Automated detection of congestive heart failure from electrocardiogram signal using Stockwell transform and hybrid classification scheme
Introduction
Congestive heart failure (CHF) is a cardiac ailment, resulting when the heart is unable to deliver a sufficient amount of oxygen and other nutrients to the different parts of the body [1], [2]. Its progression causes the heart muscle walls to become weaker so that the lower chambers of the heart are ineffective to pump the blood. The major causes of CHF are the valvular disease, alcoholism, hypertension, myocardial infarction, and diabetes [3]. The cardiac output is reduced due to the cardiac ailments such as left ventricular hypertrophy and left ventricular systolic dysfunction [4], [5], [6]. Therefore, the heart fails to deliver the required amount of blood to the whole body conveying to CHF. In developing countries, more than five million elderly people are suffering from CHF pathology, and this number may increase due to the consequence of the diseases like diabetes, myocardial ischemia, and hypertrophic cardiomyopathy [7], [8]. The electrocardiogram (ECG) and cardiac echocardiography are normally used for the diagnosis of CHF. The ECG is a simple and low-cost non-invasive diagnostic modality for measuring the electrical activity of the cardiac chambers [9], [10]. The CHF is diagnosed through the pathological symptoms such as the changes in the RR-interval and other morphological features in ECG [9], [11]. The medical practitioners visually assess the pathological changes in the ECG data during arrhythmia by monitoring inside the coronary care unit. This process is cumbersome due to the low sensitivity in detecting the heart ailments [10]. The early detection of CHF is an important and challenging issue in clinical practice to minimize its progression. Thus, an automated system based on the analysis of the ECG signal is required for the detection of CHF.
In recent years, different approaches have been proposed for the automated detection of CHF using the ECG signal [12]. These approaches are focused on the extraction of various diagnostic features from ECG and RR-time series, and the classification of CHF and normal sinus rhythm (NSR) [13], [14], [15]. Thuraisingham [16] has used features from the second order difference plot of RR-time series and K-nearest neighbor (K-NN) classifier to discriminate between NSR and CHF. Kuntamalla et al. have calculated sequential trend analysis and multiscale entropy features from the RR-time series [17], [18] and have reported a higher classification result for NSR and CHF classes. Hossel et al. [19] have used power spectral densities from the sub-band signals of the RR-time series as features for the detection of CHF. In [20], the same authors have proposed wavelet filters and soft decision algorithm for the detection of CHF from the RR-time series. Yu et al. [21] have used heart rate variability (HRV) features with mutual information based on feature ranking and support vector machine (SVM) for the detection of CHF. Isler and Kuntalap [22] have considered HRV features, wavelet entropy features, and K-nearest neighbor (K-NN) to classify CHF and NSR. Pechia et al. [23] have introduced the time-domain and frequency-domain HRV features, and a simple threshold-based classifier to detect CHF. Despite the aforementioned approaches have demonstrated a suitable performance for the detection of CHF, they have only used the features from the RR-time series. In addition to the HRV features, other morphological features of ECG have presented different values in CHF pathology [9]. Thus, a robust approach founded on the analysis of the ECG signal can be useful for the automated detection of CHF. Note that, a few methods have been reported for the detection of CHF using the features from ECG signal. Among them, Masetic and Subasi have introduced the auto-regressive (AR) model based on features from the ECG signal and the random forest classifier for the detection of CHF [24]. Sudarshan et al. [5] have used the dual-tree complex wavelet transform based on features from ECG and various classification techniques such as K-NN and decision tree for the classification of CHF and NSR.
The abovementioned corroborates that the search of novel feature extraction methods and classification techniques is an important step for the automated detection of heart ailments from the ECG signal [25]. Recently, different time-frequency analysis methods have been widely used for the analysis of RR-time series and ECG signals [26], [27], enabling to quantify the diagnostic features from physiological signals in the time-frequency domain; such methods are based on the short-time Fourier transform (STFT) and wavelet transform (WT). The demerits of the STFT are the poor temporal resolution for the high-frequency components and improper frequency resolution for the low-frequency events of a multicomponent signal [28]. Whereas the wavelet transform does not provide phase information since it just produces time-scale plots which require the postprocessing such as synchrosqueezing for obtaining the time-frequency representation of a signal [29]. The Stockwell (S)-transform is the generalization of STFT for a Gaussian window, which has been used to denoise the ECG signal and to detect the QRS-complex [30], [31]. This time-frequency analysis method uses variable window length to control the frequency parameter, and the phase information can be obtained with the help of the Fourier kernel. When a CHF occurs, there are variations arising in amplitudes and the durations of the clinical patterns such as QRS-complex, P-wave and T-wave of the ECG signal [32], [33]. In fact, it can be expected that the extraction of features using the S-Transform based time-frequency decomposition will be helpful to efficiently assess various pathological changes in ECG signal during CHF. This paper advocates a novel approach to develop an automated system for the detection of CHF. The time-frequency entropy features are evaluated using the Stockwell (S)-transform of the ECG signal and the Shannon Entropy. A hybrid classification scheme-based system that is composed of the class-specific sparse representation based classifier (SRC) and the average distances has been developed for the pattern classification problem in [34]. The kernel SRC has been used for the detection of diabetes using the HRV features in [35]. Unlike the classifiers such as SVM, neural network and deep neural networks, the SRC has fewer parameters (sparsity and iterations), requiring less training instances for the evaluation of the optimal training parameters [34]. It can be expected that the combination between the time-frequency entropy features from ECG and the hybrid classification scheme will be effective to classify CHF and NSR episodes. The rest of this paper is organized as follows. The proposed system is described in Section 2. The results and the discussion of the results are shown in Section 3, and the conclusions of this paper are written in Section 4.
Section snippets
Method
The flow-chart of the proposed approach for the detection of CHF is shown in Fig. 1. The approach consists of four major stages such as the preprocessing of ECG signals, the decomposition of ECG signal into time-frequency sub-band matrices based on the S-transform and frequency division, the evaluation of the time-frequency entropy features, and the hybrid classification scheme. The following sections briefly describe each stage of the proposed approach.
Results
The proposed method (as outlined in Fig. 1) is evaluated using both NSR and CHF ECG signals. The time-frequency entropy features are calculated from each ECG instance for both CHF and NSR classes. The statistical analysis of the time-frequency entropy features for NSR and CHF classes and the performance of the hybrid classifier are described in this section. The plots of the probability density function (PDF) associated with the selected time-frequency entropy features for NSR and CHF classes
Discussion
The objective of this study is the detection of CHF based on the time-frequency analysis of the ECG signal. The time-frequency features from LF, MBF and HF components of ECG signal are extracted using S-transform and entropy measure. The hybrid classification technique is used for assessing the performance of the proposed time-frequency entropy features of the ECG signal. To verify the effectiveness of the proposed approach, we have compared our results with existing CHF (type III -IV)
Conclusions
In this paper, a novel approach for the automated detection of CHF based on the time-frequency analysis of the ECG signal has been demonstrated. The time-frequency analysis of ECG was performed using S-transform. From the S-transform coefficients of the ECG signal at different frequency scales, the time-frequency entropy features have been computed. A hybrid classifier based on the combination of the residual of SRC and nearest distance for individual classes was used. The important observation
Conflict of interest
I declare that there is no conflict of interest for this paper.
Acknowledgments including declarations
We would like to thank Editor-in-Chief, Associate editor and anonymous reviewers of this journal for considering our paper for publication. This is purely an academic research work, and the first author (RK Tripathy) has received OPERA award with the grant number as FR/SCM/150618/EEE from BITS Pilani, Hyderabad Campus.
References (70)
- et al.
2013 Accf/aha guideline for the management of heart failure: executive summary: a report of the american college of cardiology foundation/american heart association task force on practice guidelines
J. Am. Coll. Cardiol.
(2013) - et al.
Heart failure: diagnosis, severity estimation and prediction of adverse events through machine learning techniques
Comput. Struct. Biotechnol. J.
(2017) Acc/aha 2005 guideline update for the diagnosis and management of chronic heart failure in the adult: a report of the american college of cardiology/american heart association task force on practice guidelines (writing committee to update the 2001 guidelines for the evaluation and management of heart failure)
J. Am. Coll. Cardiol.
(2005)- et al.
Automated diagnosis of congestive heart failure using dual tree complex wavelet transform and statistical features extracted from 2s of ecg signals
Comput. Biol. Med.
(2017) - et al.
Association of health aging and body composition (abc) heart failure score with cardiac structural and functional abnormalities in young individuals
Am. Heart J.
(2010) - et al.
Positive pressure ventilation in the management of acute and chronic cardiac failure: a systematic review and meta-analysis
Int. J. Cardiol.
(2005) - et al.
A new approach to early diagnosis of congestive heart failure disease by using hilbert–huang transform
Comput. Method. Progr. Biomed.
(2016) Real-time chf detection from ecg signals using a novel discretization method
Comput. Biol. Med.
(2013)- et al.
A wavelet-based soft decision technique for screening of patients with congestive heart failure
Biomed. Signal Process. Control
(2007) - et al.
Conditional mutual information-based feature selection for congestive heart failure recognition using heart rate variability
Comput. Method. Progr. Biomed.
(2012)