Empirical mode decomposition based ECG enhancement and QRS detection
Introduction
Automatic computerized analysis of physiological signals is a major field of interest since last few decades. The purpose of automation is to reduce the human effort and time required for analysis and interpretation. This helps in handling a large number of data, for fast processing and decision making, specially in intensive care services. In some cases the long duration biomedical data are compressed and stored in storage elements and processed and analyzed later by experts to detect abnormalities if any. Persons having critical and multiple abnormalities require simultaneous analysis of different physiological signals. Computer based automatic procedure helps a lot in this application as an assistive diagnosis and monitoring tool.
Electrocardiogram (ECG) is one of the major physiological signals generated from heart's rhythmic polarization and depolarization. ECG is characterized by a number of waves as P, QRS, T related to the heart activity. It is recorded, studied, analyzed and interpreted for diagnosis of cardiac abnormalities. Automation in the entire process may lead to better clinical evaluation and resulting medication. Automatic analysis consists of the steps shown in Fig. 1. In this paper an adaptive QRS detection method in presence of high and low frequency noise is discussed.
The first problem faced by an automatic ECG signal processing technique is the unwanted frequency components present in it as noise generated due to non-cardiac reasons. The low frequency noise is originated from respiration or body movement of the subject during recording. It shifts the ECG signal from zero potential baseline in a nonconventional way and thus makes it very difficult to estimate certain low frequency phenomena like ST segment morphology. Moreover, larger baseline drift may cause clipping of the peaks of ECG by saturation of ECG amplifier. A high frequency noise called power line interference appears due to interference of supply frequency with the recorded ECG signal. Cables carrying the ECG signal from patient to monitoring system are susceptible to electromagnetic interference of power frequency of 50 Hz (or 60 Hz) and thus act as the source of the noise. It also appears due to poor transmission channel conditions. Some other sources of high frequency noise in ECG are electromyographic noise during recording, noises from the signal conditioning circuit etc. So it is very difficult to eliminate it during recording or transmission.
The second step of ECG signal processing is QRS detection. The QRS complex is the most prominent waveform of the signal and hence detection of QRS complex becomes the entry point of almost all ECG analysis algorithms. The temporal and spatial information of QRS complex and its texture are used for heart rate measurement and diagnosis of several abnormalities like myocardial infarction, cardiac arrhythmias, conduction abnormalities, ventricular hypertrophy, etc.
Several methods are proposed for ECG enhancement by eliminating noise. Most of the methods are based on designing a digital filter [1], [2], which passes the required high frequency parts of ECG suppressing the baseline wander and a low pass filter [2], [3], [4] for removing power frequency interference. The main problem of low pass or band pass filtering is that ECG frequency spectra is quite overlapping with that of the noise spectra specially due to the presence of QRS complex as high frequency component in ECG. An adaptive filtering method [5] is also proposed for baseline correction. Some modern techniques like Principle Component Analysis [6], Neural Network [7], Wavelet Transform [8] etc are proposed for high frequency noise elimination from ECG wave but all have their inherent shortcoming mainly due to the dependence on ECG frequency whose bandwidth is not constant.
Automatic detection of QRS complex is a well visited problem in biomedical signal processing. As a result a number of detection techniques are reported. In many nonsyntactic methods like [9], [10]; for QRS detection, P and T waves and noises are suppressed by bandpass filtering and some nonlinear transformation is performed for QRS complex enhancement. Then some rule based technique is used to identify QRS region. All the filter based approaches suffer from a genuine problem of selecting the signal pass band. It is seen that signal pass band of QRS region may overlap with noise frequency. Moreover, it is different for different persons or even in same person at stressed condition or due to a long interval. Another method [11] uses adaptive matched filtering technique based on artificial neural network (ANN). The low frequencies are modeled by an ANN based adaptive filter and the residual signal is passed through a matched linear filter for the detection of QRS location. A fuzzy hybrid neural network based approach is also proposed to recognize different type of beats resulting from same or different source [12]. However, in most of the cases the efficiency of the algorithms is accompanied by higher computational time and cost. Hidden Markov models [13] and pattern recognition techniques are also used for the detection of QRS complex [14]. Wavelet transform based multiresolution analysis for signal decomposition technique [15], [16] is also used for QRS detection. Wavelet has the advantage that it does not require any predefined cutoff frequency for detection. But it is seen that wavelet functions that support compactness and symmetry with the test signal provide better result.
All the methods including wavelet based approach are non adaptive and hence not globally applicable. Basically due to dynamic changes in the behavior of heart and related organs, the ECG signals may exhibit time-varying as well as non-stationary behavior. Moreover, the unpredictable nature of high and low frequency noises makes the task of noise elimination and QRS detection a difficult one for conventional filtering technique or other non adaptive approaches. Hence a fully adaptive approach can perform better in almost all cardiological conditions. Recently Huang et al. have proposed the Empirical Mode Decomposition method (EMD) [17] as a new tool for the analysis of nonlinear and non-stationary time domain data. Being a completely data driven approach, it extracts the basis function from the signal itself. Thus it can be used for any kind of ECG signals. Recently EMD is being used for signal decomposition and analysis in different fields of biomedical domain [18], [19]. It is also used for QRS detection [20], [21] of ECG signal. But these methods require pre-filtering of ECG using standard band pass filters prior to EMD based QRS extraction. Thus it requires two fold operation of each ECG signal—(1) for noise elimination and (2) for QRS detection.
In this paper an EMD based single run approach for noise elimination and QRS detection is proposed. Here noises are estimated by statistical technique from the set of decomposed signals and then the QRS region is reconstructed from the relevant components of decomposed signals.
Section snippets
Empirical mode decomposition (EMD)
Empirical Mode Decomposition is relatively new signal processing technique used for nonlinear, nonstationary time series decomposition. It is different from Fourier Transform (FT) or Wavelet Transform (WT) because of the fact that the basis functions are directly derived from the signal under test. In a priori basis analysis like FT or WT, the harmonics are definitely like the basis function in one form or other. According to the principle of EMD, it decomposes a signal into a sum of
Baseline wander correction
It is known from the previous section that Empirical Mode Decomposition decomposes a signal into IMFs of gradually decreasing frequency and baseline wander is expected to present in some higher order IMFs. The residue of EMD operation may contain some parts of total baseline drift but it is not possible to have the entire baseline problem contained in the residue. This is because baseline wander may contain multiple extrema and zero crossings, which the residue cannot have as per its property.
Result and analysis
The proposed method is tested against a set of arbitrarily chosen databases taken from Physionet PTB diagnostic database and MIT-BIH arrhythmia database having sampling frequency 1 kHz and 360 Hz, respectively. The algorithm is tested with beats of different cardiac conditions as a cardiac disorder results in a change in beat morphology. Normal cardiac rhythm, Myocardial Infarction (MI), Bundle Branch Block (BBB), Hypertrophy, Dysrhythmia beats are taken from PTB database and Premature
Conclusion
In this paper, an Empirical Mode Decomposition based ECG enhancement and QRS detection technique is proposed. In almost all methods of QRS detection requires first pre-filtering and then some other signal processing algorithm is used for QRS detection. Thus it involves two fold processing of each signal. In this method a single decomposition operation is required followed by a statistical approach for noise elimination and a QRS enhancement operation for QRS detection. In most of the cases
Conflict of interest
We hereby disclose that there is no financial and personal relationships with any other people or organizations/institutes that could inappropriately influence (bias) our work.
References (27)
- et al.
Characterization and elimination of AC noise in the electrocardiogram: a comparison of digital filtering methods
Comput. Biomed. Res.
(1993) - et al.
A new mathematical based QRS—detector using continuous wavelet transform
Comput. Electr. Eng.
(2008) - L. Sornmo, Time-varying filtering for removal of baseline wander in exercise ECGs, in: Proceedings of Computers in...
- et al.
Removal of base-line wander and power-line interference from the ecg by an efficient fir filter with a reduced number of taps
IEEE Trans. Biomed. Eng.
(December 1985) - et al.
Comparing stress ECG enhancement algorithms
IEEE Eng. Med. Biol. Mag.
(1996) - R. Jane, P. Laguna, et al., Adaptive baseline wander removal in the ECG: Comparative analysis with cubic spline...
- et al.
QRS morphology representation and noise estimation using the Karhunen–Loève transform
Comput. Cardiol.
(1989) - et al.
One-pass training of optimal architecture auto-associative neural network for detecting ectopicbeats
Electron. Lett.
(Aug. 2001) Nonlinear wavelet and wavelet packet denoising of electrocardiogram signal
Biol. Cybern.
(April 1999)- et al.
A real time QRS detection algorithm
IEEE Trans. Biomed. Eng.
(1985)