1

ECG Signal Feature Extraction Using Wavelet Transforms

Ruqaiya Khanam*, Ranjana Prasad**, Vinay Arora***

AbstractThis paper deals with the study of ECG signals using wavelet transform analysis. Biomedical signals like heartbeat ECG signals tend to be nonstationary. Wavelet transforms are a powerful tool for this kind of signals. Using wavelet transform, the ECG signal is denoised by removing the corresponding wavelet coefficients at higher scales. We make use of wavelets to detect the positions of the occurrence of the QRS complex during the period of analysis. Index TermsECG, transform QRS, Signal de-noising, Wavelet

I. INTRODUCTION

An Electrocardiogram (ECG) is one of the most important

physiological signals with essential information for heart disease diagnosis. It is the electrical manifestation of the contractile activity of the heart. ECG is a quasi-periodical, rhythmically repeating signal, synchronized by the function of the heart, which acts as the generator of bioelectrical events. A typical ECG cycle is defined by the various waves (P, Q, R, S and T) of the electrical wave. The P wave marks the activation of the atria, which are the chambers of the heart that receive blood from the body. The activation of the left atrium which collects oxygen-rich blood from the lungs, and the right atrium, which gathers oxygen-deficient from the body, takes about 90msec. The QRS complex corresponds to the current that causes contraction of the left and right ventricles; the T wave represents the repolarization of the ventricles. The presence or absence of these waves as well as the QT interval and PR interval are important parameters for screening and diagnosis of cardiovascular diseases. Figure 1. A typical single cycle ECG tracing New ECG monitors having multiple filters for signal processing. The common settings are monitor mode and diagnostic mode. In monitor mode, the low frequency filter is se at either 0.5 Hz or 1 Hz and the high frequency filter is set at 40 Hz. The high pass filter reduce wandering baseline and the low pass filter reduces 60 Hz power line noise. In diagnostic mode, the high pass filter is set at 0.05 Hz, which allows accurate ST segments to be recorded. The low pass filter is set to 40, 100 or 150 Hz. Consequently, the monitor mode ECG display is more filtered than diagnostic mode, because its bandpass is narrower. To detect the possible cardiovascular abnormalities, the extraction of ECG signal by locating the interested characteristic points, analyzing ECG accurately become especially important in these application. The extraction become more complicated because of either corrupted ECG signal or embedded noises in it. The wavelet analysis is the solution to all these problems. Wavelet transform: A wavelet is simply a small wave which has energy concentrated in time to give a tool for the analysis of transient, nonstationary or time-varying phenomena such as a wave shown in fig 2.

Electrical and Electronics Engineering , Lingayas University, Faridabad* Electronics and Communication Engineering , IEC College of Engineering & Technology, Gr. Noida ** Electronics and Communication Engineering, Brown Hills College of Engineering & Technology, Faridabad*** E-mail: [email protected]

(5)

Where (t) is called the scaling function associated with the wavelet function (t) and is governed by the following condition: Figure 2. Wavelet function Continuous wavelet transform CWT: The CWT-Wf(s,)- is the inner product of a time-varying signal f(t) and the set of wavelets s,(t), it is given by (t) . dt = 1 (6)

(1) The scaling and shifting the mother wavelet () with a factor of s and , respectively, generate a family of functions called wavelets given by:

DWT uses two filters, a low pass filter (LPF) and a high pass filter (HPF) to decompose the signal into different scales. The output coefficients of the LPF are called approximations while the output coefficients of the HPF are called details. The approximation signal may be passed down to be decomposed again by breaking the signal into many levels of lower resolution components. This is called multilevel decomposition and may be represented in a wavelet decomposition tree. Only the last level of approximation is save among all levels of details, which provides the sufficient data to fully reconstruct the original signal using complementary filters. The automatic detection of ECG wave is an important topic, especially for extended recordings; because it gives various aspects can be derived from the information found in the intervals and amplitudes defined by the significant points[4]. The performance of such automatic systems depends on the accuracy and reliability in the detection of the QRS complex, which necessary to determine the heart rate. The QRS complex is the most prominent waveform of the signal with higher amplitudes. The extraction method of significant points of ECG signal in this paper concentrates on (1) Detection of QRS complexes and characteristic points. (2) Attempting to locate the associated P and T waves if any. Because of the fact that it can be used both in frequency and time domain, wavelet transform is a very promising technique. It may be used to distinguish ECG waves from serious noise, artifacts, and baseline drift. It is also used the temporal features of a signal at different resolution using wavelet transform provide better analysis of ECG signal, which is characterized by cyclic occurring pattern at different frequencies. Wavelet transform is easy to apply as a mathematical tool in decomposition of signal. The difficult part is to choose a mother wavelet that optimally fixed the signal depending the application and the signal itself. In ECG analysis discrete wavelet transform have its advantages. Normally, before ECG feature extraction the signal is

(2) With s>0 Discrete wavelet transform DWT: The DWT consist of applying the discrete signal to a bank of octave band filters based on low and high pass filters h(n) g(n) respectively. The original time-varying signal f(t) would be expanded using the following formula :

(3) Where

(4)

3 processed through a bandpass or matched filter to suppress the P and T waves and noises before sending it to characteristics detection. II.
MATERIALS AND METHODES

We have opted, in our study, for the MIT-BIH Arrhythmia database (100.dat) [5] a commonly so-used free down able annotated database. Each record consist of dual channel ECG data and is around 30 min long.. The ECG signal is sampled at rate of 360 Hz with 11 bit resolution over a 10mV range [5]. The obtained results showed on Table 1, for wavelet analysis, we used the Matlab program, which contains a Wavelet toolbox. First step: Considered signal was decomposed using three level wavelet decomposition. We used in our analysis the Daubechies db2 and db4 wavelets, the symlet sym4 and coiflet 5 wavelets. Fig.3 shows the Daubechies db4 wavelets waveform.

The high frequency components of the ECG signal decreases as lower details are removed from the original signal. After removing lower details, the signal becomes smoother and the noises are marked by high frequency components picked up during the transmission. Noise filtration using by discrete wavelet transform implicitly.

Original signal

Denoised signal Figure 5. Removing background noise from the ECG signal Second step: The obtained data is then squared to stress the signal. To set a practical lower limit to help remove the unrelated noisy peaks. A threshold of 35% of the maximum value is occurred. Now the data set is ready to peak detection using a simple search algorithm and produces a very accurate result.

Figure 3. Daubechies db4 wavelet In Fig .4 is shown here a 3-level signal decomposition of a sample ECG waveform using the db4 wavelet.

Denoised ECG signal

Extracted QRS signal Figure 4. ECG signal decomposition using db4 wavelet Figure 6. QRS detection from ECG signal

The results obtained with the proposed wavelets using several different noise levels which are in Table 1. The excellent resemblance between db4 wavelet and the actual ECG signal keeps this wavelet ahead of other wavelets. TABLE 1 SNR[dB ] 10 7 3 10 7 3 10 7 3 10 7 3

Wavelet Db2

Db4

Sym4

Coif5

QRS Detected 0.975 0.943 0.878 0.975 0.975 0.832 0.945 0.849 0.845 0.956 0.942 0.856

Errors[%] 0 3.28 9.94 0 0 14.6 3.07 12.9 13.3 1.94 3.38 12.2

III. CONCLUSION In this paper we explore different advantages using wavelet transform associated with noise strategy .We also put across a simple detection algorithm and detecting the position of QRS complexes in ECG signals. All relevant noises are removed from the signal allowing the utilization simple detection logic for the QRS detection through wavelet thresholding. The result is less time consuming analysis for a long time ECG signal. REFERENCES
[1] Carmona R.A., Torresani B., Characterization of Signals by the Ridges of their wavelet transforms IEEE Trans. Signals Processing, vol.45,no.10, oct.1997. [2] Mahmoodabadi, S.Z, Ahmadian, A. Abolhasani, M.D, ECG Feature Extraction Using Daubechies Wavelets, Proceedings of the Fifth IASTED International Conference, VISUALIZATION, IMAGING, AND IMAGE PROCESSING, Benidorm, Spain. September 7-9, 2005 [3] Martinez, J.P.,Almeida, R., Olmos, S., Rocha, A.P., Laguna, P., A Wavelet based ECG Delineator: Evaluation on Standard databases, IEEE Transactions on Biomedical Engineering, Vol, 51, no.4, April 2004 [4] Cuiwei Li, Chongx Zheng, and Chengfeng Tai. Detection of ECG Characteristic points using wavelet transforms. IEEE Trans. Biomed . Eng., 42(1) : 21-28, Jan.1995 [5] MIT-BIH Arrhythmia Database , www.physionet.org [6] Wikipedia the free encyclopedia http:// en.wikipedia.org/wiki/Electrocardiogram