduction of noises such as supply-line
malities [2]. The ECG signal consists of P, Q, R, S and T interference and motion artefacts. Therefore, an efficient
waves and each wave represents a function of the heart: algorithm for removing the lower frequency compo-
the P-wave represents the atrial depolarization; the Q- nents present in ECG signals is much needed.
wave represents the depolarization of the interventricu- After the removal of noise, the features in the ECG sig-
lar septum; the R-wave represents the depolarization of nal should be extracted [3]. The number of fiducial
the ventricles and the S-wave represents the final depo- points in the ECG signal can be removed using the fea-
larization of the ventricles. ture extraction technique by applying an efficient
The ECG signal is recorded from the surface of a threshold to the peak points. An efficient feature extrac-
patient’s body and, while recording, this system will tion technique called Fast Fourier Transform is employed
introduce some noise in the ECG signal. The amplitude to detect these peak points and the detected peak

CONTACT B. V. P Prasad [email protected], [email protected]

© 2017 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use,
distribution, and reproduction in any medium, provided the original work is properly cited.
184 B. V. P. PRASAD AND V. PARTHASARATHY

pseudo detection of peak signal in the classification.

Therefore, a median filtering technique is employed to
remove the noise from the ECG signal. The trade-off
between noise attenuation and signal details has to be
verified [7]. This median filtering technique will act as a
low-pass filter for the removal of high frequency compo-
nents in the ECG signal.
The second step of this classification technique is fea-
ture extraction. Fast Fourier Transform technique is
effectively introduced to extract feature components
such as PQRST signals from the ECG signal [8]. This tech-
nique will decompose the original signal from time
domain to frequency domain. High frequency signals
and maximum values of particular frequency will be
Figure 1. Peaks in a normal ECG signal. extracted from the frequency domain based on the
thresholding technique [8].
The third step of the classification is abnormality
points consist of some additional points other than the
detection, where the condition (objective) of the patient
PQRST signal; then, by using an efficient threshold tech-
will be initialized. After initialization, a fitness function
nique, the P, QRS and T-wave in the ECG signal can be
for the multi-objective genetic algorithm is implemented
separated. The peak amplitudes that are in the input
using the area of convergence for a particular abnormal-
ECG are shown in Figure 1.
ity [9]. If the fitness function is less than the area of con-
The identification of the type of abnormalities from
vergence, it will be marked as an abnormality. Then, the
the detected peak points of the ECG signal poses some
second step is to determine the stage of the abnormali-
difficulties, due to the fact that two signals may have
ties with the help of the fuzzy-based scheme and this
similar patterns but may show different diseases. In
condition will deliver the degree of deviation to the out-
other cases, two signals might have different patterns
put. Hence, the abnormality type and its condition can
but indicates same disease [5]. Hence, efficient detec-
be predicted. A flow chart of our proposed abnormality
tion algorithms for detecting abnormalities in the heart
detection algorithm is shown in Figure 2.
need to be developed. Therefore, we propose a multi-
objective genetic algorithm that will differentiate the
type of abnormality based on the condition of the
Methods
patient. In the proposed scheme, the susceptibility of
the patient to the disease (more-severe, severe and Pre-processing
normal condition) were determined with the help of a
ECG are inclined to noises, since the ECG signals are
fuzzy-based scheme.
taken from the surface of the patient’s body, so many
wavelet algorithms have been introduced to de-noise
the output signal, but the Daubechies wavelet transform
General concepts
was predicted as more efficient for de-noising [10]. The
The efficiency of the classification technique is based on noise of the input signal is filtered by a moving averages
its ability to differentiate cardiac abnormalities of algorithm and this algorithm will consider the moving
patients by considering their profile. Previous classifica- slope of the ECG signal [11].
tion techniques do not hold this kind of feature, because Noise removal requires numerous approaches from dif-
they consider one and the same detection criteria for all ferent sources. Thakor and Zhu [12] achieved noise
patients [6]. The main objective of the proposed research removal by an adaptive filtering technique, considering
is divided into three phases: (i) pre-processing (noise constant reference input to cancel noise such as baseline
removal); (ii) feature classification (to separate PQRST wandering in the input ECG. However, this technique was
peak values and QRS wave functions) using FFT and (iii) found inefficient for diagnostic applications. The main
classification of abnormalities using a fuzzy-based multi- concept of the wavelet technique is to decompose an
objective genetic algorithm. ECG signal into numerous subsamples at different gauges
Initially, ECG signals contain noises due to baseline [3,13]. The wavelet transform technique does not differen-
wandering and attenuation losses and these high fre- tiate noise and signal coefficients of wavelet decomposi-
quency components will lead to false detection or tion which have a very low signal-to-noise-ratio (SNR)
 BIOTECHNOLOGY & BIOTECHNOLOGICAL EQUIPMENT 185

Discrete Wavelet Transform (DWT) is the most common

algorithm for feature extraction and the input signal is
decomposed into many levels. This decomposition tech-
nique will give all the information (waveform shape and
peak values) of the ECG signal [15]. Many feature extrac-
tion techniques such as the threshold technique are also
available [16]. However, the peak value of the ECG signal
obtained from DWT was found to be inaccurate for noisy
signals. Hence, an adaptive wavelet decomposition
method is proposed; the least square values are used for
approximation of the ECG signal. Fast Fourier Transform
(FFT) is used to convert the ECG signal, i.e. in the time
domain, to an ECG signal in the frequency domain for
more accurate extraction of peak values. QRS complex
waves of ECG signal are obtained with the help of FFT,
QRS complex, peak values of P amplitude, QRS wave and
amplitude values; then the results are fed to the classifica-
tion unit to detect the abnormality in the ECG. Peak wave
detection by soft computing techniques such as Support
Vector Transform (SVM), Particle Swarm Optimization tech-
niques was accessible to extract the peak value of ECG
[17]. This kind of soft computing will lead to a separate
condition for each peak. To define all the peak signals, a
separate technique for feature extraction is required.

Abnormality classification
The block-based neural network was used to classify the
abnormality of the ECG signal and the values can be
optimized with the help of a Particle Swarm Optimiza-
Figure 2. Flow chart of the proposed system. tion (PSO) technique. The classification of abnormality of
different sampling frequency was found inefficient [18].
The Artificial Neural Network (ANN) with Mixture of
value. This proves it an attractive method for noise Experts (MOE) technique is introduced using a local clas-
removal with improved SNR. sifier with a combination of global classifier for categoriz-
ing abnormalities. This kind of approach is quite efficient;
therefore an accurate global classifier is used to define
Feature extraction
the normal and abnormal region for ECG features. Classi-
Contour Wavelet Transform (CWT) [13] is used as an fication phase learning algorithms such as Support Vec-
efficient method for multi-lead to record ECG signals. tor Machine (SVM), genetic algorithm and neural
However, it was found that it is not suitable for single network are used to find abnormalities in ECG signals.
lead ECG signals, as they are susceptible to noises [14]. But, many of the algorithms, such as genetic algorithm
Hence, an optimization algorithm is suitable for varying and SVM, are not robust for varying characteristics of
sampling frequency. Many wavelet algorithms have patients, such as age, weight and physical condition, so
been introduced to de-noise ECG signals, but the Dau- the output of the algorithm will be based on assumption
bechies wavelet transform has been identified to be [19]. If a patient with different conditions (objectives) is
more efficient for de-noising [10]. The features of the selected, then the ranges of abnormalities are common.
ECG signal such as P-wave (both peak and amplitude) Therefore, an improved learning algorithm for abnormal-
of the ECG signal should be extracted from the de- ity detection of different patients has been framed.
noised signal. As a result, an efficient feature extraction Therefore, multi-objective genetic algorithm is proposed
technique should be introduced to remove all noise in to detect the abnormality of a particular patient based
the ECG signal [15]. on the condition of the patient [20].
186 B. V. P. PRASAD AND V. PARTHASARATHY

Classification
Pre-processing
The ECG signal raw input will be inclined to noises in the
output because of the electrical potential produced by the
heart, leading to attenuation losses. Hence, removal of
noise is essential for better abnormality prediction [10]. The
removal of noise in ECG signals will be carried out with
the help of a relaxed median filter; the trade-off between
noise attenuation and signal details has to be verified [4].
The median filter is considered as ‘m’ and the median filter
function is characterised by function [17]. Every input ECG
wave in the time domain is given by Equation (2):
Figure 3. Histogram of the signal and the standard deviation curve.

Algorithm I – Noise Removal Using Median Filter

Feature extraction

Input: Raw ECG Signal; In this phase, the main features of the ECG (amplitude
Output: Smooth ECG Signal; and time period of peak signal) such as R-peak, QRS-
Step 1: Input ECG signal is given by the wave function y:
Step 2: Signal will be subjected to median function complex wave peak, wave function and T-amplitude
were extracted. The PQRST wave extraction was done
y ¼ fm ðvðiÞÞ (1) with the help of FFT. FFT is the simplest way of analyse a
yðiÞ ¼ fm fvðiÞ; i 2 w g (2) signal with the help of Discrete Fourier Transform (DFT)
[8]. The response of the time domain and frequency
Step 3: For median filtering, the window size for filtering should be assigned domain is shown in Figure 5; FFT will eliminate lower-
first window sizew ¼ j 4f1s lengthðvÞj, where fs is the sampling frequency,
and length (v) is the total number of samples. order harmonics and the signal needs to be decom-
Step 4: The total signal is divided based on the sample size; the value of the posed into subsamples then the input should be peri-
amplitude of the signal is sorted in the window size.
!
odic in nature which is the summation of many
Xw
sinusoidal signals of different frequency. The input sam-
y ¼ sort v (3)
j¼1 ples contain N number of samples.
Step 5: Find the median of the signal with median
The basic equation for FFT is given as
functionm ¼ medianðyÞ
Step 6: After initialization of the median value, the value of median is then 2 1
N
2 1
N

compared with the value in the entire signal.

X X ð2nþ1Þk
Step 7: The window size will be incremented w ¼ 2m þ 1 and the corre- xðkÞ ¼ xð2nÞWN2nk þ xð2n þ 1ÞWN (5)
sponding value median will be calculated. n¼0 n¼0
Step 8: Stop if all the criteria are met.
where k = 0, 1, 2, 3…n

After the computation of the median, the vector

will be stored in G, where G is the subset of signal
vector S:

G[S (4)

The value of window size will be increased in corre-

sponding stages and this type of computation will
remove different frequency components in every step.
Then, this gradient vector is subtracted from the
original signal (i.e. original signal minus the gradient vec-
tor) [16]. The whole algorithm will form a low-pass filter
to remove the sharp edges in the ECG signal [7]. The
deviation of amplitude in the ECG signal is shown in
Figure 3. The pure ECG signal without noise is shown in
Figure 4. Figure 4. Filtered ECG signal.
 BIOTECHNOLOGY & BIOTECHNOLOGICAL EQUIPMENT 187

Figure 6. Two-level four-point DFT.

where

2 1 2 1
N N
X X
xe ¼ xð2nÞWN2nk and x0 ¼ xð2n þ 1ÞWN2nk (9)
n¼0 n¼0

By this formula, the value of the original signal is decom-

posed into many subdivisions in the frequency domain.
The decomposition process can be obtained by re-order-
ing of the signal. The basic functional diagram of the FFT
algorithm is shown in Figure 6.
The radix-2 FFT algorithm divides the entire DFT cal-
culation into two parts of 2-point DFTs. Each 2-point DFT
will consist of a multiply-and-accumulate operation
called a butterfly, as shown in Figure 6. Eight input time
Figure 5. Response of signal for time and frequency domain. samples are first split into two groups of 4-point DFTs.
The two 4-point DFTs are then combined into four 2-
Note: sine and cosine functions can be expressed in
point DFTs. Then the four 2-point DFTs are combined to
polar or rectangular coordinates using Euler’s equation
produce the final output xðkÞ.
[21]. The output spectrum of DFT 0 xðkÞ0 is represented as
The FFT algorithm is the simplest form to evaluate the
correlation between input time samples, N cosine and N
DFT [8]. The computational efficiency of the DFT algo-
sine equation. In complex notations, the DFT algorithm
rithm will be N2 for computing N-points, whereas the
consists of both a real and a complex part, but for real-
computational efficiency of the FFT algorithm will be
time applications the complex function will be removed. N
2 log N. The magnitude and phase response of FFT are
N N shown in Figure 7.
1      1
2 2X 2pnk 2pnk 1 2X
xðkÞ ¼ xð2nÞ cos  jsin þ xWNk
n n¼0
N N n n¼0
(6) RR intervals
2pi
where WN ¼ e is the nth primitive root of unity and xð
N The heartbeat of the patient will be determined with the
nÞ is the signal amplitude. The above equation is divided help of RR intervals, which are calculated by the distance
into odd and even elements. between two R peaks. This value of the RR-interval will
also be used to understand the heart rhythm of a partic-
2 1 2 1
N N
X X ð2nþ1Þk
ular patient. The heartbeat is defined as the number of
xðkÞ ¼ xð2nÞWN2nk þ xð2n þ 1ÞWN (7) R-peak values in a particular minute. The rhythm of the
n¼0 n¼0
heart can be determined by the following formula:
XðkÞ ¼ xe þ WNk xo (8)
RRint ¼ Ri  Ri1 (10)
188 B. V. P. PRASAD AND V. PARTHASARATHY

Figure 7. Magnitude and phase response of FFT. Figure 8. Detection of peaks in an ECG signal.

where Ri is the time period of the ‘R’ peak in the ith beat (normal beat). The peak signals are plotted in an
wave and Ri1 is the time of the ‘R’ peak in the ði  1Þth ECG wave (Figure 8).
wave.
RR interval calculation provides useful information for
clinical diagnosis and identification of symptoms for Multi-objective genetic algorithm
arrhythmia events that are associated with heart-rate
Objectives that are taken into consideration will conflict
variation in patients.
with each other for many problems in the real world.
Hence, an optimizing value x with respect to single
QRS detection objective is required. An optimal solution for the multi-
objective optimization problem is to consider a set of
The duration of the QRS wave varies with the origin and solutions, which has capability to satisfy objectives at
conduction path of the activation pulse in the heart. So, an optimum level without affecting other solutions.
QRS will be the main feature for the classification of the The multi-objective genetic algorithm uses two opera-
conditions of the heart. In this work, the QRS duration is tors to determine new solutions and they are mutation
represented by the time interval between the two peak and crossover. In the crossover operation, generally
Q and S signals which can be calculated by. two parent random numbers are combined together to
form new offspring. The selection of parent in the exist-
QRS ¼ Qti  Sti (11) ing random number is considered according to the
strength of the parent to produce better offspring [21].
where Qti and Sti are the time values of the Q and S wave. The iteration efficient random numbers will undergo
In each sub-band, signal variance can be found with mutation to produce better offspring. The selection of
the help of average power in the sub-band. The variance parent will be based on the fitness function of a particu-
of a QRS signal is calculated by the formula, lar node. The fitness value can be initialized by calculat-
2
ing the particular offspring with a corresponding
1X N
fitness function. For example, every different objective
s 2X ¼ ½jxðnÞ  xj (12)
N n¼1 will possess a different fitness function; factors such as
age, weight and physical condition will have a different
where x is the sample mean of signal and N is the num- objective function when compared to the physical
ber of samples in the given input segment. The number activity of a particular patient. The offspings will
of samples N differs from one ECG to another based on undergo repetitive mutation until an effective result is
whether its type is large beat (ectopic beat) or narrow obtained [9].
 BIOTECHNOLOGY & BIOTECHNOLOGICAL EQUIPMENT 189

Algorithm II – Multi-Objective Genetic Algorithm signal is then stored in a vector which should be equal
Input: Extracted Features QRS Wave; to the length of the ECG signal. The input ECG signal will
Output: Detected Abnormality; contain noise and it should be removed for better classi-
Step 1: Set i = 1, N number of random populations are selected for every
variable fication of the ECG peak signal amplitude. The ECG signal
Step 2: Initialize the number of objective functions i, where i = 1,2,3… .k. de-noising will be done with the help of a relaxed
Xk
2.1: Calculate weight for every random number xk ¼ ð1 6 uk Þ i¼1 ui , median filter. In the relaxed median filter technique, the
Where uk is the random number for that particular population. The weight
of the random number is the measure of significance of the random num- window will be assigned as 1 6 4fs , where fs is the sam-
ber to remaining numbers. pling frequency. The sampling frequency of the filter will
2.2: Calculate the selection probability
be incremented at each step and the high frequency sig-
X nals other than fundamental ECG signals will then be
PðxÞ ¼ ðf ðxÞ  fmin Þ1 ðf ðyÞ  f ðminÞÞ (13)
y¼p1 removed from the ECG signal. The median filter will act
as a simple low-pass filter to remove the high frequency
where f ðxÞ is the objective function, f ðminÞ is the minimum of the function,
f ðyÞ is the common objective. components from the input ECG signal. The output of
Selection probability is the chance of each random number to be selected as the median filter will be a pure ECG signal and the signal
the best fit.
Step 3: Based on the objective function, the rank of that particular random will be forwarded to the feature extraction phase. Fea-
number is calculated by the following formula: ture extraction will be carried out with the help of FFT.
The input ECG signal will be seperated into two 4-point
f ¼ zk ðxi Þ (14)
DFT; the 4-point DFT signal will be subdivided into four
2 32 3 2-point DFT signals. This combination is then compared
z1 x1p1 x1p2 x1p3 x1p4
6 z 76 x x2p4 7 to form an input signal in the frequency domain. The
6 2 76 2p1 x2p2 x2p3 7
f ¼ 6 76 7 (15)
4 z3 54 x3p1 x3p2 x3p3 x3p4 5 outputs from the FFT are the peak amplitudes of the P-
zk xkp1 xkp2 xkp3 xkp4 wave, QRS waves amplitude and T-wave amplitude.
The amplitude and position values are given to the
multi-objective genetic algorithm and the Parametric
where zk is the objective function; the rank of an individual in the matrix is
selected from the fitness value of that particular individual.
equation of objective will be considered as objective
Step 4: Rank matrix of all the populations are then compared to obtain the function for the genetic algorithm [21]. Random num-
parent individuals. bers will be generated for feature extracted output [23,
Step 5: The parent nodes are moved to cross-over in this cross-over phase; N
number of off-springs will be produced. 24]. The disease susceptibility will be determined with
Step 6: The above steps are repeated until a best offspring is produced. the help of fuzzy concept, such as very severe, severe or
Step 7: After initialization of offspring, the best offspring will be obtained
based on the rank of that particular offspring. normal.
Step 8: Repeat the step until a best solution is determined. If the condition
gets satisfied, then stop the iteration.
Results and discussion
Datasets We have shown an experiment on automated detection
of arrhythmia and other heart disease using the MIT-BIH
The proposed method uses the MIT-BIH arrhythmia arrhythmia database. In this experiment, we have used a
database [22]. The data consist of 48-h excerpts from combination of linear and non-linear methods to extract
two channels ambulator containing about 134 samples the features of ECG signals. The data obtained consist of
and the sampling frequency of the ECG signal is taken as 11,000 samples in the time interval of about 10 s with
360 samples per second with a resolution of 11-bit over sampling frequency of 360 Hz. In this part, the classifica-
a maximum range of 10 mV amplitude. The input data tion performance of arrhythmia by the proposed multi-
are grouped in the same directory and the recording of objective genetic algorithm and existing methods such
ECG contains maximum of 13,000 data values. The data as PSO, SVM and Genetic Algorithm (GA) algorithm was
retrieval technique will differ for varying databases. compared.
Hence, the system will not be robust; effectiveness calcu-
lation will be carried out from the same dataset.
Performance evaluation – denoising or pre-
processing
Experimental procedure
The analysis of the proposed work was carried out in
The input raw ECG signal is then read with the given MATLAB 2015a environment; raw ECG data were
input sample frequency of 360 samples per second. The obtained from the MIT-BIH database and then de-noised
value of the amplitude and position of the amplitude with a median filter to remove the noise in the input
190 B. V. P. PRASAD AND V. PARTHASARATHY

Table 2. Sensitivity and specificity comparison of various

algorithms.
Classifier Sensitivity Specificity
Hybrid fuzzy NN 96 99.58
SVM-linear 87.79 91.11
SVM-quadratic 90.93 91.11
SVM-polynomial 92.79 97.49
NN 97 98.22
FFT 97.5 98.3

that the false detection of ECG peaks by FFT is reduced

when compared to the existing DWT algorithm [27] as
shown in Table 2. The error detected by the FFT algorithm
can be calculated as follows:

ðTP þ TNÞ
error ¼ (16)
Figure 9. Comparison of filtered and raw ECG signal. ðtotal no of beats Þ

ECG signal [25]. Then the filtered ECG signal was plotted
over the raw ECG signal to determine the variation in Sensitivity (se ) and specificity (sp ) give the measure of
the ECG signal. The filtered output is shown in Figure 9. positives and negatives that can be identified from the
The SNR of ECG signals were identified for different sig- database. The sensitivity and specificity of beat classifica-
nals in the database and the results were compared with tion can be calculated as follows.
existing Hilbert transform. The result was found more
TP
efficient than existing systems [26]. Sensitivity ¼ (17)
TP þ FN
SNR is the ratio of the power of the pure signal to the
TP
power of the noise signal of the input ECG. Increased value Specificity ¼ (18)
TP þ FP
of SNR will lead to better efficiency. The SNR values of the
median filtered ECG signal are given in Table 1. The median It is evident that the FFT algorithm was found to be
filter was found efficient in removing the high frequency sensitive for detection of peak signals (Table 2) and the
components that are present in the input ECG signal. detected features are compared in Figure 10. The plot
shows the histogram value of the peak signal and the
standard deviation curve for the detected ECG signal.
Performance evaluation – beat classification The major area of the histogram above the standard
FFT was used for efficient beat classification and to detect deviation curve illustrates low sensitivity of ECG signal
peak amplitudes. The proposed FFT was compared with and it can be found that data above 100 and 101 show
existing discrete wavelet transform algorithm and it was some abnormal behaviour of ECG signals. The standard
found that the proposed method has efficiency of 98.7% deviation curve gives some clear idea about peak detec-
in peak detection. The incorrect detections of peak waves tion by the FFT algorithm (Figure 10).
will be calculated by identifying the values of true positive The peak detection of ECG signals by FFT is then com-
(TP), true negative (TN) and false positive (FP) of detected pared with DWT. True positive is the number of correct
ECG signals. The TP, TN and FP values were detected by peaks detected, whereas True negative is the number of
the FFT algorithm and then compared with existing dis- peaks detected incorrectly by the detector. The true posi-
crete wavelet transform algorithm. The results showed tives and true negatives of the two algorithms are com-
pared in Table 3 and Figure 11. It is clear that the detection
Table 1. SNR comparison value of filtered signal. of genuine peak values from FFT are more efficient than
Number SNR using Hilbert transform SNR using median filter the existing DWT algorithm (Table 3 and Figure 11).
100 24 29
101 17 20
102 22 26 Performance evaluation – abnormality detection
103 20 22
104 19 23 A multi-objective genetic algorithm was used to differ-
105 22 24
106 28 32 entiate between normal ECG signals and abnormal ECG
107 25 29 signals. The ECG data with detected arrhythmia from the
 BIOTECHNOLOGY & BIOTECHNOLOGICAL EQUIPMENT 191

Figure 11. Prediction comparison.

Table 4. Abnormality detected ECG signals.

Class Records number
Normal 100,101,103,105,106,112,113,114,
115,116,117,121,122,123,201,202,
205,209,213,215,219,220,222.
Abnormal 104,108,109,111,118,119,124,200,
203,207,208,210,212,214,217,221,
223,228,230,231,232,234.

Figure 10. Standard deviation curve of the peak amplitude. percentage of our proposed system is around 98.7%
(Table 4 and Figure 12).
database are given in Table 4. Abnormalities detected by The proposed multi-objective genetic algorithm was
both the genetic and the multi-objective genetic algo- implemented using MATLAB 2015a environment in 64-
rithm were compared as shown in Tables 5 and 6. It is bit Windows operating system. The important advantage
shown that out of 45 data entries, 22 ones were found to of the proposed multi-objective algorithm is that it has
be abnormal with the help of the multi-objective genetic very low computational cost for feature classification.
algorithm (Tables 5 and 6). The proposed algorithm was The total time of computation is about 0.28 ms of single
found to be accurate in detection of abnormalities from CPU implementation for the resolution of 360 samples
ECG [28]. The classification accuracy of our proposed per second [29]. The speed of computation is faster than
multi-objective genetic algorithm was compared with the real time requirement. The hybrid fuzzy NN, Mixture
existing classification algorithms such as Neural Network of Experts, ANN-PSO, ANN, Probabilistic NN, MLPNN,
(NN), genetic algorithm etc. and the proposed fuzzy SVM and MO-GA algorithm provided accuracy upto 96%,
based multi-objective genetic algorithm was found to 98%, 95.5%, 98.4%, 96.9%, 97.5%, 97.6%, 97.7% and
be more accurate than the existing system. The accuracy 98.8%, respectively (Figure 12 and Table 4). Recently, we

Table 3. Beat classification by fast Fourier transform.

Number Total beats Detected beats from True True False Detected beats from True True False
(Actual) DWT positive negative positive FFT positive negative positive
100 2273 2272 2272 1 0 2273 2273 0 0
101 1865 1866 1863 2 3 1864 1863 2 1
102 2187 2180 2174 13 6 2189 2184 3 5
103 2084 2083 2083 1 0 2086 2084 0 2
104 2229 2260 2224 5 36 2229 2228 1 1
105 2572 2579 2569 3 10 2571 2570 2 1
106 2027 2026 2022 5 4 2033 2025 2 8
107 2137 2128 2128 9 0 2135 2134 3 1
108 2532 2522 2520 12 2 2528 2527 5 1
109 2278 2250 2248 30 2 2280 2278 0 2
110 2601 2585 2585 16 0 2615 2600 1 15
192 B. V. P. PRASAD AND V. PARTHASARATHY

Table 5. Performance evaluation of different algorithms. are 22, 20 and 22, respectively. The true value, genetic
Reference Classifier Year Accuracy algorithm and multi objective genetic algorithm for the
Osowski and Linh [5] Hybrid fuzzy NN 2001 96.06%
Shen et al. [14] Mixture of experts 2002 98.00%
number of detected normal ECGs are 23, 25 and 23,
Kumar et al. [29] ANN-PSO 2015 95.58% respectively (Table 6). For instance, for the combined
Gothwal et al. [30] ANN 2011 98.48% dataset with 184 subjects, the WDIST2-5 and ML-5
Yu and Chen [31] Probabilistic NN 2007 96.92%
Das and Ari [32] MLPNN 2013 97.58% attained 96.31% and 98.33% accuracies, respectively,
Faziludeen and Sabiq [33] SVM 2013 97.50% while the cascaded two 99.52% subject verification accu-
Proposed method MO-GA Proposed 98.70%
racy. It should be noted that the unified dataset includes
subject ECG measured from mobile phones, subject ECG
measured in the presence of arrhythmia and subject
Table 6. Comparison of abnormality detection. ECG data measured 2–20 times over a 6-month period.
Algorithms True Genetic Multi-objective
value algorithm genetic algorithm
Number of detected 22 20 22 Conclusions
abnormal ECG
Normal ECG 23 25 23
The classified individual characteristics of heartbeats
from standard 10-s, 12-lead ECG signals database were
used to identify arrhythmia in patients. The features are
have reported a new ECG signal classification system, extracted by the FFT algorithm and then it is fed to
which employs the Butterworth filtering technique for the multi-objective genetic algorithm by considering the
pre-processing (denosing); Integral Model of Peak Esti- age, weight and physical condition of the patient. The
mation (IMPE) technique for estimation of the peaks of multi-objective genetic algorithm was determined to
the P, Q, R and S waves; the Peak-to-Peak Sampling Pat- be more sensitive to identify abnormalities in patients.
tern (PPSP) method for prediction of the signal features The results from the multi-objective genetic algorithm
and the Multivariate Machine Learning with Neural Net- were then compared with existing genetic algorithm
work (MML-NN) classification technique to classify and the simulation results illustrated that the multi-
whether the ECG signal is normal or abnormal [34]. It has objective genetic algorithm was more efficient for vary-
the advantage of multiple abnormality level classifica- ing factors. The final result showed 20% increase in the
tion, correct sequence rate monitoring and efficient efficiency in detecting abnormalities when compared to
peak-to-peak analysis. the existing system. This present research provides bet-
In this study, the recording numbers for normal class ter efficiency for both SNR and mean square error with
ECG signals are 100, 101, 103, 105, 106, 112, 113, 114, efficiency of around 98.70%. This proposed work sets
115, 116, 117, 121, 122, 123, 201, 202, 205, 209, 213, 215, out an automatic method to identify arrhythmia in
219, 220 and 222, respectively. The recording numbers patients from 12-lead ECG signals by classifying heart-
of abnormal ECG conditions are 104, 108, 109, 111, 118, beats using machine-learning methods. The results
119, 124, 200, 203, 207, 208, 210, 212, 214, 217, 221, 223, showed that it is possible to obtain better efficiency
228, 230, 231, 232 and 234, respectively (Table 5). The using multi-objective genetic algorithm. In future
true value, genetic algorithm and multi objective genetic research, analysing and modelling the sequence of
algorithm for the number of detected abnormal ECGs heartbeats using advanced machine learning methods
can be implemented to achieve better performance.

Disclosure statement
No potential conflict of interest was reported by the authors.

ORCID
B. V. P Prasad http://orcid.org/0000-0003-2099-3081

References
[1] Hao W, Chen Y, Xin Y. ECG baseline wander correction by
mean-median filter and discrete wavelet transform. In:
Figure 12. Accuracy comparison chart for different algorithms. [editors unknown].33rd Annual International Conference
 BIOTECHNOLOGY & BIOTECHNOLOGICAL EQUIPMENT 193

of the IEEE Engineering in Medicine and Biology Society, [17] Mehta SS, Lingayat NS. Application of support vector
EMBC; 2011 Aug 30–Sep 3; Boston, Massachusetts (USA): machine for the detection of P-and T-waves in 12-lead elec-
IEEE; 2011. p. 2712–2715. trocardiogram. Comput Meth Prog Biomed. 2009;93:46–60.
[2] AlMahamdy M, Riley HB. Performance study of different [18] Shadmand S, Mashoufi B. A new personalized ECG signal
denoising methods for ECG signals. Pro Com Sci. classification algorithm using block-based neural network
2014;37:325–332. and particle swarm optimization. Biomed Signal Process
[3] Agante PM, Marques De Sa JP. ECG noise filtering using Control. 2016;25:12–23.
wavelets with soft-thresholding methods. In: Murray A, [19] Subasi A, Gursoy MI. EEG signal classification using PCA,
Swiryn S, editors. Computers in cardiology. Vol. 26. Hann- ICA, LDA and support vector machines. Exp Syst Appl.
over (Germany): IEEE; 26-29 Sept, 1999. p. 535–538. 2010;37:8659–8666.
[4] Strumillo P. Nested median filtering for detecting T-wave [20] Murata T, Ishibuchi H. MOGA: multi-objective genetic
offset in ECGs. Electron Lett. 2002;38(14):682–683. algorithms. In: [editors unknown]. Proceedings of 1995
[5] Osowski S, Linh TH. ECG beat recognition using fuzzy IEEE International Conference on Evolutionary Computa-
hybrid neural network. IEEE Trans Biomed Eng. 2001;48 tion. Vol. 1; 1995 29 Nov–1 Dec; Perth, WA. (Australia):
(11):1265–1271. IEEE; 1995. p. 289–294.
[6] Konak A, Coit DW, Smith AE. Multi-objective optimization [21] Zidelmal Z, Amirou A, Ould-Abdeslam D, et al. ECG beat
using genetic algorithms: a tutorial. Reliab Eng Syst Safety. classification using a cost sensitive classifier. Comput
2006;91:992–1007. Methods Programs Biomed. 2013;111(3):570–577.
[7] Ahmed W, Khalid S. ECG signal processing for recognition [22] Moody GB, Mark RG. The impact of the MIT-BIH arrhyth-
of cardiovascular diseases: a survey. In: [editors unknown]. mia database. IEEE Eng Med Biol Mag. 2001;20(3):45–50.
Proceedings of the Sixth International Conference on [23] Pan J, Tompkins WJ. A real-time QRS detection algorithm.
Innovative Computing Technology (INTECH); 24-26 Aug, IEEE Trans Biomed Eng. 1985;32(3):230–236.
2016; Dublin. (Ireland): IEEE; 2016. p. 677–682. [24] Olesen LL, Andersen A. ECG as a first step in the detection
[8] Cain ME, Ambos HD, Witkowski FX, et al. Fast-Fourier of left ventricular systolic dysfunction in the elderly. ESC
transform analysis of signal-averaged electrocardiograms Heart Fail. 2016;3(1):44–52.
for identification of patients prone to sustained ventricu- [25] Choi S, Adnane M, Lee GJ, et al. Development of ECG beat
lar tachycardia. Circulation. 1984;69(4):711–720. segmentation method by combining lowpass filter and
[9] Deb K. Multi-objective genetic algorithms: problem diffi- irregular R–R interval checkup strategy. Expert Syst Appl.
culties and construction of test problems. Evol Comput. 2010;37(7): 5208–5218.
1999;7(3):205–230. [26] Joshi V, Verma AR, Singh Y. De-noising of ECG signal using
[10] Kestler HA, Haschka M, Kratz W, et al. De-noising of high- adaptive filter based on MPSO. Proc Comp Sci.
resolution ECG signals by combining the discrete wavelet 2015;57:395–402.
transform with the Wiener filter. In: [editors unknown]. [27] Khorrami H, Moavenian M. A comparative study of DWT,
Computers in cardiology. Vol. 25. Cleveland (Ohio): IEEE; CWT and DCT transformations in ECG arrhythmias classifi-
13-16 Sept, 1998. p. 233–236, . cation. Expert Syst Appl. 2010;37:5751–5757.
[11] Blanco-Velasco M, Weng B, Barner KE. ECG signal denois- [28] Elhaj FA, Salim N, Harris AR, et al. Arrhythmia recognition
ing and baseline wander correction based on the empiri- and classification using combined linear and nonlinear
cal mode decomposition. Comp Biol Med. 2008;38:1–13. features of ECG signals. Comput Methods Programs
[12] Thakor NV, Zhu YS. Applications of adaptive filtering to Biomed. 2016;127:52–63.
ECG analysis: noise cancellation and arrhythmia detection. [29] Kumar VR, Sivanantharaja A. Feed forward neural network
IEEE Trans Biomed Eng. 1991;38(8):785–794. optimized using PSO and GSA for the automatic classifica-
[13] Morlet D, Peyrin F, Desseigne P, et al. Time-scale analysis tion of heartbeat. Middle-East J Sci Res (MEJSR). 2015;23
of high-resolution signal-averaged surface ECG using (5):896–901.
wavelet transformation. In: Degani R,. Computers in cardi- [30] Gothwal H, Kedawat S, Kumar R. Cardiac arrhythmias detection
ology. Venice (Italy): IEEE; 23-23 Sept, 1991. p. 393–396. in an ECG beat signal using fast Fourier transform and artificial
[14] Shen TW, Tompkins WJ, Hu YH. One-lead ECG for identity neural network. J Biomed Sci Eng. 2011;4: 289–296.
verification. In: [editors unknown]. Proceedings of the Sec- [31] Yu SN, Chen YH. Electrocardiogram beat classification
ond Joint EMBS/BMES Conference; 2002 Oct 23–26; Hous- based on wavelet transformation and probabilistic neural
ton, TX. (NJ, USA): IEEE; 2002. p. 62–63. network. Pattern Recognit Lett. 2007;28:1142–1150.
[15] Homaeinezhad MR, Tavakkoli E, Ghaffari A. Discrete wavelet- [32] Das MK, Ari S. ECG beats classification using mixture of
based fuzzy network architecture for ECG rhythm-type features. Int Sch Res Notices. 2014 [cited 2017 Apr
recognition: feature extraction and clustering-oriented 6];2014:178436. [12 p.]. DOI:10.1155/2014/178436.
tuning of fuzzy inference system. IJSP-IMPPR. 2011;4(3):107– [33] Faziludeen S, Sabiq PV. ECG beat classification using
130. wavelets and SVM. In: [editors unknown]. Proceedings of
[16] Ravindrakumar S, Shruthilaya K, Rekha M, et al. An hybrid the IEEE Conference on Information & Communication
method using wavelet transform and slope thresholding Technologies (ICT); 2013 Apr 11–12; Thuckalay, Tami
for cardiogram de-noising and delineation. In: Rajasingh I, Nadu. (India): IEEE; 2013. p. 815–818.
Desikan K, editors. Proceedings of the International Con- [34] Prasad BVP, Parthasarathy V. A novel electrocardiogram
ference on Mathematical Computer Engineering-ICMCE. signal feature prediction technique using peak-to-peak
Vol. 1; 29–30 Nov, 2013; VIT University, Chennai: Bonfring; sampling pattern for multiple abnormality level classifica-
2013. p. 105–111. tion. J Comput Theor Nanosci. 2017;14(7):3382–3392.