You are on page 1of 7

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/265854337

Heart diseases prediction based on ECG signals’ classification using a genetic-


fuzzy system and dynamical model of ECG signals

Article  in  Biomedical Signal Processing and Control · November 2014


DOI: 10.1016/j.bspc.2014.08.010

CITATIONS READS

28 760

3 authors, including:

M. H. Vafaie Mohammad Ataei


University of Isfahan University of Isfahan
9 PUBLICATIONS   125 CITATIONS    62 PUBLICATIONS   350 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Nonlinear system time delay state estimation View project

MIMO radar waveform design View project

All content following this page was uploaded by M. H. Vafaie on 14 January 2019.

The user has requested enhancement of the downloaded file.


Biomedical Signal Processing and Control 14 (2014) 291–296

Contents lists available at ScienceDirect

Biomedical Signal Processing and Control


journal homepage: www.elsevier.com/locate/bspc

Heart diseases prediction based on ECG signals’ classification using a


genetic-fuzzy system and dynamical model of ECG signals
M.H. Vafaie, M. Ataei ∗ , H.R. Koofigar
University of Isfahan, Department of Electrical Engineering, Isfahan, Iran

a r t i c l e i n f o a b s t r a c t

Article history: The early detection of abnormal heart conditions is vital to identify heart problems and avoid sudden
Received 25 February 2014 cardiac death. The people with similar heart conditions almost have similar electrocardiogram (ECG) sig-
Received in revised form 15 July 2014 nals. By analyzing the ECG signals’ patterns one can predict arrhythmias. Since the conventional methods
Accepted 21 August 2014
of arrhythmia detection rely on observing morphological features of the ECG signals which are tedious
and very time consuming, the automatic detection of arrhythmia is more preferable. In order to automate
Keywords:
detection of heart diseases an adequate algorithm is required which could classify the ECG signals with
Electrocardiogram (ECG) signals
unknown features according to the similarities between them and the ECG signals with known features.
classification
ECG signals dynamical model
If this classifier can find the similarities precisely, the probability of arrhythmia detection is increased
Genetic algorithm and this algorithm can become a useful means in laboratories. In this article a new classification method
Genetic-fuzzy system is presented to classify ECG signals more precisely based on dynamical model of the ECG signal. In this
Fuzzy logic proposed method a fuzzy classifier is constructed and its simulation results indicate that this classifier
can segregate the ECGs with an accuracy of 93.34%. To further improve the performance of this classifier,
genetic algorithm is applied where the accuracy in prediction is increased up to 98.67%. This proposed
method increases the accuracy of the ECG classification regarding more precise arrhythmia detection.
© 2014 Elsevier Ltd. All rights reserved.

1. Introduction consuming. Automation of the ECG signal analysis based on com-


puter programming is a better system to diagnose heart diseases.
The electrocardiogram (ECG) signals are the recordings of the In this article a new automatic categorization method is proposed
bioelectrical activities of the cardiac system and some or all parts that can classify similar ECG signals into one group and predict the
of these signals are changed from normal condition when heart dis- potential of the heart diseases in each group.
ease occurs [1,2]. People with similar heart conditions have almost Several techniques on automatic detection of the abnormal con-
ECG signals with similar features. If an ECG signal with unknown ditions are reported. In these reports, the analysis of the ECG signals
features has a morphological pattern similar to that of an ECG signal to detect these abnormal conditions is conducted by using auto-
with an especial arrhythmia, it could be deduced that this unknown correlation function, frequency domain features, time frequency
signal has the same arrhythmia. In such cases, it is possible to detect analysis and wavelet transform [3–7]. Since these techniques
the heart disease by analyzing the pattern of the ECG signals. ignore the nonlinear behavior of the signal dynamics, the informa-
Early detection of abnormal heart conditions can avoid sudden tion provided on ECG signals are insufficient and their classification
cardiac death and other dangerous illnesses caused by heart dis- based on these techniques is not acceptable. To avoid this deficiency
ease. Many researchers have reported results on analyzing the ECG several articles in the area of the nonlinear analysis and chaos
signals and detecting the abnormal conditions thereof. To detect theory in studying the behavior of the ECG signal from an exper-
such conditions in laboratory, a continuous ECG signal monitoring imental time series are published. The objective of these articles
is required for each patient. This procedure is tedious and very time is to determine whether the dynamical measures such as correla-
tion dimension and especially the greatest Lyapunov exponent can
serve as useful tools for ECG classification. Though there exist good
results on distinguishing between normal and abnormal patients,
∗ Corresponding author at: Department of Electrical Engineering, Faculty of Engi-
segregation between abnormal ECG signals and their acceptable
neering, University of Isfahan, Isfahan, Iran. Tel.: +98 0311 7934068/0913 3344056;
fax: +98 3137933071.
classification is yet to be perfected [8–18].
E-mail addresses: mh.vafaie@eng.ui.ac.ir (M.H. Vafaie), ataei@eng.ui.ac.ir, Attempt is made in this article to introduce a method based on
mataei1971@yahoo.com (M. Ataei), koofigar@eng.ui.ac.ir (H.R. Koofigar). dynamical model of the ECG signal that can classify ECG signals in

http://dx.doi.org/10.1016/j.bspc.2014.08.010
1746-8094/© 2014 Elsevier Ltd. All rights reserved.
292 M.H. Vafaie et al. / Biomedical Signal Processing and Control 14 (2014) 291–296

a more precise manner which would enhance the prediction of the Table 1
Parameters of the ECG model for generating ECG signal illustrated in Fig. 1(a).
illnesses related to the heart diseases.
This method is found on the fact that similar ECG signals have Index (i) P Q R S T
almost similar heart conditions and when these ECG signals are ti (s) −0.2 −0.05 0 0.05 0.4
modeled a similarity is observed among their parameters. When  i (radians) −0.33 −0.08␲ 0 0.08 0.50
the parameters of the dynamical model of the ECG signals are ai 1.2 −5.0 30.0 −7.5 0.75
obtained, by applying adequate classifier the ECG signals can be bi 0.25 0.1 0.1 0.1 0.4

classified according to these parameters. Through this classifica-


tion, since the ECG signals placed in one group indicate similar Table 2
heart condition, the heart condition of the unknown signals can ECG parameters of a normal person and a person who has an AF arrhythmia.
be predicted.
Parameters Normal heart beat Atrial fibrillation
In order to construct this classifier: first, the genetic algorithm
(GA) is applied to estimate unknown parameters of the dynami- ti = [t1 t2 t3 t4 t5 ] [−0.2 −0.05 0 0.05 0.4] [−0.1 −0.02 0 0.02 0.2]
ai = [a1 a2 a3 a4 a5 ] [1.2 −5.0 30.0 −7.5 0.75] [0 −5.2 29.8 −7.3 0.95]
cal model of the clinically obtained ECG signals from PhysioBank
bi = [b1 b2 b3 b4 b5 ] [0.25 0.1 0.1 0.1 0.4] [0 0.1 0.1 0.1 0.4]
database [19], second, by studying a group of ECG signals with
known features, a set of fuzzy rules is generated and through them
a fuzzy classifier is constructed. Here, to improve the performance
f2 is the respiratory frequency [22]. The details of these three
of this fuzzy classifier, the GA is applied to optimize the mem-
differential equations are presented in Appendix I.
bership functions and fuzzy rules which lead to the construction
According to [22], distinct points on the ECG signal like the P,
of a genetic-fuzzy classifier. Finally, this genetic-fuzzy classifier is
Q, R, S, and T used in (1) can be configured by setting ai , bi , and
implemented for ECG signal classification.
 i parameters. The location of these points is set by  i , as well as
The structure of this article is as follows: the dynamical model
the amplitude and the width of the ECG waves are set by ai and bi .
of the ECG signal, the genetic algorithm and its application in esti-
Proper setting of these parameters would generate a desirable ECG
mation of the unknown parameters of the dynamical model are
signal; therefore, by applying the parameters of Table 1 in (1), the
presented in Section 2. The theory of the fuzzy and genetic-fuzzy
ECG signal is generated through MATLAB and illustrated in Fig. 1a.
classifiers is introduced in Section 3 which are used in classifica-
The real ECG signal [19] which has characteristics similar to the
tion of the ECG signals. The results and discussion are presented in
simulated ECG is demonstrated in Fig. 1b to compare real ECG and
Section 4 and finally, the conclusions in Section 5 end the article.
ECG signal generated by simulating (1).
Since heart diseases change the ECG signal model parame-
2. Materials and methods ters, recognizing the effects of these parameters’ changes through
related computer programs make the prediction of the heart dis-
2.1. Dynamical model describing the ECG signals eases more rapid in an automatic manner.

If an ECG signal can be modeled precisely by differential equa- 2.2. Genetic algorithm and its application in parameter
tions, one can study this signal using certain mathematical tools. estimation
Modeling of the ECG signal by differential equations is of essence,
where the effect of heart diseases on ECG signals can be analyzed In the ECG model by changing the parameters ai , bi , and  i a
by using the mathematical software like MATLAB. When the effect desirable ECG signal can be generated. The value of these parame-
of the heart diseases are modeled, by a suitable computer program, ters is different in the patients with different heart conditions, and
the prediction and diagnosis of the heart diseases can be made and these differences are illustrated in Table 2, where one individual is
conducted, respectively, on an automatic basis. normal and the other has an Atrial fibrillation (AF) arrhythmia.
The initial model of the ECG signal was reported in 1972 by After the values of these parameters for various heart diseases
Zeeman [20], where the effects of the sympathetic and parasym- are identified, then, the values obtained from ECG signal of a patient
pathetic nervous system are not considered. A new model for ECG with unknown disease are compared to these identified values
signal is proposed in [21], where a new parameter according to the which would contribute to the diagnosis of the heart diseases.
activities of sympathetic and parasympathetic systems is defined The ECG signals in clinics always are obtained in the form of a
and applied in the Zeeman model. Afterwards, an improved and continuous signal and a direct access to accurate values of these
flexible model for generating ECG signal which generates a trajec- parameters is not possible; therefore, in order to obtain these
tory in the three-dimensional state-space with coordinates (x, y, z) parameters, the ECG signal must be sampled and these parameters
is introduced; in addition to the fact that the variation of the ECG must be calculated through estimation techniques.
is produced through the trajectory motion at z direction [22]. This Because the dynamical model of the ECG contains three nonlin-
model is described using three ordinary differential equations pre- ear differential equations, the classic estimation methods cannot
sented in (1). The accuracy and some applications of this model are be adopted in estimating the unknown parameters of this model
studied in [23]. [24–26].
The GA is applied in order to estimate these parameters. In this
ẋ = ˛x − wy proposed method, the unknown parameters of the ECG signal are
ẏ = ˛y − wx estimated in a manner where the error function defined in (2) is
   (1) minimized (or its inverse is maximized). In (2), s1 is the sampled
i2
ż = − ˛i i exp − − (z − z0 ) ECG signal obtained from continuous ECG signal acquired in clin-
2b2i
i∈{P,Q,R,S,T} ical environment with unknown parameters, and s2 is the signal
generated by simulating (1). Both the s1 and s2 are sampled in the

where ˛ = 1 − x2 + y2 , i = ( − i ) mod2,  = ˛ tan2 same frequency.
(y, x) (the four quadrant arctangent of the real parts of the ele- 
ments of x and y, with − ≤ ˛tan2(y, x) ≤ ) and w is the angular e= (s1(i) − s2(i)), F = 1/e (2)
velocity of the trajectory, zo = Asin(2f2 t) where A = 0.15 mV and i
M.H. Vafaie et al. / Biomedical Signal Processing and Control 14 (2014) 291–296 293

Fig. 1. An ECG signal: (a) generated by simulation (b) obtained in clinic.

In this estimation technique, the parameters of s2 are changed Table 3


The estimated ECG parameters obtained by adopting the GA.
by GA in a manner that the value of the error function is decreased to
a value lower than predefined threshold, or the value of F reaching Parameters Real values [19] Estimated values
a value higher than predefined threshold. ai [1.20 −5.00 30.00 [1.23 −5.15 29.97
The objective of the GA is to search and find the unknown −7.50 0.75] −7.51 0.73]
parameters (ai , bi , and  i ) in order to maximize the F value. The bi [0.25 0.10 0.10 0.10 [0.21 0.10 0.11 0.10
chromosomes used in this estimation have 112 genes consisting 0.40] 0.38]
i [−0.19 −0.05 0.00 [−0.16 −0.03 0.00
of a1 , a2 , a3 , a4 , a5 , b1 , b2 , b3 , b4 , b5 ,  1 ,  2 ,  4 and  5 parameters,
0.07 0.98] 0.08 0.92]
where the samples are: a1 is the value of the parameter a for the
P-wave, b2 is the value of the parameter b for the Q-wave, b3 is the
value of the parameter b for the R-wave and  4 is the value of the that, GA is a suitable method for estimating the unknown param-
parameter  for the S-wave. In this estimation, since the parameter eters of the ECGs obtained in clinics. Based on the results obtained
 3 representing R-wave position is always zero, it is ignored. Each from this proposed estimation, the ECG signals can be classified
one of these parameters is coded in a binary form and consists of with respect to the estimated parameters.
eight genes. The real values of these parameters are obtained by In this article, the ECG signals of four groups: normal heart
converting these binary values to decimal values. beat (NHB), congestive heart failure beat (CHF), ventricular tach-
The first GA population used in this estimation consists of fifty yarrhythmia beat (VT), and AF (Appendix II), each with seventy
random chromosomes. In this estimation, the values of 0.06 and five ECG signals are collected from PhysioBank; the parameters of
0.6 are selected as the mutation and cross-over rates, respectively, which are obtained through this proposed estimation method. By
according to the rule reported in [24]. For selecting the 0.06 and 0.6 using these values, the membership function of each parameter is
values the trial-and-error method is adopted here. After thirty iter- generated, which contribute to the construction of the fuzzy rules.
ation of the GA, the best chromosome with maximum F is selected These fuzzy rules in turn contribute to the classification of the ECG
as an optimal solution. In order to select a suitable number of signals and prediction of heart diseases.
iterations, some simulations are made at the beginning with fifty
iterations. The simulation results indicate that the optimal chro- 3. Theory and calculation
mosome is found by GA after about 26 iterations, and hence, to be
on the safe side, the thirty iterations is selected. After obtaining the 3.1. Fuzzy classifier and its application in ECG signal classification
optimal chromosome, the optimal values of the parameters of s1 are
obtained. For examining the accuracy of this estimation method, Here a fuzzy classifier is applied in ECG signals. In this fuzzy clas-
an ECG signal of the normal person with known parameters [19] sifier the membership functions are used for describing the input
is selected and this proposed technique is applied for estimating data. The procedure of generating membership functions for ECG
its parameters. The outcome of the thirty iterations, the real values parameters is described here. This procedure is used for generating
and the estimated values of the ECG parameters are illustrated in membership function of all ECG model parameters. These member-
Table 3. In this estimation, parameter i is equal to 65,536 (number of ship functions are obtained after estimating the parameters of one
data points obtained from the ECG signal sampling) and the value of hundred ECG signals with known heart conditions. After identify-
the error function for the optimal chromosome is 0.139, indicating ing these values, according to the upper and lower limits of these
294 M.H. Vafaie et al. / Biomedical Signal Processing and Control 14 (2014) 291–296

Table 4
Upper and lower limits of the estimated ECG parameters.

Parameter NHB VTC AF CHF

Min Max Min Max Min Max Min Max

a1 0.98 1.47 −0.06 0.31 −0.07 0.45 0.68 1.53


a2 −5.39 −4.63 −23.6 −18.33 −5.01 −4.12 −5.98 −5.23
a3 29.38 30.71 28.15 33.03 29.11 31.45 28.98 31.72
a4 −8.33 −6.90 −24.07 −19.03 −8.00 −6.28 −8.67 −5.39
a5 0.31 1.04 −0.19 0.38 0.23 1.40 0.40 0.99
b1 0.06 0.43 0.01 0.03 0.01 0.04 0.04 0.38
b2 0.01 0.22 1.61 1.93 0.01 0.20 0.02 0.28
b3 0.08 0.17 1.21 1.57 0.10 0.21 0.05 0.31
b4 0.01 0.22 1.50 1.77 0.01 0.20 0.02 0.32
b5 0.21 0.62 0.03 0.07 0.24 0.76 0.19 0.45
1 −0.43 −0.07 −0.18 −0.03 −0.21 −0.05 −1.23 −0.76
2 −0.23 −0.05 −0.17 −0.01 −0.11 −0.02 −0.33 −0.10
4 0.02 0.17 0.01 0.10 0.02 0.10 0.06 0.23
5 0.88 1.10 0.31 0.68 0.43 0.55 0.91 1.34

Fig. 2. The membership function of a1 parameter according to the data illustrated in Table 4.

parameters (Table 4), the membership functions are constructed. in a manner where in the defuzzification phase the ECG groups are
After studying the values of a1 parameter of all one hundred sig- segregated perfectly. Furthermore, this membership function must
nals, the membership function of this parameter is generated and be generated in a manner where the incorrect selection of a group
demonstrated as a sample in Fig. 2. is prevented (Fig. 3).
In this proposed fuzzy classifier a set of IF THEN rules is gen- In general, several rules can be corrected under the same cir-
erated based on the study conducted on the features of the same cumstance, while only one group must be selected for unknown
one hundred ECG signals and the constructed membership func- ECG signal. To obtain proper classification, the contribution of
tions. These one hundred ECG signals are in four groups of (NHB), each rule must be combined in an appropriate manner. Here, the
(CHF), (VT), and (AF) mentioned above, where each one consists sup-min (supermom-minimum) composition method is adopted.
of twenty-five ECG signals with known heart conditions in accor- Moreover, for defuzzification of the output, the center of gravity
dance with the PhysioBank database. The form of these fuzzy rules (COG) method, presented in (5) is adopted, as follows:
is as follows: K
i=1
(zi · (zi ))
Rule 1 : IF a1 is ZE AND 3 is PS THEN S is AF. (3) Z∗ =  K
(5)
i=1
(zi )
Rule 2 : IF 1 is NL AND 5 is PL THEN S is CHF. (4)
where, Z* is the value of the output obtained from defuzzification,
where a1 , a2 , a3 , a4 , a5 , b1 , b2 , b3 , b4 , b5 ,  1 ,  2 ,  3 ,  4 and  5 are K is the number of the fuzzy rules, and values of zi and (zi ) are
the ECG parameters, NL, NM, NS, ZE, PS, PM and PL are the fuzzy calculated through the sup-min composition [27]. By applying Z*
sets and S is an ECG signal with unknown features the condition according to Fig. 3 the ECG signals are classified. The performance
of which must be identified. In fuzzy set definition, NS, ZE, PM and of this classifier is investigated in Section 4.
PL are negative small, zero, positive medium and positive large,
respectively. 3.2. Genetic-fuzzy system and its application in ECG signal
After studying all ECG signals with known features, twelve rules classification
are generated. Because the desired output of this proposed fuzzy
classifier is an ECG group and is not a number; therefore, assigning For increasing the accuracy of the fuzzy classifier the GA is
unknown ECG signals to these groups using fuzzy rules must be applied in generating a genetic-fuzzy classifier. In this approach,
made in a proper manner. To do this, one membership function is the fuzzy sets, membership functions and fuzzy rules are opti-
generated for the output and one fuzzy set is defined for each one of mized through GA, where twelve rules with optimum configuration
these ECG groups. Construction of the fuzzy sets must be presented are generated. The simulation results indicate that the accuracy
M.H. Vafaie et al. / Biomedical Signal Processing and Control 14 (2014) 291–296 295

Fig. 3. The membership function of the output.

Table 5 Table 6
Fuzzy classifier results. Genetic-fuzzy classifier results.

ECG groups with 75 members Number of ECG signal(s) placed in each ECG groups with 75 members Number of ECG signal(s) placed in each
group using fuzzy classifier group using genetic-fuzzy classifier

NHB CHF VT AF NHB CHF VT AF

NHB 67 2 4 2 NHB 73 0 0 2
CHF 0 71 2 2 CHF 0 75 0 0
VT 1 1 72 1 VT 0 0 74 1
AF 1 3 1 70 AF 0 0 1 74

of the classification is increased significantly; therefore, the heart other arrhythmia is zero while, 73 signals are placed in NHB group
diseases can be predicted more precisely. The performance of this and 2 signals are placed in AF using the genetic-fuzzy classifier;
classifier is investigated in Section 4. hence, these 2 signals are detected wrongly. Moreover, in the fourth
row the actual number of AF signals is 75 and the actual num-
4. Results and discussion ber of the other arrhythmias is zero while, 74 signals are placed
in AF group and one signal is placed in VT using the genetic-fuzzy
4.1. Fuzzy classifier performance analysis classifier; hence one inaccurately predicted signal.
After investigating the performance of the genetic-fuzzy clas-
After implementing this proposed fuzzy classifier, in order to sifier, the results indicate that the ECG signals are classified with
examine its performance, three hundred ECG signals are selected an accuracy of 98.67% (from three hundred ECG signals, 296 signals
and applied. These signals consist of four ECG groups each with 75 are segregated correctly). This high accuracy percentage means that
members. An ideal classifier can segregate ECG signals perfectly, the unknown ECG signals can be placed in the corresponding ECG
while in practice this is not so, consequently the performance of groups in a more precise manner.
the classifier must be investigated on a constant basis by calculating
the ratio of the ECG signals classified correctly with respect to all 5. Conclusion
ECG signals applied to the classifier.
Here, after applying three hundred ECG signals, the obtained Since in laboratories, arrhythmias are detected by continuous
results indicate that this classifier can classify ECG signals with an monitoring of the ECG signals which is very time consuming,
accuracy of 93.34% (from three hundred ECG signals, 280 signals this article presents a new method in detecting heart diseases
are segregated correctly) (Table 5). in an automatic manner. In automatic arrhythmia detection, the
accuracy is the most important factor. Here, two classifiers are
4.2. Genetic-fuzzy classifier performance analysis presented to increase the accuracy based on the fact that people
with similar heart condition almost have similar ECG signals, so
The above procedure is repeated once more with the same ECG the parameters of the ECG model for them are almost identical. In
signals for the genetic-fuzzy classifier and the result is presented this method the status of a new ECG signal is predicted through
in Table 6. Each row of Tables 5 and 6 is for one ECG group with 75 estimating its parameters and finding the similarities between
members. them and the parameters of the ECG signals with known heart
To describe the results only Table 6 will be analyzed. Each fig- conditions. In order to implement these classifiers, the param-
ure in the related column represents the number of the ECG signals eters of the clinically obtained ECG signals are estimated using
placed in each group using genetic-fuzzy classifier. In the first row GA and then by studying the features of one hundred signals,
the actual number of NHB signals is 75 and the actual number of a fuzzy classifier with twelve fuzzy rules is constructed. Three
296 M.H. Vafaie et al. / Biomedical Signal Processing and Control 14 (2014) 291–296

hundred ECG signals are applied to this classifier to test its per- References
formance. The ECG signals are classified with accuracy of 93.34%.
For increasing the accuracy of the classification, the membership [1] P. McSharry, G. Clifford, L. Tarassenko, Method for generating an artificial RR
tachogram of a typical healthy human over 24-hours, Comput. Cardiol. 29
functions and fuzzy rules are optimized through the GA leading (2002) 225–228.
to the introduction of a newly constructed genetic-fuzzy classifier. [2] D. Kaplan, M. Furman, S. Pincus, Techniques for analyzing complexity in
The classifier and its simulation results indicate an increase in the heart rate and beat-to-beat blood pressure signals, Comput. Cardiol. 3 (1990)
243–246.
accuracy up to 98.67%. By adopting this method, the arrhythmia [3] J.P. Martinez, R. Almeida, S. Olmos, A.P. Rocha, P. Laguna, A wavelet-based
is detected more precisely with high contribution to the clinical ECG delineator: evaluation on standard databases, IEEE Trans. Biomed. Eng.
practice. 51 (2004) 570–581.
[4] K. Minami, H. Nakajima, T. Toyoshima, Real-time discrimination of ventricular
tachyarrhythmia with Fourier-transform neural network, IEEE Trans. Biomed.
Appendix I. Details about the dynamical model of the ECG Eng. 46 (1999) 179–185.
signal [5] Y.H. Hu, S. Palreddy, W. Tompkins, A patient adaptable ECG beat classifier using
a mixture of experts approach, IEEE Trans. Biomed. Eng. 44 (1997) 891–900.
[6] S.Z. Mahmoodabadi, A. Ahmadian, M.D. Abolhasani, ECG feature extraction
The ECG model presented in (1) generates a trajectory in a three- using Daubechies wavelets, in: Int. Conf. on Vis. Imaging and Image Process,
dimensional state-space. Quasi-periodicity of the ECG is reflected 2, 2005, pp. 343–348.
[7] C. Alexakis, H.O. Nyongesa, R. Saatchi, N.D. Harris, C. Davies, C. Emery, R.H.
by the movement of the trajectory around an attracting limit cycle
Ireland, S.R. Heller, Feature extraction and classification of electrocardiogram
of unit radius in the (x, y) plane. Each revolution on this circle cor- (ECG) signals related to hypoglycaemia, in: IEEE Conf. on Comput. in Cardiol.,
responds to one RR-interval or heart-beat. Interbeat variation in 2003, pp. 537–540.
the ECG is reproduced using the motion of the trajectory in the z [8] J.I. Salisbury, Y. Sun, Assessment of chaotic parameters in nonstationary elec-
trocardiograms by use of empirical mode decomposition, Ann. Biomed. Eng. 32
direction. Distinct points on the ECG, such as the P, Q, R, S, and (2004) 1348–1354.
T are described by events corresponding to negative and positive [9] M. Small, D. Yu, R.G. Harrison, C. Robertson, G. Clegg, M. Holzer, F. Sterz,
attractors/repellors in the z direction [22]. Deterministic nonlinearity in ventricular fibrillation, Am. Inst. Phys. 10 (2000)
268–277.
[10] Y. Zhao, J. Sun, M. Small, Evidence consistence with cardiac data: surrogate and
Appendix II. A brief description about the arrhythmia nonlinear dynamical modeling, Int. J. Bifurc. Chaos 18 (2008) 141–160.
[11] M.I. Owis, A.H. Abou-Zied, A.M. Youssef, Y.M. Kadah, Study of features based on
investigated in this paper nonlinear dynamical modeling in ECG arrhythmia detection and classification,
IEEE Trans. Biomed. Eng. 49 (2002) 733–736.
B.1. Atrial fibrillation (AF) [12] K.K. Jen, Y.R. Hwang, ECG feature extraction and classification using cepstrum
and neural networks, J. Med. Biol. Eng. 28 (2008) 31–37.
[13] E.D. Ubeyli, Adaptive neuro-fuzzy inference system for classification of ECG
AF is the most common abnormal heart rhythm. It may cause signals using Lyapunov exponents, J. Comput. Methods Prog. Biomed. 93 (2009)
no symptoms, but is often associated with palpitations, fainting or 313–321.
[14] S. Karpagachelvi, M. Arthanari, M. Sivakumar, ECG feature extraction tech-
chest pain. The cause of an individual’s AF may not be identified. AF
niques – a survey approach, Int. J. Comput. Sci. Inf. Secur. 8 (2010) 76–80.
may be identified clinically when taking a pulse and its presence [15] T.M. Nazmy, H. El-Messiry, B. Al-Bokhity, Adaptive neuro-fuzzy inference sys-
can be confirmed with an ECG that demonstrates the absence of P tem for classification of ECG signals, J. Theor. Appl. Inf. Technol. 7 (2010) 71–76.
waves and an irregular ventricular rate [28]. [16] A. Voss, S. Schulz, R. Schroeder, M. Baumert, P. Caminal, Methods derived from
nonlinear dynamics for analysis heart rate variability, Philos. Trans. R. Soc. 367
(2008) 277–296.
B.2. Congestive heart failure (CHF) [17] K. Stemickel, Automatic pattern recognition in ECG time series, J. Comput.
Methods Prog. Biomed. 68 (2002) 109–115.
[18] J. Fell, K. Mann, J. Roschke, M.S. Gopinathan, Nonlinear analysis of continuous
CHF means the heart does not pump as well as it should to meet ECG during sleep II, dynamical measures, Biol. Cybern. 82 (2002) 485–491.
[19] The MIT-BIH Arrhythmia database. Available from: http://physionet.org/
the body’s oxygen demands, often due to heart diseases such as car-
physiobank/database/mitdb/ [online].
diomyopathy or cardiovascular disease. CHF can result from either [20] E.C. Zeeman, Differential Equations for the Heartbeat and Nerve Impulse, Math-
a reduced ability of the heart muscle to contract or from a mechan- ematics Institute, University of Warwick, Coventry, UK, 1972.
ical problem that limits the ability of the heart’s chambers to fill [21] A. Ayatollahi, N. Jafarnia Dabanloo, D. McLemon, V. Johari Majd, H. Zhang, A
comprehensive model using modified Zeeman model for generating ECG sig-
with blood. CHF occurs most frequently in those over age 60 and nals, Iran. J. Electr. Electron. Eng. 1 (2005) 88–93.
is the leading cause of hospitalization and death in that age group. [22] P. McSharry, G. Clifford, L. Tarassenko, A dynamical model for generating syn-
In over 50 percent of cases, sudden death occurs due to a cardiac thetic electrocardiogram signals, IEEE Trans. Biomed. Eng. 50 (2003) 289–294.
[23] G. Clifford, P. McSharry, A realistic coupled nonlinear artificial ECG, BP and
arrhythmia, or irregular heartbeat [29]. respiratory signal generator for assessing noise performance of biomedical sig-
nal processing algorithms, in: Proc. of SPIE Int. Symp. on Fluct. and Noise, 2004,
pp. 290–301.
B.3. Ventricular tachycardia (VT) [24] M. Gen, R. Cheng, Genetic Algorithm and Engineering Optimization, John Wiley
& Sons, New York, USA, 1999.
Ventricular tachycardia or ventricular fibrillation (VF) refers to [25] Y. Goletsis, C. Papaloukas, D.I. Fotiadis, A. Likas, L. Michalis, Automated ischemic
beat classification using genetic algorithm and multicriteria decision analysis,
any rhythm faster than 100 (or 120) beats/min, with 3 or more IEEE Trans. Biomed. Eng. 51 (2004) 1717–1725.
irregular beats in a row, arising distal to the bundle of His. The [26] N.G.B. Amma, Cardiovascular disease prediction system using genetic algo-
rhythm may arise from working ventricular myocardium, the dis- rithm and neural network, Comput. Commun. Appl. 5 (2012) 1–5.
[27] T.W. Chua, W.W. Tan, Non-singleton genetic fuzzy logic system for arrhythmia
tal conduction system, or both. During VT, the following may be classification, Eng. Appl. Artif. Intell. 24 (2011) 251–259.
observed: hypotension, tachypnea, high jugular venous pressure [28] M. Drinnan, J. Allen, A. Murray, Relation between heart rate and pulse transit
and variation in intensity of the first heart sound, caused by loss of time during paced respiration, J. Physiol. Meas. 22 (2001) 425–432.
[29] http://www.hopkinsmedicine.org/heart vascular institute
atrioventricular (AV) synchrony [29].

View publication stats

You might also like