Original Research ArticleRecognition of ECG signals using wavelet based on atomic functions
Introduction
Diverse heart disease affects the lives of millions, and this disease is the leading cause of death. As people get older, they have more probability of suffering heart disease. The United Nations calculated that the number of people aged 60 years would increase by 56% in 2030 [1]. The name of the medical test that detects heart abnormalities is electrocardiogram (ECG). The ECG measures the electrical activity of the heart through electrodes attached on the skin. The electrodes can be attached in different parts of the body; the most common configuration uses five electrodes [2]; other configuration uses ten electrodes [3]; these configurations measure the electrical signal. Different settings could be applied, so many kinds of leads could be created. The most common are; Lead I, Lead II, and Lead III.
Fig. 1 shows the ECG of Normal Rhythm; the waves on the ECG correspond electrical phenomena on the heart surface. P wave represents the Atrial depolarization. The QRS complex wave represents the Ventricular depolarization and the Atrial repolarization, while T wave represents the Ventricular polarization. When the electrical heart signals do not work correctly, the waveforms on the ECG change, then we know that an arrhythmia is present.
The term arrhythmia means any change in the electrical activity of the heart. This activity may happen fast, slowly, or erratically, thus the heart beats fast, slowly or erratically. Some people experiment in irregular heartbeats, which may feel like a racing heart or fluttering. Many arrhythmias are harmless, but if they occur in a damaged heart or they are severely abnormal, these arrhythmias could be dangerous. Arrhythmias consist of an early beat (premature contraction), slow beat (bradycardia), fast beat (tachycardia), and irregular beat (fibrillation or flutter).
The arrhythmias produce alterations on the ECG signal; in consequence, the ECG signals are the most common technique used to diagnose heart diseases. This technique could be complicated for humans; for example, in a complete ECG record, each beat has to be analyzed, sometimes the records last minutes, hours, or even days. After hours of analysis, human error may appear because of tiredness. For that reason, computational methods are an alternative to interpret the ECG signals.
Some years ago, several algorithms for ECG classification had been widely developed. The classical proposes consist of the next stages: Preprocessing, Segmentation, Feature extraction, and Classifier Algorithms. About the preprocessing stage, many techniques have been applied like FIR Digital Filter [5], Adaptive Filters [6], Multiadaptive Bionic Wavelet Transform [7], Nonlinear Bayesian Filters [8], among other things. The Segmentation stage uses techniques like Quad Level Vector [9], Neural Networks [10], Genetic Algorithms [11], Wavelet Transform [12,13], Filter Banks [14] among others. The feature extraction stage applies many algorithms, these are the most common: RR Interval [15,16], proposed Normalized RR Interval, ECG segments or ECG intervals [17], another features extraction algorithms are applied in time or frequency domain [18], Principal Component Analysis (PCA) [19,20], Kernel Principal Component Analysis (KPCA) [21], Independent Component Analysis (ICA) [22] among others techniques. The final stage uses techniques like Artificial Neural Network (ANN) [23], Support Vector Machine (SVM) [24], Linear Discriminant [25], Decision Trees [26], Nearest Neighbors [27], Clustering [28], Hidden Markov Models [29], Convolutional Neural Network (CNN) [1,[30], [31], [32], [33]], inter alia.
Advanced signal processing and learning methods have been applied to develop computational methods to recognize ECG signals. Researchers in the literature have applied one or more of the techniques previously mentioned. In Ref. [34] used a modified U-net model to perform analysis on ECG segments. The authors in Ref. [1] applied CNN to detect the ECG signals. In Ref. [35] preprocessed the ECG signal using a low-complex digital hardware implementation for arrhythmia detection. In Ref. [36] the features of the ECG signal are calculated using Welch's method and a discrete Fourier transform, in the Classifier stage, they use two genetic ensembles based on SVM. An automated system using a combination of CNN and long short-term memory (LSTM) for detection ECG signals is proposed by Ref. [33]. In Ref. [37] extracted the features of the ECG signal using the spectral power density then a machine learning algorithms were used. In Ref. [38] the apply to deep learning(1D-CNN) to detect 17 classes of cardiac arrhythmia. A 9-layer deep CNN to identify 5 different heartbeats in ECG signals is developed by Ref. [30]. In Ref. [31] a CNN structure comprises of four convolutional layers, four max pooling layers and three fully connected layers for the diagnosis of Coronary artery disease was proposed. In Ref. [32] presented a CNN to automatically detect the different ECG segments. The main features of the ECG signals are obtained through discrete wavelet transform, followed by principal component analysis on each decomposed level, the features were reduced through statistical analysis as an input to SVM in Ref. [39]. Arrhythmias detection algorithm that combines a number of ECG parameters using SVM is investigated in Ref. [40]. In Ref. [41] presented an automated method for using a single lead of ECG sensors using PCA, DWT, and SVM. In Ref. [42] the proposal applied PCA of segmented ECG beat, then these features were independently classified using neural network and LS-SVM. In Ref. [43] the authors applied an optimum support vector machine in which the dataset is reduced to 18 features using PCA. A summary of the approaches mentioned is presented in Table 1.
The approaches in the literature present the problem that the ECG signals have noise, the authors face the problem applying DWT or other filtering techniques. In this proposal, the signals were preprocessed to eliminate the noise and baseline wander using Wavelet based on Atomic Functions (WAF) due to it eliminates the noise better than the classic wavelets [44], then Z-score was applied to normalize the signals. After these signals were divided into segments, these segments long one second, two seconds, five seconds, and ten seconds. These segments were fed into the classifier stage to make the final decision. The rest of the paper is organized as follows: Section 2 is a summary of the databases, Section 3 is a description of the proposal, the experimental results are shown in Section 4, Section 5 provides the discussions, and finally, the conclusions.
Section snippets
Databases
The Databases used in this paper are obtained from 4 publicly available databases MIT-BIH [45,46], MIT-VFDB [45,47], Fantasia [45, 59] and St.-Petersburg INCART [3,45]. The summary of the four databases is shown in Table 2.
Proposed approach
This paper proposes to classify different ECG signals. To achieve this goal, we use different databases. First, the performance of each database is calculated according to the block diagram in Fig. 2. Next, the databases are gathered creating a new database. Then, the performance of the new database is calculated using the block diagram in Fig. 2 too. The block diagram of the proposed approach is shown in the next Figure. First, an ECG signal is fed into the preprocessing stage, which consists
Training & testing
In this paper, 10-fold cross-validation is performed to evaluate the proposed approach. We use 90% of the signals to training the approach; the other 10% is used to test the method. The ECG signal samples, after the preprocessing step, were directly fed to the classifier stage. The results are shown in Table 8, Table 9, Table 10, Table 11 use as classifier Fine Gaussian SVM [53,55,56], and the results shown in Table 12 use Fine Gaussian SVM and Fine tree [54,57].
Results
In this section, the results for the databases are shown in Table 8, Table 9, Table 10, Table 11, Table 12 shows the results when the previous databases were combined. Table 8, Table 9, Table 10, Table 11, Table 12 show the results when the ECG signals were segmented (1 s, 2 s, 5 s, and 10 s), and these signals were resampled, starting with the original sample frequency and gradually decreasing to 250 Hz. The frequency 250 Hz ensures standardization between databases. Table 3, Table 4, Table 5,
Discussion
This paper faces the problem of recognizing different ECG signals. The proposed approach uses ECG signals (1 s, 2 s, 5 s, and 10 s). We used the databases MIT-BIH, MIT-VFDB, Fantasia, and St Peter. These databases are prevalent in the literature. The results in Table 8, Table 9 show that if the ECG signals are resampled, the accuracies oscillate in a maximum of 1.3%, as we can see in Table 8 when the 10 s signals were analyzed. If the sampling frequency is reduced, the number of samples per
Conclusion
This paper introduces the Wavelet based on Atomic Functions in the preprocessing stage to classify different ECG signals. The results are compared to other reported works; which used classic Wavelet in the preprocessing scene. The proposal recognizes up to 11 kinds of ECG signals with an accuracy of 98.9%.
The results show that it is possible to reduce the sample frequency without considerable losses in accuracy, which means lower computational cost. Then, classification techniques as CNN could
Disclosure of potential conflicts of interest
We declare that we do have no commercial or associative interests that represent a conflict of interests in connection with this manuscript. There are no professional or other personal interests that can inappropriately influence our submitted work.
Research involving human participants and/or animals
This article does not contain any studies with human participants or animals performed by any of the authors.
CRediT authorship contribution statement
Andres Hernandez-Matamoros: Conceptualization, Data curation, Visualization, Software, Writing - original draft. Hamido Fujita: Acquisition, Supervision, Visualization, Formal analysis, Methodology, Writing - review & editing. Enrique Escamilla-Hernandez: Software, Validation. Hector Perez-Meana: Project administration, Funding, Writing - review & editing. Mariko Nakano-Miyatake: Investigation, Writing - review & editing.
Acknowledgements
The authors would like to thank to the National Science and Technology Council of Mexico for the financial support during the realization of this research. This study is supported by JSPS KAKENHI (Grants-in-Aid for Scientific Research) #JP20K11955.
References (60)
- et al.
Automated identification of shockable and non-shockable life-threatening ventricular arrhythmias using convolutional neural network
Future Gener Comput Syst
(2018) - et al.
ECG beat classification using particle swarm optimization and radial basis function neural network
Expert Syst Appl
(2010) - et al.
A new method for classification of ECG arrhythmias using neural network with adaptive activation function
Digit Signal Process
(2010) Combining recurrent neural networks with eigenvector methods for classification of ECG beats
Digit Signal Process
(2009)- et al.
Robust multiple cardiac arrhythmia detection through bispectrum analysis
Expert Syst Appl
(2011) - et al.
Analyzing ECG for cardiac arrhythmia using cluster analysis
Expert Syst Appl
(2012) - et al.
A deep convolutional neural network model to classify heartbeats
Comput Biol Med
(2017) - et al.
Automated detection of coronary artery disease using different durations of ECG segments with convolutional neural network
Knowledge Based Syst
(2017) - et al.
Automated detection of arrhythmias using different intervals of tachycardia ECG segments with convolutional neural network
Inf Sci
(2017) - et al.
Automated diagnosis of arrhythmia using combination of CNN and LSTM techniques with variable length heart beats
Comput Biol Med
(2018)
Automated beat-wise arrhythmia diagnosis using modified U-net on extended electrocardiographic recordings with heterogeneous arrhythmia types
Comput Biol Med
Novel genetic ensembles of classifiers applied to myocardium dysfunction recognition based on ECG signals
Swarm Evol Comput
Novel methodology of cardiac health recognition based on ECG signals and evolutionary-neural system
Expert Syst Appl
Arrhythmia detection using deep convolutional neural network with long duration ECG signals
Comput Biol Med
Application of principal component analysis to ECG signals for automated diagnosis of cardiac health
Expert Syst Appl
Effects of principle component analysis on assessment of coronary artery diseases using support vector machine
Expert Syst Appl
The six second ECG: a practical guidebook to basic ECG interpretation
Methodological considerations in the design of trials for safety assessment of new drugs and chemical entities
Curr Control Trials Cardiovasc Med
Recursive digital filters for biological signals
Med Biol Eng Comput
ECG enhancement by adaptive cancellation of electrosurgical interference
IEEE Trans Biomed Eng
Multiadaptive bionic wavelet transform: application to ECG denoising and baseline wandering reduction
EURASIP J Adv Signal Process
A nonlinear Bayesian filtering framework for ECG denoising
IEEE Trans Biomed Eng
ECG signal compression and classification algorithm with quad level vector for ECG holter system
IEEE Trans Inf Technol Biomed
Application of artificial neural networks for ECG signal detection and classification
J Eletrocardiol
Genetic design of optimum linear and nonlinear QRS detectors
IEEE Trans Biomed Eng
The wavelet transform, time/frequency location and signal analysis
IEEE Trans Inform Theory
A wavelet-based ECG delineator: evaluation on standard databases
IEEE Trans Biomed Eng
ECG beat detection using filter banks
IEEE Trans Biomed Eng
Advanced methods and tools for ECG data analysis
Cited by (38)
Design of subject independent 3D VAD emotion detection system using EEG signals and machine learning algorithms
2023, Biomedical Signal Processing and ControlThe prediction of cardiac abnormality and enhancement in minority class accuracy from imbalanced ECG signals using modified deep neural network models
2022, Computers in Biology and MedicineClassification of cardiac diseases from ECG signals through bio inspired classifiers with Adam and R-Adam approaches for hyperparameters updation
2022, Measurement: Journal of the International Measurement ConfederationA new approach to adaptive threshold based method for QRS detection with fuzzy clustering
2022, Biocybernetics and Biomedical EngineeringECG signals-based automated diagnosis of congestive heart failure using Deep CNN and LSTM architecture
2022, Biocybernetics and Biomedical EngineeringCitation Excerpt :The constructed synthetic data was utilized by this BiRNN-based CHF detection approach for building models that plays an anchor role in differentiating between the considered signals with maximized accuracies. An Atomic functions-based ECG recognition scheme was proposed for detecting CHF disease with the elimination of baseline wander and noise [27-30]. It adopted a new processing stage that aided in sampling and resampling the frequency of ECG signals.