You are on page 1of 38

Accepted Manuscript

Automated Characterization and Classification of Coronary Artery
Disease and Myocardial Infarction by Decomposition of ECG Signals:
A Comparative Study
U Rajendra Acharya , Hamido Fujita , Muhammad Adam ,
Oh Shu Lih , Vidya K Sudarshan , Tan Jen Hong , Joel EW Koh ,
Yuki Hagiwara , Chua K. Chua , Chua Kok Poo , Tan Ru San

INS 12572

To appear in:

Information Sciences

Received date:
Revised date:
Accepted date:

23 June 2016
13 August 2016
5 October 2016

Please cite this article as:
U Rajendra Acharya ,
Hamido Fujita ,
Muhammad Adam ,
Oh Shu Lih , Vidya K Sudarshan , Tan Jen Hong , Joel EW Koh , Yuki Hagiwara , Chua K. Chua ,
Chua Kok Poo , Tan Ru San , Automated Characterization and Classification of Coronary Artery Disease and Myocardial Infarction by Decomposition of ECG Signals: A Comparative Study, Information
Sciences (2016), doi: 10.1016/j.ins.2016.10.013

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service
to our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and
all legal disclaimers that apply to the journal pertain.









Graphical Abstract









 Classification of three types of ECG beats are proposed.
 Normal, CAD and MI are three classes considered in this work.
 DCT, DWT and EMD decomposition methods are employed on ECG beat.
 Decomposed signals are fed to Locality Preserving Projections method.
 KNN classifier yielded 98.5% accuracy with seven features for DCT method.

Singapore. University of Malaya. Vidya K Poo1. Joel EW Koh1. Singapore M 5 AN US 1 PT ED *Corresponding Author Postal Address: Iwate Prefectural University (IPU). Iwate 020-0693 Japan Telephone: +81-19-694-2578. National Heart Centre. 2Department of Biomedical Engineering. Malaysia 4Iwate Prefectural University (IPU).org ABSTRACT CE Cardiovascular diseases (CVDs) are the main cause of cardiac death worldwide. Email Address: HFujita-799@acm. the presence of these cardiac conditions is primarily reflected on the electrocardiogram (ECG) signal. if not diagnosed and treated at an early stage may eventually lead to an irreversible state of heart muscle death called Myocardial Infarction (MI). SIM University. Thus.ACCEPTED MANUSCRIPT Automated Characterization and Classification of Coronary Artery Disease and Myocardial Infarction by Decomposition of ECG Signals: A Comparative Study U Rajendra Acharya1. 3Department of Biomedical Engineering. it is challenging and requires rich experience to manually interpret the visual subtle changes occurring in the ECG waveforms. Iwate 020-0693 Japan Department of Cardiology. School of Science and Technology. Tan Jen Hong1. CAD AC condition progresses rapidly. Singapore. Chua K Chua1. Yuki Hagiwara1. However. Hamido Fujita4*. Muhammad Adam1. Faculty of Engineering. Tan Ru San5 CR IP T Sudarshan1. The Coronary Artery Disease (CAD) is one of the leading causes of these CVD deaths.3. Chua Kok Department of Electronics and Computer Engineering.2. Faculty of Software and Information Science. many automated diagnostic systems are developed to overcome . Oh Shu Lih1. Ngee Ann Polytechnic. Normally. Faculty of Software and Information Science.

which accounts for 31% of global death.ACCEPTED MANUSCRIPT these limitations. INTRODUCTION ED cosine transform. Buja et al. Then. in which fibrous plaques begin to form thick regions in the inner wall of the arteries (Willerson et al. CAD is primarily the result of atherosclerosis. empirical mode decomposition. discrete wavelet transform. the LPP features are ranked using F-value. 1990. In 2012. These coefficients are reduced using Locality Preserving Projection (LPP) data reduction method. DWT CR IP T and EMD to obtain respective coefficients. 2014).5% accuracy. electrocardiogram. a total of 17. Finally. Our proposed system yielded highest classification results of 98. CAD is the leading cause of death in the United .. cardiovascular diseases (CVDs) remain as the number one cause of cardiac death. 99. Empirical Mode Decomposition (EMD) and Discrete Cosine Transform (DCT) are compared. discrete PT 1.. 1987. These atherosclerotic plaques are basically made up of fats and fibrous tissues (Buja et al. ECG signals are subjected to DCT.. The screening system can help the cardiologists in detecting the CAD and hence presents any possible MI by prescribing suitable medications. CE Globally. Buja et al. old age homes.. It can be employed in routine community M screening. polyclinics and hospitals.5% specificity using only seven features obtained using DCT technique.. myocardial infarction. Keywords: coronary artery disease. an approximately 7. 1982. the performance of an automated diagnostic system developed for detection of CAD and MI using three methods such as Discrete Wavelet Transform (DWT).5 million deaths due to CVDs are reported AC worldwide. Buja et al.4 million deaths are due to Coronary Artery Disease (CAD) (Mendis et al. In this study. the highly ranked coefficients are fed into the K-Nearest Neighbor (KNN) classifier to achieve the best classification performance. Of these. 2007). In this study. 2007).7% sensitivity and AN US 98..

5% of women are affected with MI. Clinically. 2014).. first MI. rupturing of these atherosclerotic plaques leads to acute Ischemic Heart Disease (IHD) that comes with spectrum of Acute Coronary Syndrome (ACS). the dead cardiac muscle tissue is replaced with scar tissue that has no myocardial contractility. infarction or arrhythmias due to CAD (Katz et al.000 silent MIs happens ED every year (Mendis et al. In severe condition.. the T waves appear abnormally high and the QT intervals are longer . In other words. In addition.000 Americans are experiencing fresh coronary attack. Thus.. 16% and 10% of male and female deaths respectively are due to CAD. Chee et al. MI irreversible muscle death occurs when there is a disruption or reduction of blood flow to the heart muscles (myocardium) (Chee et al.. 2014). Within seconds of MI. Herring et al. The ECG signal provides vital information with regard to the function and rhythm of the heart. With severe and prolonged MI.. the AN US pathogenesis for majority of MI conditions is caused by atherosclerotic plaques that occlude the local coronary artery (Guyton et al. 2006). approximately 74000 deaths in total (Townsend et al.000 are having repeated episodes. US reported mortality of 370213 due to CAD. The morphological changes in the ECG signal are observed as the heart experiences episode of ischemia. Townsend et al. 2007). approximately 1% of men and 0. 2012.. Every year..ACCEPTED MANUSCRIPT Kingdom (UK) with close to half (45%) of CVDs death are due to it. 2015. These epidemiological studies on CAD and MI indicate the significance for the CE need of an early detection system. In 2012.000 people are hospitalized PT due to MI. which is equivalent to 1 in every 7 Americans (AHA CR IP T 2016).. 660. in UK.. Electrocardiogram (ECG) is the most AC commonly preferred diagnostic tool due to its non-invasiveness. 160. Levin 2013. 2006. As of 2012. a total of 175. Evidently. in particularly Myocardial Infarction (MI). which is around 1 in every 3 minutes (Townsend et al. myocardium permanently loses its contractility at the M region of MI. and 305. It is reported that. 2007). standard or resting ECG can be a convenient noninvasive diagnostic method for detecting the CAD and MI. In 2013. 2014). and low cost.

. 2016). 2008) or ECG signals (Rahhal et al. 2012..ACCEPTED MANUSCRIPT than normal (Chee at al. 2013. Arif et al. The proposed study compares the performance of three different techniques namely.. about 70% of CAD patients do not exhibit any significant variations in their CR IP T ECG signals (Silber et al. Many signal processing algorithms such as linear (time and frequency domain) M (Al-Kindi et al. 2014). Jayachandran et al. . Safdarian et al. Arif et al...... an automated diagnostic system is essential which can overcome these limitations of manual evaluation of ECG signals (Martis et al. DWT (Desai et al. Lee et ED al. 2015) have been developed to accurately extract features from Heart Rate (HR) or ECG signals for the classification of CE normal or abnormal (CAD or MI) signals (Table 1)... 2014b). Sun et al.... Summary of various detection and diagnosis studies of CAD and MI is tabulated in Table 1. Also.. However.. automated classification of heart rate (Acharya et al. Thus. thus. 2012. CAD and MI) using ECG signals. Kaveh et al. To best of our knowledge. 2010. 1975).. nonlinear methods (Lee et al. 2016) and Discrete Cosine Transform (DCT) (Desai et al. 2014.. in detection of normal. 2007. 2016) for the detection of CAD and MI have been developed. this is the first study on the characterization and classification of three classes (normal.. Empirical Mode Decomposition (EMD) (Desai et al... AC Thus. Therefore. Safdarian et al. Lu et al.. 2007. Kim et al. 2000. 2013. it is tedious to manually or visually examine and infer these subtle morphological changes in long continuous ECG beats. 2016). 2008. Discrete Wavelet Transform (DWT) (Giri et al. In contrast. 2007). AN US Over the years. the elevation of ST segment is an indication that the heart is experiencing serious and extensive myocardial ischemia. Acharya et al... 2007).. 2011). it is evident from the literature review table 1 that all the studies reported focuses on automated classification of two classes (normal and CAD or MI). Banerjee et al. this process is time consuming and prone to errors due to fatigue. 2016).. sub endocardial ischemia is reflected as a ST segment depression (Chee et al. 2011. and PT Tunable Q Wavelet Transform (TQWT) (Patidar et al. 2012.

DCT and EMD from the pre-processed ECG beats are reduced and ranked using LPP and F-value respectively.. The highly CR IP T ranked features are classified using K-Nearest Neighbor (KNN) classifier.5% accuracy.ACCEPTED MANUSCRIPT CAD and MI classes using ECG beats. the highly ranked features are fed to the KNN classifier independently for the characterization or normal. . 99. Subsequently. 2. A total of 2. 148 MI and 52 Normal subjects ECG records were used. The proposed system yielded a highest classification performance of 98.2 Methods PT 7 CAD. Then. the ECG signals were obtained from Physionet open access databases. CAD and MI ECG signals. CE System sketch of the proposed methodology is shown in Figure 1. 2000). The robustness of the proposed system is justified with ten-fold cross validation technique. DCT and EMD) are applied to Locality Preserving Projection (LPP) reduction method. the reduced features are ranked using Analysis of Variance (ANOVA) statistical analysis technique.1 Materials For this study. M The ECG signals of Normal and MI were acquired from PTB Diagnostic ECG Database while the ECG signals for CAD were acquired from St. The features AC extracted using DWT.5% specificity using only seven features obtained from DCT AN US method. METHODOLOGY 2. The features (coefficients and IMFs) obtained from the respective transformation techniques (DWT.-Petersburg Institute of ED Cardiological Technics 12-lead Arrhythmia Database (Goldberger et al.7% sensitivity and 98.

.000 Hz for MI and normal classes. A total of 92273 ECG . In this study. 2.. The Rpeak is chosen as the distinctive point because of its visible high amplitude...2.AN US CR IP T ACCEPTED MANUSCRIPT Figure 1: System sketch of the proposed system. Iyengar et al. 2000. Pan et al.1 Pre-processing M The ECG signals from the databases (Goldberger et al.. 2006). using the detected R-peaks. sampling frequency of 1000 samples per second is chosen for this study. The segmentation of the ECG beat is done by taking 250 and 400 samples before and after the R-peak respectively. 2013). 1996) are ED sampled at 257 Hz for CAD and 1. 2. 2012. without the inclusion of the first and last beats. all the ECG signals were segmented. ECG signals are preprocessed to remove the baseline wander and CE PT noise using Daubechies-6 (db6) mother wavelet (Martis et al. In order to ensure uniformity across the databases.2.2 Beats Segmentation AC The ECG beat segmentation of the pre-processed ECG signals is subjected to R-peak detection using Pan-Tompkins technique (Martis et al.

CR IP T Table 2: Overview of the ECG beats used. MI and CAD. . Three techniques used are briefly described in the following sections. 41545 CAD and 40182 MI. MI. and CAD beats segmented from continuous ECG signals. DWT and EMD methods separately to obtain their coefficients and IMFs respectively.2. The summary of number of ECG beats used in each classes is shown in Table 2. Figure 2 shows the normal. 2. Number of Beats Class Normal 10546 MI 40182 CAD 41545 92273 PT ED M AN US Total Beats AC CE Figure 2: Typical ECG beat: normal.3 Feature Extraction Each ECG beat is subjected to DCT.ACCEPTED MANUSCRIPT beats were used consisting of 10546 normal.

Discrete Wavelet Transform (DWT) AC The DWT technique converts the time domain signal into wavelet domain to obtain the time and frequency values in terms of coefficients (Addison 2005. each ECG beat of 651 samples are compressed to of 217 DCT CE PT coefficients. In general. the DCT (Ahmed et al.. The output coefficients of the HPF are called detail coefficients and LPF are called approximations. it is clear that when ( √ M ( ) √ ) .m=0 . Olkkonen 2011). In DWT. ( ) is defined as: ( ) From (1). A total of 217 X 92273 DCT coefficients are obtained for three classes. the ECG signals undergo high pass filtering (HPF) and low pass filtering (LPF) to decompose the signals into different scales (frequency bands) (Daubechies 1990). In this study. the first DCT ED coefficient is the average value of the ECG beat samples.ACCEPTED MANUSCRIPT Discrete Cosine Transform (DCT) The DCT technique has the ability of energy compaction within a restricted number of DCT coefficients. 1974) of an ECG beat of length N is commonly defined as: ( )∑ ( ) [ ( ) ( ) ] (1) CR IP T ( ) The inverse DCT is defined as: ∑ ( ) ( ) [ ] (2) AN US ( ) For both equations (1) and (2). Hence. Then approximations are further subjected to decomposition using .m √ ∑ 0 ( ) .

AC 1) The total number of zero crossing and extrema have to be equal or at least differ by one and. IMFs are obtained for a signal (Huang et al. At each level of decomposition. Thus.. a total 17 X 92273 DWT coefficients are obtained for the three Empirical Mode Decomposition (EMD) AN US ECG classes. 1998). . Intrinsic Mode Functions (IMFs) that has its own characteristic oscillations (Huang et al. 1998. Martis et al.. the samples of a signal gets reduced due to sub-sampling by 2 (Ratnakar et al. 2015. 1998)... ED initially for a signal... the DWT is performed up to six levels of decomposition using CR IP T Daubechies 4 (db4) mother wavelet (Martis et al.. In this work. 2) The mean value of the envelope that is define by the local minima and the envelope of the local maxima have to be zero. The basis of EMD is to decompose a signal x (t) into sets of frequency and amplitude modulated signal components.ACCEPTED MANUSCRIPT both HPF and LPF and thus this procedure is repeated for different levels. obtain local maxima and local minima. Pachori et al. Later by joining all the maxima together and minima together obtain the envelopes for both respectively. 2012). 2014a). 17 detail coefficients are obtained for each ECG beat of 651 samples/length. 1998).. 2009). The EMD methodology is described in (Huang et al. This IMF extracted is tested to confirm whether it CE meets the following two conditions or not. M Using an iterative method known as sifting method. In this sifting method. At sixth level of decomposition. PT Finally extract the IMF from a signal by subtracting the average of envelopes (maxima and minima) (Huang et al.

features with high F-value and p-value <0. the LPPs are calculated by obtaining the optimal linear ED approximations to the Eigen functions of the Laplace Betrami operator (He et al.ACCEPTED MANUSCRIPT Once the conditions are satisfied. CR IP T In this study. 1998. In this work. This technique optimally protects local neighborhood details of the data.. In this method. now consider the residual signal as a new signal and repeat the above mentioned procedure. AN US 2. 2005).2.. for high dimensional data. This is to determine the difference between the three classes AC according to the F-value (Fisher’s discriminatory value) and p-value (probability value).4 Features Reduction . PT 2. 2.2. In this study. So. To obtain the subsequent (remaining) IMFs. The transformation matrix is M calculated using Laplacian of the graph which maps the data to a subspace. Thus. the features are ranked based on F-value and the highly ranked features are then fed into the classifier independently. all the features extracted undergo Analysis of Variance (ANOVA) CE statistical test (Duda 2012). the first IMF of a signal is obtained. Perform the procedure until further no more IMFs can be extracted from the residual signal (Huang et al. EMD is performed up to two levels and used only second level IMF (IMF2)..6 Classification – K-Nearest Neighbor (KNN) . 651 IMF coefficients are extracted from each ECG beat. 2005. In fact. a total of 651 X 92273 IMF coefficients are obtained using EMD for the three classes of ECG beats in this work. 2012). a graph is built by including the neighborhood details of the data set.2.0001 indicate good discrimination characteristics.Locality Preserving Projections (LPP) It is a linear dimensionality technique proposed by He et al..5 Features Ranking – ANOVA Statistical Analysis In this study. Martis et al.

ranked in descending order of F value obtained from ANOVA statistical analysis. three different transformation techniques are implemented. 17 M X 92273 DWT coefficients and 651 X 92273 IMF coefficients are obtained for three classes. the coefficients and IMFs extracted are reduced using LPP method and ED subsequently.ACCEPTED MANUSCRIPT In KNN classification. 17 DWT coefficients and 651 IMFs from each ECG beat of 651 samples. Further. In Figure 3 and Table 3 (Appendix) we can see twelve highly ranked LPP AC CE PT coefficients. For each ECG beat. a total of 92273 ECG beats of three classes (10546 normal. 3. each of the unknown samples are classified based on the majority known k nearest samples (Han et al.. 2006. RESULTS In this study. . 2016). This method is non-parametric because no assumptions CR IP T are made for the data distribution. Gonzalez et al. namely DCT. In this study. We have obtained 217 DCT coefficients. It is a type of nonparametric lazy learning technique whereby the function is locally estimated and computed in the recall phase. variation of k nearest neighbors (k) are varied between 5 to 10. A total of 217 X 92273 DCT coefficients. 40182 MIs AN US and 41545 CADs) are segmented from two separate ECG signal databases. The highest accuracy is achieved for k = 5 in this work. and obtained using DCT method. DWT and EMD methods..

in Figure 4 and Table 4 (appendix) twelve highly ranked LPP coefficients are AC CE PT ED presented and obtained using DWT method. it is evident that. there is a distinct discrimination between the three classes.CR IP T ACCEPTED MANUSCRIPT AN US Figure 3: Bar graph of ranked LPP feature values obtained using DCT technique. From Figure 3 and Table 3 (appendix). based on the mean and SD values of the features. M Likewise. . Figure 4: Bar graph of ranked LPP feature values obtained using DWT technique.

46% specificity using only seven features. The highly ranked features from the respective techniques are fed independently AC to KNN classifier.72% sensitivity and 98. in Figure 5 and Table 5 (appendix) twelve highly ranked LPP coefficients are M AN US CR IP T presented and obtained using DWT method. Ten-fold cross validation is implemented to ensure that the results obtained are reliable. the integration of DCT technique into the proposed system yielded the highest performance results of 98.5% accuracy. PT It is evident from Figure 5 and Table 5 that the mean and SD value discrimination CE between the three classes is not as strong as DCT and DWT techniques. Likewise.ACCEPTED MANUSCRIPT From Figure 4 and Table 4 (appendix). Sketch of variation of accuracies (%) for different number of features for DCT method using KNN classifier is . As shown in Table 6. 99. ED Figure 5: Bar graph of ranked LPP feature values obtained using DWT technique. it is evident that. there is a distinct discrimination between the three classes. based on the mean and SD values of the features.

58 CR IP T Method CE PT ED M AN US Acc = accuracy. Table 6: Classification results of the DCT.96 Sen (%) 99.77 93.16 81.72 99.80 99. DWT and EMD techniques. Table 7: Confusion matrix of the 3 classes.25 51. PPV = positive predictive value (Akobeng 2006).69 97. Sen = sensitivity. Original/ Predicted Normal MI CAD .23 Spec (%) 98. Spec = specificity.ACCEPTED MANUSCRIPT presented in Figure 6. AC Figure 6: Sketch of accuracies (%) for different number of features using DCT method with KNN classifier.50 98. Table 7 shows the confusion matrix of our work using KNN classifier for DCT method.46 98. No of Features DCT DWT EMD 7 12 9 Acc (%) 98.34 PPV (%) 99.

. This decomposition method results in fewer coefficients without any loss of signal information or energy during both wavelet and inverse transform (Abo-Zahhad 2011. 4. DWT and EMD techniques for the diagnosis of CAD and MI using ECG beats are compared.7% sensitivity and 98.ACCEPTED MANUSCRIPT Normal 10384 96 66 MI 133 39642 407 CAD 94 586 40865 From Table 7... . Allen et al. 99. 40182 MI and 41545 CAD) beats are used. 2010). In this study.. Desai et al. 651 samples of the individual ECG beat are compressed to 217 DCT coefficients (one-third).5% specificity using M only seven features by employing DCT technique.. AC DWT is also used for ECG data compression due to its non-stationary and localized properties (Abo-Zahhad 2011). 1974. Hence. 2011. DWT has the ability to analyze ECG signal at different resolution by breaking down the ECG signal into several frequency bands (Banerjee et al. 2008). DISCUSSION AN US In this study. performances of DCT. 2012). Alam et al. DCT is a class of orthogonal transforms used for ECG data compression due to ED its energy compaction and decorrelation properties (Bendifallah et al. 1992). it can be observed that majority of the beats are correctly classified into CR IP T their respective classes and only few beats are incorrectly classified and treated as outliers. 92273 (10456 normal.. the proposed system yielded a highest classification accuracy of 98. allowing classification to be performed with only few CE significant features.5%. In this study. Overall. It has the capabilities to concentrate most of the ECG signal information within a PT small subset of the transform coefficients with insignificant error (Ahmed et al. Moreover.

this is the first study involving the classification of three classes achieving high classification performance using 92273 ECG beats with just seven features. the proposed diagnosis system is insensitive to the ECG signal noise. in this work. DCT method yielded highest classification accuracy due to its energy compaction AC CE Our proposed system has the following key advantages: a) As summarized in Table 1. However. the accuracy achieved depends on the assumption that the shape of the basis function is similar to that of the ECG signal (Muhidin et al. . have classified five arrhythmias using DCT-PCA method with highest performance (Martis et al. This helps to capture the subtle changes in the ECG signal. 2014). The characteristic oscillation of the individual IMFs are presented on separate time scale..ACCEPTED MANUSCRIPT EMD is signal dependent and adaptive approach. 1998). 2014a).. Hence. using each ECG property. in our classification also. b) The DCT method aids in noise suppression or removal. 2013b).. 1998). the analysis of IMFs together will provide instantaneous CR IP T frequency information with respect to time (Huang et al. In the previous work by ED Martis et al. predicting the shape of the basis function is challenging especially for nonstationary and nonlinear signal. which could be related to a particular type of disease.. This is because the exact time and time-scale that M the signal changes are not known (Martis et al.. Based on the results. The ECG signal is decomposed into a finite numbers of different IMFs. Moreover. AN US In the case of DWT. which is highly efficient (Huang et al. 2013). we have used Daubechies 4 (db4) mother wavelet (Martis et al.. However. PT beat. the proposed system performed well using DCT and DWT techniques and achieved significantly better accuracy than EMD method.

only 133 MI and 94 CAD ECG M beats are wrongly classified as normal.5% specificity by using seven features using DCT technique. repetitive. non-invasive. fast and simple to use. This clearly shows that. and coronary angiogram etc. PT Cardiovascular diseases are the leading cause of death in the world. The proposed system provides a reliable and accurate detection of CAD and MI (high classification performance).7% AC sensitivity and 98. CAD and MI classes. so that the clinician can further investigate the subject when a beat is classified as either CAD or MI.35%). performance of diagnostic support system developed using DCT. CONCLUSION ED robust and number of false positive is very less (<0. we are able to detect the CAD with an accuracy . In this study. our method is 5. AN US e) Proposed technique is robust and reliable due to the 10-fold cross validation technique. f) Another important finding of our work is that.5% accuracy. The most important finding of this work is that. our technique is cost effective and fast as compared to other cardiac diagnostic imaging tools Magnetic Resonance Imaging (MRI). and CE EMD methods are compared in the classification of normal. The proposed system yielded highest classification performance of 98.ACCEPTED MANUSCRIPT c) The proposed system is completely automated. Intravascular Ultrasound (IVUS). Thus. DWT. 99. d) The main application of our proposed method is to spot abnormal beat in a long ECG recording. it can be used in hospitals to aid the clinicians in their CR IP T diagnosis. In addition.

This contributes in preventing patients from having MI and save life AC CE PT ED M AN US significantly.23% of CAD ECG signals and 0. authors aim extend their work for different stages of CAD CR IP T and MI.ACCEPTED MANUSCRIPT of 98.75%. In future study. Only 0. . This performance can be further improved by using more subjects in each class and better nonlinear features as parameters for the classifiers.33% of MI ECG beats are wrongly classified as normal ECG signals (Table 7).

94 Normal Source: supine.8% Sensitivity = 100% Specificity = 93. Calicut. Time domain. India Lead: Lead II Subjects: 10 CAD. 2015 PT ED M AN US Lee et al. 10 Normal Source: Iqraa hospital. Approximate entropy SVM.CR IP T ACCEPTED MANUSCRIPT Table 1: Overview of study using HRV and ECG signals in the diagnosis of CAD and MI. 64 CAD (51 angina pectoris. Poincare plot.90% Giri et al. Approximate entropy Support vector machine (SVM) Classification based on multiple association rules (CMAR) Naïve Bayesian (NB) Decision tress (DT) SVM: Accuracy = 90% Kim et al. Fractal scaling measures. right lateral. 94 Normal Source: supine. Hurst exponent. left lateral positions Frequency domain. 2007 Lead: 12 leads Subjects: 20 Normal. right lateral. Poincare plots. 2007 . Time domain.7% Accuracy = 99. 15 Normal Source: Iqraa hospital. right lateral. DFA. India DWT SVM. 2013 Lead: Lead II Subjects: 10 CAD.6% Specificity = 99. Poincare plot. Kerela. Gaussian mixture model (GMM) K-Nearest Neighbors (KNN) Probabilistic neural network (PNN) TQWT and CC Least squares SVM (LS-SVM) GMM classifier: Accuracy = 96.5 – 84. CMAR. MDA. 2008 Lead: lead II Subjects: 99 CAD. left lateral positions Frequency domain. DT and Classification based on predictive association rules (CPAR) CPAR & SVM: Accuracy = 85. left lateral positions Frequency domain. Calicut.8% AC CE Patidar et al. Author (Year) Database Techniques/ Features Normal and CAD using HRV signals Efficiency Lead: lead II Subjects: 99 CAD. 13 acute coronary syndrome) Source: supine.7% Sensitivity = 99. NB. Complexity estimations Multiple discriminant analysis (MDA) Accuracy = 72. Time domain. Kerela.6% Lee et al.

Calicut. 2005 PT Sood et al. 2001 Lead: 12 leads Subjects: 479 CAD. LS-SVM Accuracy = 100% Sensitivity = 100% Specificity = 100% No accuracy reported. 2014 Lead: Lead II Subjects: 10 CAD. Proposed range of parameters for two classes with (p<0. 1988 Lead: Lead V3 to V4 Subjects: 62 men. Nonlinear methods (Poincare geometry. amplitude Source: Iqraa Hospital.2016 Acharya et al. EMD) Lead: Lead II EMD / second-order difference plot area. 220 Normal Source: Bicycle exercise ECG test Men: Sensitivity = 84.5% Lewenstein et al. India Lead: Lead II Subjects: 10 normal 10 CAD ECG signals Source: Iqraa Hospital. DFA. 10 Normal Source: Iqraa hospital. 31 women (21 CAD & 10 normal) Source: resting ECGs Lehtinen et al. Proposed range of parameters for two classes with (p<0. 1998 Lead: 9 leads Subjects: 127 CAD. 2010 M BCT Blinded test Fisher’s exact test ED CE Arafat et al.05) Schreck et al. India modulation bandwidth. 297 Normal Source: Exercise stress ECG Lead: lead v5 Subjects: 8 CAD.05) RBF neural networks Average specificity and sensitivity of about 97% Fuzzy uncertainty Probabilistic uncertainty Combined uncertainty PCA. correlation dimension. SVM Combined Uncertainty: 80% correct classification percentage (CCP) Accuracy = 79. Calicut. (51 CAD & 11 normal). sample entropy. frequency modulation bandwidth. Recurrence quantification analysis. 2016 AN US Time domain and frequency domain analysis. Kerela. Subjects: 10 normal 10 CAD ECG signals analytical signal representation area.17% . India CR IP T ACCEPTED MANUSCRIPT Flexible Analytic Wavelet Transform and Entropy Features. approximate entropy.3% Specificity = 81... Calicut.8% Women: Sensitivity = 76. HOS. Fourier-Bessel expansion Normal and CAD using ECG signals Artificial neural network (ANN) Receiver operating characteristics (ROC) analysis No accuracy reported.2% Specificity = 80% ROC = 91.Kumar et al. 7 Normal Source: Exercise stress ECG Lead: 12 leads Subjects: 480 patients Source: Exercise stress ECG AC Babaoglu et al.

Petersburg Institute of Cardiological Technics 12-lead Arrhythmia database.CR IP T ACCEPTED MANUSCRIPT Lead: 12 leads Subjects: 480 patients Source: Exercise stress ECG Binary particle swarm Genetic algorithm SVM Yin et al.17% Sensitivity = 94.39% Lu et al. 2013 AN US Babaoglu et al. 2000 Lead: 12 leads Beat(s): ST-segment & T wave Subjects: 104 MI.6% Specificity = 90% . 2011 Lead: 12 leads Subjects: 46 Source: Long-term ST database (motion activity) Lead: Lead II Beat(s): raw ECG waveform Subjects: 43 CAD. 49 Normal Source: Multi-parameter Intelligent Monitoring in Intensive Care (MIMIC II) Wavelet decomposition R-wave peaks detection ST segment detection DWT.99% Sensitivity = 97.34% Using 31 cumulant features: Accuracy = 98. 20 Normal AC CE PT ED M Kaveh et al. Specificity = 100% Testing set: Sensitivity = 84. HOS bispectrum and cumulants PCA KNN and DT Using 13 Bispectrum features: Accuracy = 98. 2017 Lead: 12 leads Beat(s): 182013 ECG beats Subjects: 40 normal 7 CAD Source: St..57% Specificity = 99. Fantasia open access database. 2010 Normal and MI using ECG signals ANN Fuzzy logic system Binary particle swarm optimization – feature selection technique: Accuracy = 81.75% Specificity = 99.46% Accuracy = 80% Training set: Sensitivity = 94.2%. PCA and SVM Accuracy = 88% Acharya et al.

6 Normal DWT.CR IP T ACCEPTED MANUSCRIPT Lead: 12 leads Beat(s): 64680 R-peaks Subjects: 177 MI.7% . 2010 Lead: lead II Beat(s): 2282 Normal. ST level deviation PCA.6% MIT BIH database Sensitivity = 99.1% Localization: Accuracy = 93. V4 Beat(s): 582 MI. 718 MI Subjects: 2 MI.5% Specificity = 99. 2009 Accuracy = more than 95% Denoise –DWT QRS complex band selection R peak detection Baseline detection QRS vector computation PTB database Sensitivity = 99. 125 Normal Detection of R-peaks Phase space fractal dimension (PSFD) Neuro-Genetic algorithm classifier Efficiency = 96% Chang et al. Energy-entropy Energy and entropy plots Banerjee et al. 1840 Normal Subjects: 148 MI. Q wave. 2010 PTB database Lead: leads V1-V4 Beat(s): 42852 Subjects: 5 patients M ED PT MIT BIH arrhythmia database Lead: leads V1-V4 Beat(s): 19062 Subjects: Lead: 12 leads Beat(s): 10580 MI. V2. 52 Normal (594 records) AC CE Arif et al.8% T wave amplitude.5% Jayachandran et al. 547 Normal Subjects: 1129 samples of heartbeats from clinical data ECG beat sampling Hidden markov models Gaussian mixture models Hidden markov models: Sensitivity = 79% Accuracy = 71. 2010 AN US Lahiri T et al. V3. Back propagation neural networks Detection: Sensitivity = 97. 2009 Lead: lead V1.

2012 ED M Arif et al.4% Q wave amplitude ST level deviation T wave amplitude KNN classifier MI detection: Sensitivity = 99. 20 Normal R-complex time index detection Q & S time index detection J-point time index detection Reference point detection (REF) ST segment analysis Sensitivity = 85% Specificity = 100% Banerjee et al.97% Specificity = 99.9% MI localization: Sensitivity & specificity = more than 99% Accuracy = 98.CR IP T ACCEPTED MANUSCRIPT Lead: 12 leads Beat(s): ST segment analysis Subjects: 20 MI. 1 Normal Lead: Lead 3 Beat(s): 1 beat Subjects: 148 MI. 2011 PTB database Lead: leads V1-V4 Beat(s): 42852 Subjects: 5 patients DWT QRS complex band selection & finding the QRS window R peak detection Q & S point detection T wave detection QRS vector computation Mahalanobish distance based classification MIT BIH Arrhythmia database: Sensitivity = 99. 2011 R peak registration Time normalization of cardiac cycles Cross transform and Wavelet coherence of ECG beats . MIT BIH arrhythmia database Lead: leads V1-V4 Beat(s): 19062 Subjects: Lead: 12 leads Beat(s): 16960 MI. 3200 Normal Subjects: 10 types of MI. 52 Normal AC CE PT Banerjee et al.8% PTB database: Sensitivity = 99.8% ECG patterns shows difference characteristics over QRS and T waves regions.84% Accuracy: 96. 2012 AN US Al-Kindi et al.

NB Localization: Accuracy = 76% Detection: Accuracy = 94% Liu et al. 125652 Normal Subjects: 148 MI of 10 types. 79 normal ST segment Polynomial fitting KNN ensemble Safdarian et al. Kolmogrov-Sinai.74% Sensitivity = 99. 2015 Lead: 12 leads Beat(s): Subjects: 148 MI.58% MI detection: Accuracy = 98. 52 Normal AC CE PT ED M AN US Sun et al. KNN. Multilayer Perceptron (MLP). 2014 Lead: lead 2 Beat(s): 1 beat Subjects: 290 subjects (549 records) T-wave integral ANN.4% Sharma et al. 2016 Lead: 12 leads Beat(s): 485753 MI.CR IP T ACCEPTED MANUSCRIPT Lead: 12 leads Subjects: 369 MI.55% Specificity = 99. Approximate. Renyi.27% MI localization: Accuracy = 98. PNN.16% . 52 Normal ECG polynomial fitting algorithm (PolyFit) PolyFit-based ECG parameterization algorithm (PolyECG) Akaike information criterion (AIC) MI Detection: Accuracy = 94.8% Sensitivity = 99. Fractal dimension Kolmogrov complexity Largest lyapunov exponent Normal. Wavelet entropies. 2012 Wavelet transform of multi-lead ECG Multi-scale energy Multiscale Eigen space analysis DWT on ECG beats Signal energy. Tsallis. 2015 Lead: 12 leads Beat(s): 1 beat Subjects: 148 MI.45% Specificity = 96. 52 Normal (549 ECG records) Acharya et al. CAD and MI using ECG signals Sensitivity = 91% Specificity = 85% MI Detection: Accuracy = 96% Sensitivity = 93% Specificity = 99% MI Localization: Accuracy = 99. Permutation. Shannon. Fuzzy.

Sensitivity = 99.7% Specificity = 98.5%.Lead: lead 2 Beat(s): 41545 CAD. 10546 Normal Subjects: 7 CAD. 52 Normal DCT coefficients. 148 MI. 40182 MI. DWT coefficients and IMFs of EMD AC CE PT ED M AN US In this work CR IP T ACCEPTED MANUSCRIPT DCT coefficients and KNN classifier: Accuracy = 98.5% .

Adam M. Sudarshan VK. Addison PS. 2014. Faust O. 3. ECG signal compression using discrete wavelet transform. Poo CK. . Knowledge-Based Systems. 113: 55-68. Kadri NA. Tan JH. 4. Acharya UR. Rao KR. Sudarshan VK. Wavelet transforms and the ECG: a review. Chua CK. Linear and AN US nonlinear analysis of normal and CAD-affected heart rate signals. Xiang C. 99: 146-156. 2016. Ghista DN. Computers Methods and Programs in Biomedicine. IEEE. Physiological Measurements. Nayak J. 23: 90-93. Natarajan T. 7. CE electrocardiogram signals. Application of higher- PT order spectra for the characterization of coronary artery disease using 43. Martis RJ. REFERENCES 1. Ahmed N. Suri JS. Transactions on Computers. 2008. Tamura T. Mookiah MRK. Biomedical Signal Processing and Control. 2. 178: 4571-4582. Muhammad A. Acharya UR. Chua KC. Juuso T.). Chua KP. Martis RJ. Olkkonen (Ed. 143 -169. Tan JH. Hagiwara Y. 2005. 2011. Acharya UR. Discrete wavelet transforms – applications and theory. Tan RS. Oh SL. Sankaranarayanan M. 31: 31- AC 6. 26: 155-199. Dr. Acharya UR. Information Sciences. ED 5. Vinitha S. Koh JEW. 2017. Abo-Zahhad M. Fujita H. ISBN: CR IP T 978-953-307-185-5. Koh JEW. Oh. InTech. Martis RJ. SL. 1974. Tan RS. Automated detection and localization of M myocardial infarction using electrocardiogram: a comparative study of different leads. Automatic identification of cardiac health using modeling techniques: A comparative study.ACCEPTED MANUSCRIPT 6. Discrete cosine transform. Swapna G.

Malagore IA. M 13. 5th International Conference on Electrical and Computer Engineering (ICECE). AHA. A report from the American Heart Association (AHA). 1992: 687-690. Malagore IA. 2016: e2e11. 4th International PT Conference on Bioinformatics and Biomedical Engineering (iCBBE). IEEE. 17. Findik O. Acta Paediatrica. Dohrmann M. Alam MS. 37: 3177-3183. Effects of principle component analysis on assessment of coronary artery diseases using support vector machine. ECG data compression using the discrete cosine transform (DCT). Expert Systems with Applications. Babaoglu I. A comparison of feature selection models utilizing binary particle swarm optimization and genetic algorithm in determining coronary artery disease using support vector machine. 2006. 2010. IEEE. . 12. Afsar FA. Akobeng AK. Towards real-time detection of myocardial infarction Biomedical Engineering. Babaoglu I. Proceedings of Computers in Cardiology. 1st Middle East Conference on 10. 2008: 53-58. 2005. IEEE. M. 37: 2182-2185. Detection and localization of myocardial infarction CE using k-nearest neighbor classifier. Skubic M. 15. CR IP T by digital analysis of electrocardiograms. Journal of Medical Systems. Al-Kindi SG. AC 16. 11. Arif. ED 14. Ali F. Arif M. Heart disease and stroke statistics – 2016 Update. Findik O. specificity and AN US predictive values. Farghaly A.ACCEPTED MANUSCRIPT 8. 36: 279-289. Bayrak M. IEEE. 2010. IEEE. 2010. Belina J. Ulker E. American Heart Association. Understanding diagnostic tests 1: sensitivity. Afsar FA. 2012. Arafat S. 96: 338-341. Compression of ECG signal based on its deviation from a reference signal using discrete cosine transform. 9. Classification of coronary artery disease stress ECGs using uncertainty modeling. 2011. Circulation. Automatic detection and localization of myocardial infarction using back propagation neural networks. Rahim NMS. Allen VA. Expert Systems with Applications.

In: Willerson JT. Clubb Jr FJ. International Conference on Recent Trends in Information Systems. Buja LM. AC 25. Electronic Letters. the Watanabe heritable hyperlipidaemic rabbit. London: Springer. Improved ECG compression method using discrete cosine transform. Willerson JT. International Conference on Systems in Medicine and Biology. Electronics and Computer Engineering. Hsieh JC. 26. ECG feature extraction and classification of anteroseptal myocardial infarction and normal subjects using discrete wavelet transform. 19. Cross wavelet transform based analysis of electrocardiogram signals. Liu CH. A classification approach for myocardial infarction using CR IP T voltage features extracted from four standard ECG Leads. 11: 41-52. 18: 451-461. Mitra M. Boulemden M. Cardiovascular medicine. 3rd ed. IEEE. PT 24. 2010. Human Pathology. Atherosclerosis: pathologic anatomy and pathogenesis. Coronary artery disease: pathological anatomy and pathogenesis. 1: 88-92. 3rd edition. 23. YH. AN US 21. Holmes Jr DR. Buja LM. coronary arteriographic. 22. CE Cardiovascular Medicine. 47. Buja LM. 2007: 1581-91. Wellens HJJ. 2012. 2011. and experimental M studies. Benzid R. Chang PC. Bendifallah A.ACCEPTED MANUSCRIPT 18. Willerson JT. Cohn JN. International Journal of Electrical. Cohn JN. European Heart Journal. Lin JJ. Banerjee S. Banerjee S. Mitra M. 2009. . IEEE. 1987. A hybrid system with hidden markov models and Gaussian mixture models for myocardial infarction classification with 12-lead ECGs. 1990. Buja LM. editors. Wellens HJJ. editors. 2011. Bilheimer DW. 20. 11th IEEE Conference on Hugh Performance Computing and Communications. Holmes Jr DR. McAllister Jr HA. The role of coronary artery lesions in ischemic heart disease: insights from recent clinicopathologic. Banarjee S. Pathobiology of human familial ED hypercholesterolemia and a related animal model. In: Willerson JT. London: Springer. Chou. McAllister Jr HA. Mitra M. 2007: 593-610.

Amaral LAN. Acharya UR. time-frequency localization and signal analysis. Triguero I. Knowledge-Based Systems. Martis RJ. Sankhe MS. PCA. Advances in Cardiac Signal Processing. Mechanical Systems and Signal Processing. 328: 42-59. Goldberger AL. Spaan JAE. Shetty RK. The Electrocardiogram. Benitez JM. Ivanov PCh. . Sarika K. Information Sciences. 2007. Seow SC. 32. 2016. Krishnan SM (Eds. 2013. 35. Martis RJ. Desai KD. 28. 2012. Chee J.1-53. Decision support system for arrhythmia beats using ECG signals with DCT. Sheshikala G. Peng C-K. Rodriguez Y. IEEE Transactions on Information Theory. PhysioBank. Gonzalez M. 2000. 2001. 31. 2016. 16: M 1640012 (19 pages). Mark RG. Daubechies I. Bergmeir C. PhysioToolkit. Nayak CG. 101: 215-220. Journal of Mechanics in Medicine and Biology. SpringerVerlag Berlin Heidelberg. Lim TC. Circulation. Suri JS. Cheng JS. 29. and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals. DWT and EMD methods: A comparative study. Ahamed TVI. Giri D. Hart PE. The wavelet transform. 33. In Acharya UR. Glass L. Research on the intrinsic mode function (IMF) criterion in CR IP T EMD method. Hausdorff JM. Yu DJ.). Pattern Classification 2nd Edition. 30. Stanley HE. Automated PT diagnosis of coronary artery disease affected patients using LDA. 1990: 961-1005. CE 34. 2006: 817-824. Mietus AC JE.ACCEPTED MANUSCRIPT 27. Desai U. 5th International Conference on Biomedical AN US Engineering and Informatics (BMEI). IEEE. A Wiley-Inter ED science Publication. Moody GB. A real-time fetal ECG feature extraction using multiscale discrete wavelet transform. pp. ICA and discrete wavelet transform. Suri JS. Yang Y. On the stopping criteria for k-nearest neighbor in positive unlabeled time series classification problems. Sree SV. Stork DG. Duda RO. 37: 274-282.

454: 903-995. Jin SH. 1998. Acharya UR. Long SR. IL. CR IP T 38. Locality preserving projection. Text Book of Medical Physiology. 11th Edition. New York. Guyton AC. Iyengar N. Lipsitz LA. Shen Z. 99: 219-30. NY. Chicago. Coronary artery disease (CAD). Analysis of myocardial infarction using PT discrete wavelet transform. Jayachandran ES. 2007. present and future. Kuching. 14: 3480-3483. 45. IEEE Conference on Wireless Sensors. Chung W. Godlberger AL. The empirical mode decomposition and the Hilbert spectrum for nonlinear and nom-stationary time series analysis. Wu MC. American Journal of ED Physiology. 2010. M 41. Tung CC. Sarawak. . Liu HH. Hall JE. Zheng Q. University of Chicago. Age-related alterations in the fractal scaling of cardiac interbeat interval dynamics. 2006. 37. 2013. 2006. Han J. Katz MJ. 271: 1078-1084. Paterson DJ. Choi HM. AN US 40. 34: 985-992. 2005.ACCEPTED MANUSCRIPT 36. Yen NC. Kaveh A. Journal of Medical Systems. He X. Shih HH. ECG diagnosis of acute ischemia and infarction: past. Park YK. 39. Kim WS. AC 44. Kamber M. QJM: monthly journal of the Association of Physicians. CE 2015. IFMBE Proceedings. Joseph KP. In Proceedings of the Royal Society. USA: Elsevier. Huang NE. A study on development pf multi-parameter measure of heart rate variability diagnosing cardiovascular disease. 42. 2nd Edition. Ness SM. Peng CK. Niyogi P. 2006. Automated classification of coronary atherosclerosis using single lead ECG. Wild Iris Medical Education. Herring N. Data mining: Concepts and techniques morgan kaufmann. 1996. USA. 43. Morin R.

61: 178-184. Lu HL. Acharya UR. Journal of Scientific 48. 2015. Artificial M neural network for exercise electrocardiographic detection of coronary artery February 1998. Liu B. 2001. IEEE. Lehtinen R. Wang G. Mishra H. Zheng Y. Kumar U. 2nd International Conference on Bioelectromagnetism. Lee HG. Lewenstein K. CR IP T principle to help automatic diagnosis of myocardial infarction. A novel AC electrocardiogram parameterization algorithm and its application in myocardial infarction detection. Medical and Biological Engineering &Computing. Petznick A. A data mining approach for coronary heart disease prediction using HRV features and carotid arterial wall thickness. 39: 1-6. Pachori RB. 54. 2007. Ryu KH. 51. An Efficient Automated Technique for CAD Diagnosis Using Flexible Analytic Wavelet Transform and Entropy Features Extracted from HRV signals. IEEE. Chua CK. Tong L. Noh KY. Chia P. Mining biosignal data: coronary artery disease diagnosis using linear and nonlinear features of HRV. Martis RJ. Malmivuo J. Radial basis function neural network approach for the diagnosis of PT coronary artery disease based on the standard electrocardiogram exercise test. Yanti R. Noh KY. Liu J.ACCEPTED MANUSCRIPT 46. Zhou F. Pahlm O. Kumar M. Acharya UR. 2016. 2009. Computers in Biology and Medicine. 2000. International Conference on Biomedical Engineering and Informatics. 50. Ng EYK. 53. ED disease. Tan JH. 27: 387-390. 63: 165-172. Huang K. Li F. Lee HG. Luo Y. Ryu KH. Roy AD. CE 52. 68: 866-870. Holst H. Analysis of ECG signal by chaos & Industrial Research. 218-228. Ong K. Sarkar S. Turjanmaa V. Expert Systems with Applications. AN US 49. Edenbrandt L. 47. An automated ECG classification system based on a neurofuzzy system. Melbourne Australia. Computers in Cardiology. Lahiri T. Springer-Verlag Berlin Heidelberg. 2008. pp. Application of empirical mode decomposition (EMD) for automated detection of .

Martis RJ. Lim CM. 22: 1250027. Biomedical Signal Processing and Control. 58. Suri JS. Mohamed. In Machine Learning in Healthcare Informatics. AC 62. Global Status Report on non-communicable diseases 2014. World ED 60. Expert Systems with Applications. 2015. Ray AK. Olkkonen J. LDA. Pachori RB. 2014. 2012. Tompkins WJ. Acharya UR. An approach for ECG feature extraction using daubechies 4 wavelet. 2013a. 42: 4567-4581. Mohamed A. 2014a. 8: CR IP T 437-448. Chakraboty C. Acharya UR. International Journal of Neural Systems.ACCEPTED MANUSCRIPT epilepsy using EEG signals. Current methods in electrocardiogram characterization. Computers in Biology and Medicine. 57. 2006. Martis RJ. Martis RJ. 2013b. 56. et al. Deriche. 63. . Application of empirical mode decomposition for analysis of normal and diabetic RR-interval signals. Adeli H. Martis RJ. Health Organization. Philadelphia. A real time QRS detection algorithm. Shashank K. Muhidin A. Avinash P. Lim CM. 2011. ICA and Discrete Wavelet Transform. Croatia. ECG beat classification using PCA. 2014. International Journal of Computer PT Applications. Characterization of ECG beats from cardiac arrhythmia using discrete cosine transform in PCA framework. Discrete Wavelet Transforms – Theory and applications. WB Saunders Co. Mendis S. published by CE InTech. Acharya UR. 2014b. 45: 76-82. Knowledge-Based Systems. Pan J. M 59. Acharya UR. 96: 36-41. 55. 61. Wavelet-based machine learning techniques for AN US ECG signal analysis. Sharma R. 56: 2545. 48: 133-149. 11th Edition.

Sun L. Pachori RB. 2014. Patidar S. 2016. J. 72. Zacharias D. Yager RR. 16: 1640002 (20 pages).ACCEPTED MANUSCRIPT 64. Attarodi G. Automated diagnosis of coronary artery disease using Tunable-Q wavelet transform applied on heart rate signals. 82: 1-10. 17: 132-134. Heart Disease Macmillan publishing Co. 1988. Knowledge-Based Systems. Ng L. ED 69. IEEE. Alajlan N. Sharma LN. Tripathy RK. Signal filtering using discrete wavelet transform. Annals of Emergency Medicine. 345: 340-354. Ratnakar M. 66. 2012. Lu Y. Li S. Sunil KS. International journal of recent trends in engineering. . A new pattern recognition method for AN US detection and localization of myocardial infarction using T-wave integral and total integral as extracted features from one cycle of ECG signal. Katz N. Rahhal MM. Wortzel JV. Sood S. CE 70. Bazi Y. Bosco SF. 67. Safdarian N. Acharya UR. New York (1975). Melgani F. ECG analysis using multiple instance learning for myocardial infarction detection. 2015. Dabanloo NJ. Application of empirical mode AC decomposition-based features for analysis of normal and CAD heart rate signals. Acharya UR. Allegra JR. Nitisha J. 2. 498 71. 62: 1827-1837. Kumar M. Yang K. 65. Detection of coronary artery disease from the normal resting ECG using nonlinear PT mathematical transformation. 2016. Deep learning CR IP T approach for active classification of electrocardiogram signals. Silber EN. Dandapat S. 59: 3348-3356. Biomedical Science and Engineering. Pachori RB. IEEE. Transaction on Biomedical Engineering. 2015. p. Information Sciences. Schreck BS. AlHichri H. 7: 818-824. Journals of Mechanics in Medicine and Biology. 2009. Transaction on Biomedical Engineering. Multiscale energy and eigenspace approach M to detection and localization of myocardial infarction.. 68. Schreck DM.

378 751.3632 57. Townsend N.0017 646. 1982. Townsend N.2096 172.7059 305. Chen Y. Rayner M.9676 329. British Heart Foundation: London.672 5. Beijing. 2011.7197 534. 74.386938 -171.948 695.5498 PT 3232.07488 231.4445 32. Buja LM.945 268. September 18-21.509 311. APPENDIX Normal Mean CE MI F-Value SD Mean SD Mean SD 1564.98502 -28.826 2193.0001).172 -141.7374 3717.8666 543.6713 4074. Hillis CR IP T Cardiovascular disease statistics.7363 -51.28165 Table 4: Results of LPP coefficients obtained using DWT method for the normal. Nichols M. Wickramasinghe K.049 9.956 -438.96811 196.256 -1039. British Heart Foundation: London.77 -45.1662 270. Bhatnagar P.766 803. Rayner M.3952 1627. Ji W. Williams J.7637 286. A novel method of diagnosing coronary heart disease by analyzing ECG signals combined with motion activity. Features Normal Mean SD CAD Mean MI SD Mean F-Value SD .777961 72. Willerson JT.ACCEPTED MANUSCRIPT 73.6197 743. 75.76763 31.1881 1458. Bhatnagar P. 2014.8649 834.5564 577. Coronary heart disease statistics.48351 -99.5191 3184.797 86. CAD and MI ECG signals (p<0.1463 1015. Wickramasinghe K. LD. Leal J. Luengo-Fernandez R.719 -180.0001).379 662.3527 558.677 384.5519 383.244 830. IEEE International Workshop AN US on Machine Learning for Signal Processing. a compendium of health statistics 2012 edition.926 -646.078 477.612 1216.037 -374. Smolina K. New York: Raven.06408 38.2547 972.2325 243. 12826.602 77.05236 20.994 4016.1472 268. disease clinical and 76.1001 -131.6169 AC LPP3 LPP4 LPP6 LPP2 LPP5 LPP1 LPP10 LPP9 LPP8 LPP11 LPP12 LPP7 CAD ED Features M Table 3: Results of LPP coefficients obtained using DCT method for the normal. China. Ischemic heart pathophysiological aspects.311 -436.93 4092.1795 2109.5426 671.927 -102.109 623.4614 505.9786 673.2805 545. Yin L.025 910.666 4419. CAD and MI ECG signals (p<0.9099 2439.5911 449.7575 681.9879 2758. 2012.6623 -0.74155 234.3687 1042.7941 387.255 2204.117 189.738 -424.3159 292.

303186 1.3126 40.40835 1333.293276 2.504627 0.068 900.81201 300.68047 66.11746 -104.76112 20.926139 M 2.2259 76.308159 2.084 678.555882 1. CAD and MI ECG signals (p<0.55041 9.284995 -0.708488 0.581 2.856 -77.580344 0.785 3347.662832 0.176529 -0.6026 665.673 12.1606 515.0001).555597 0.155968 0.8902 93.594684 0.499149 87.17982 CE LPP4 LPP3 LPP2 LPP7 LPP11 LPP10 LPP1 LPP9 LPP12 LPP5 LPP6 LPP8 Normal Mean SD ED Features AN US Table 5: Results of LPP coefficients obtained using EMD method for the normal.596162 0.091849 3.97 -98.004148 0.771773 1.331978 0.ACCEPTED MANUSCRIPT -74.382 -62.13768 1.08003 0.1268 18.0044 530.2858 597.1445 7.6901 13.42925 0.37565 31.7272 -78.4954 115.48845 795.374641 1.305453 0.8429 93.7729 -5.03764 -0.369248 0.991 2137.8773 -95.824 3742.7623 124.045123 2.64179 12.16914 108.735 1705.557028 1.61875 2.00854 0.2482 -86.2879 188.65347 -89.777704 0.03578 -79.40985 -1029.3519 740.23136 -0.609805 .470724 -0.846 1003.391799 0.43097 83.79 -95.4666 137.092582 -0.846943 0.37896 193. 6678.37151 104.7695 660.52394 15.26369 652.759511 -0.479644 2.9709 24.8095 463.28786 1.901259 0.0004 21.351888 83.968 25.05469 468.895 40.76502 1.015333 1.00439 35.0459 12.7201 379.1773 43.70406 20.785825 0.32691 18.135457 -0.101 3897.957 89.339736 0.52887 1.19348 MI F-Value SD Mean SD 1.16441 0.51134 16.3416 -79.146 1125.54842 11.78288 CR IP T LPP7 LPP2 LPP4 LPP8 LPP1 LPP6 LPP12 LPP9 LPP3 LPP11 LPP10 LPP5 AC CAD Mean 1.068406 PT 3.65405 0.9712 25.01065 0.87 -103.075401 -0.0473 47.684123 0.199115 0.941 6.06508 -49.2606 92.09479 -0.455717 -0.48275 89.698257 3.4176 -101.551883 1.49819 -1527.169629 84.10654 -1392.884296 16.3016 2.1063 17.29976 460.020902 -0.39211 13.19489 -0.578863 0.8626 331.664289 5.6312 38.61883 392.92928 15.901178 0.11915 -0.49384 13.342545 0.86712 2.355374 2.7017 4062.77294 176.23874 23.997758 0.767533 0.