On an Algorithm for Detection of QRS Complexes in

Noisy Electrocardiogram Signal

Manab Kr. Das1, Samit Ari2, Swagatika Priyadharsini3
1, 2, 3
Department of Electronics and Communication Engineering
National Institute of Technology, Rourkela, Odisha-769008, India.
Email: 1manabster@gmail.com, 2samit.ari@gmail.com
Telephone: +91-661-2464464/2464

Abstract─Electrocardiogram (ECG) signal provides the valuable implemented for R-peak detection. They are of finite duration
information for detection of abnormal heart disease. Detection of and non stationary in nature. Hence, a technique likes Fourier
QRS complexes is the first step towards recognition of heart series (based on sinusoids of infinite duration) is inefficient
disease from the ECG signal. ECG would be much more useful as
for ECG signal analysis. On the other hand wavelet, which is a
a diagnostic tool if unwanted noise embedded in the signal is
removed. The aims of the work are to (i) ECG signal
very recent addition in this field of research, provides a
enhancement using empirical mode decomposition (EMD) based powerful tool for extracting information from such signals.
method. (ii) Detection of QRS complexes using continuous
wavelet transform method from the enhanced signal. The The paper is organized as follows: Section II describes the
experiments are carried out on MIT-BIH database. The results theoretical background which includes empirical mode
show that our proposed method is very effective and an efficient decomposition and continuous wavelet transform. Section III
method for fast computation of R peak detection. shows the methodology of the proposed technique,
experimental results and analysis are described in Section IV
Keywords: ECG signal, MIT-BIH Arrhythmias database, Empirical
and Section V represents the conclusions for this paper.
mode decomposition, Wavelet transform, QRS detection.

I.INTRODUCTION
ECG Signal

R peak detection

R peaks
Enhancement

E CG is a nearly periodic signal that reflects the activity of

the heart. A lot of information on the normal and
pathology of heart can be obtained from ECG. However, the
using EMD using CWT

ECG signals being non-stationary in nature, it is very difficult Fig.1 Block diagram of the proposed method
to visually analyze them. Thus the computer based method is
required for ECG signal analysis. II.BACKGROUND
Transmission of ECG often results in the corruption of signal
due to introduction of noise [1]. Various factors are A. Empirical Mode Decomposition Method (EMD):
responsible for introduction of noise including poor channel
conditions, Baseline wander (caused by respiration), 50 or 60 A new non-linear technique, called empirical mode
Hz power line interference etc. Analyzing such a noisy signal decomposition method, has recently been developed by
is bound to give erroneous results. Thus the signal is first N.E.Huang et al for adaptively representing non-stationary
made free of noise, a process called denoising or rather we signals as sums of zero mean AM-FM components [2]. EMD
may call it enhancement. A number of methods have been is an adaptive, high efficient decomposition with which any
incorporated for enhancement of ECG signal. These are filter complicated signal can be decomposed into a finite number of
banks, neural network, adaptive filtering etc. Empirical Mode intrinsic mode functions (IMFs). The IMFs represent the
Decomposition is a recent development which provides a oscillatory modes embedded in the signal, hence the name
powerful tool for decomposing a signal into a finite number of intrinsic mode function. The starting point of EMD is to
IMFs (intrinsic mode functions). ECG analysis basically consider oscillations in signals at a very local level. It is
involves QRS complex detection and subsequent feature applicable to non-linear and non-stationary signal such as
extraction. The block diagram of the proposed method is ECG signal. An intrinsic mode function is a function that
shown in Fig. 1. However, the basic principle of all the satisfies two following conditions [2].
methods involves the transformation of ECG signals using
different transformation techniques including Fourier (i) The number of extreme and the number of zero
transform, Hilbert transform, wavelet transform etc. crossings must differ by at most 1.
Physiological signals like ECG are considered to be quasi- (ii) At any point the mean value of the envelope defined
periodic in nature. Various approaches have been by maxima and the envelope defined by minima
must be zero.
 A systematic way to extract the IMFS is called the sifting Wavelet transforms are applied to decompose the signal
process and its algorithm is given below into a set of coefficients that describe the signal frequency
i) Identify all the extremes of x (t). content at given times.
ii) Interpolate between minima, ending up with a signal
min(t) and similarly between extreme to give max(t). The continuous wavelet transform of the signal, x(t), is
iii) Compute the average: defined as [5].
e(t)=(min (t)+max (t))/2
iv) Extract the detail: d(t) = x(t)-e(t) (Steps 1-4 are
repeated till d(t) satisfies both the criteria of IMF)
v) Iterate on the residual e(t) ∞
1 t −b
In practice, after a certain number of iterations, the resulting
F ( a, b ) =
a ∫ x(t ) *ψ (
−∞
a
)dt (1)

signals do not carry significant physical information. To

prevent this, we go for some boundary conditions. We can Here Ψ(t) is the analyzing wavelet function
stop the sifting process by limiting the normalized standard
a is the dilation parameter and
deviation (nstd) [3].
where b is the location parameter of the wavelet
2
T ( x i (t ) − x i (t + 1)) Fig.2. shows that the Original ECG and seven IMFs of the
nstd = ∑ 2 signal. Actually the wavelets are generated from a single basic
t =1 ( x i (t + 1))
wavelet Ψ(t), the so called mother wavelet, by scaling and
translation.

The nstd is set between 0.2 and 0.3 for proper results [3]. 1 t −τ
The sifting process was applied on an ECG signal to
ψ s ,τ
(t ) =
s
ψ(
s
) dt (2)
obtain the various IMFs. This has been represented in Fig. 2.
Here τ is the scaling factor and √ (1/s) is for normalization
The EMD method is a powerful tool for analyzing ECG
across the different scales. Due to the scaling and translation,
signal. It is very reliable as the base functions depend on the
Wavelet Transform is localized in both time and frequency.
signal itself. It is very adaptive and avoids diffusion and
Several Mother Wavelets like Mexican-hat and Morlet have
leakage of signal.
been used in ECG signal analysis. The mother wavelet has a
1
lot of significance for the efficiency of the process. In this
ECG 0
-1 paper we have used a Haar Wavelet as the mother wavelet
0 100 200 300 400 500 600 700 800 900 1000
0.5
0
because the oscillatory nature of other mother wavelets results
IMF1
-0.5 in several ridges for each ECG component, while only one
0 100 200 300 400 500 600 700 800 900 1000
1 pair of ridges is generated via the Haar wavelet due to its
IMF2 0
-1
0 100 200 300 400 500 600 700 800 900 1000 configuration. The Haar mother wavelet function Ψ(t) as
IMF3 1
0
-1 shown in Fig. 2 can be described as:
0 100 200 300 400 500 600 700 800 900 1000
IMF4
0.5
0 Ψ(t) = 1 0 ≤ t < 1 /2
-0.5
0 100 200 300 400 500 600 700 800 900 1000
IMF5
0.5
0
= -1 1/ 2 ≤ t < 1
-0.5
0 100 200 300 400 500 600 700 800 900 1000
IMF6 0.1
=0 otherwise
0
-0.1
0 100 200 300 400 500 600 700 800 900 1000
-0.05
IMF7 -0.1
-0.15
0 100 200 300 400 500 600 700 800 900 1000
No.of Samples

Fig.2.The Original ECG and seven IMFs of the signal

B. Continuous Wavelet Transform:

Wavelets are also another powerful method for the
representation and analysis of ECG signal. They have been
implemented for the analysis of physiological waveforms like
ECG, Phonocardiogram etc. [4]-[7]. This is because wavelet
has finite duration as compared to Fourier methods based on
sinusoids of infinite duration. Fig. 3 Haar mother wavelet
We have chosen CWT over DWT because unlike DWT there proposed method. Table I shows the SNR improvement after
is no dyadic frequency jump in continuous wavelet transforms enhancement.
[5]. Also high resolution in time-frequency domain is Table II indicates the detection error rates and sensitivity
achieved between proposed (EMD and CWT) method and DOM
method. Here FP denotes false positive i.e. false peak
III. METHODOLOGY
detection and FN denotes false negative i.e. failure to detect.
The total detection error rate (D.E) is calculated as (FP +FN)/
A. Enhancement Technique using EMD:
(Total no of R peaks *100). The average detection error rate is
The basic principle of enhancement of ECG signal using found to be 0.01%. Efficiency is measured in terms of
EMD is expressing the noisy ECG as sum of a series of IMFs. sensitivity given by
It has been shown that the 1st IMF contains nothing but high TP
frequency noise [4]. So we can easily eliminate this Se = x100
component. The next two IMFs contain both noise as well as TP + FN
information. It has been shown that if we remove the 2nd IMF Here TP stands for true positive which is the total no of
there is heavy distortion of the R- peaks [8]. In order to peaks correctly detected by the detector. Sensitivity is
remove noise while preserving the information we go for calculated to be 99.84 % which shows that the method has a
filtering. good efficiency.
The whole procedure can be described by the following 1
algorithm.

Clean ECG
0 [a]
i. The ECG signal is first decomposed into IMFs. The
sum of these IMFs should represent the signal well. -1
0 200 400 600 800 1000 1200 1400 1600 1800 2000
The IMFs are obtained using the sifting process
1
described in the earlier section.
Noisy ECG

ii. The first four IMFs are removed by using low pass 0 [b]

Butterworth filter as the noise comprises the higher

-1
frequency components. We use a Butterworth filter 0 200 400 600 800 1000 1200 1400 1600 1800 2000
because of its inherent characteristics of having a flat 2
Denoisy ECG

frequency response.
0 [c]
iii. The 1st IMF is now eliminated. We reconstruct the
enhanced signal by eliminating the 1st IMF and -2
0 200 400 600 800 1000 1200 1400 1600 1800 2000
adding up the rest IMFs. No.of Samples

B. R-peak Detection using Continuous Wavelet Transform

Fig. 4. (a) Clean ECG signal (b) Noisy ECG signal after the
The basic principle is involved by using a threshold addition of White Gaussian Noise (c) Denoising signal after
detector [5]. The various steps are described below:
enhancement.
i. The enhanced signal is transformed using
continuous wavelet transform. 1

ii. Positive maximum peaks are larger than a 0.5

threshold are selected. The main threshold is
0
chosen as a fraction of root mean square of the
signal. We have chosen this to be around two -0.5
times the root mean square of the signal after
-1
carrying out a series of experiments. 0 100 200 300 400 500 600 700 800 900 1000

R peak denoted by blue dash

2
IV.EXPERIMENTAL RESULTS AND ANALYSIS
1

We have used MIT-BIH database to validate the

efficiency of our proposed method. Simulation is carried out 0

in MATLAB environment. We have added white Gaussian

noise to the clean ECG signals to obtain a collection of noisy -1
100 200 300 400 500 600 700 800 900 1000
ECG signals with SNR varying from 5 dB to 15 dB. The No.of Samples

enhancement algorithm is then used and the SNRs of the

enhanced signals are calculated to find the efficiency of the Fig.5 R peak detection using Continuous Wavelet
Transform
 TABLE I is better as compared to other methods. The average detection
EXPERIMENTAL RESULTS OF ENHANCEMENT METHOD USING EMD error rate for CWT is 0.01% .The sensitivity of CWT is
99.84%.
Noisy signal MIT BIH Record Enhanced Output
SNR(dB) SNR(dB) Thus our method of signal enhancement and R-peak
detection using empirical mode decomposition method and
200_1 8.3
5 continuous wavelet transform is a novel, efficient method
201_1 8.6
having less computation time, hence best suited for analysis of
202_1 8.7
210_1 8.7
ECG signal for clinical purposes.
230_1 8.5
200_1 13.5 REFERENCES
10 201_1 13.7
202_1 13.3
[1] Guodong Tang and Aina Qin, “ECG Denoising based
210_1 13.6 on Empirical Mode Decomposition,” 9th International
230_1 13.3 Conference for Young Computer Scientists, pp. 903-
200_1 18.5 906, 2008.
15 201_1 17.1 [2] N. E. Huang, Z. Shen, S. R. Long, M. C. Wu, H. H.
202_1 17.8 Shih, Q. Zheng, N.-C. Yen, C. C. Tung, and H. H.
210_1 18.3 Liu, “The empirical mode decomposition and the
230_1 17.5 Hilbert spectrum for nonlinear and nonstationary
time series analysis,” Proc. Roy. Soc. Lond, vol. A
TABLE II
COMPARISON BETWEEN PROPOSED AND DIFFERENCE OPERATION METHOD 454, pp. 903–995, 1998.
[3] Hualao Ling, Qiu-Hua Lin, and J.D.X.Chen,
EMD & CWT METHOD “Application of the Empirical Mode decomposition
MIT DOM METHOD
(PROPOSED) to the analysis of Esophageal Reflux Disease,” IEEE
BIH TP
DATA D.E Se D.E Se Transactions on Biomedical Engineering, Vol.52,
FP FN FP FN No.10, 2005.
(%) (%) (%) (%)
200 2601 0 01 0.04 99.9 00 00 0.00 100
[4] Sreeraman Rajan, R. Doraiswami, M. Stevenson, and
R.Waltrous,” Wavelet based bank of correlators
201 1963 2 12 0.71 99.3 03 00 0.15 100
approach for phonocardiogram signal classification”,
202 2136 8 00 0.37 100 08 00 0.37 100 IEEE proceedings, pp. 77-80, 1998.
203 2982 2 30 1.07 98.9 06 01 0.23 99.9 [5] A.Ghaffari, H.Golabayani, and M.Ghasemi, “A new
205 2656 0 01 0.05 99.9 10 00 0.38 100 mathematical based QRS detector using continuous
208 2956 0 05 0.16 99.8 08 00 0.27 100 wavelet transform”, Computers and Electrical
209 3004 0 00 0.00 100 00 02 0.06 99.9 Engineering, vol.34, pp. 81-91, 2008.
[6] Conor McCooey, Dinesh Kant Kumar, and Irena
210 2647 0 04 0.15 99.8 10 00 0.38 100
Cosic, “Decomposition of Evoked Potentials using
212 2748 0 04 0.16 99.8 01 01 0.07 99.9
Peak Detection and the Discrete Wavelet
213 3251 0 02 0.06 99.9 07 02 0.27 99.9 Transform”, IEEE proceeding , pp 2071-2074, 2005.
215 3363 6 00 0.18 100 00 10 0.29 99.7 [7] Szi-Wen Chena, Hsiao-Chen Chena, and Hsiao-Lung
217 2208 0 00 0.00 100 00 00 0.00 100 Chanb, “A real-time QRS detection method based on
219 2154 0 00 0.00 100 01 00 0.05 100 moving-averaging incorporating with wavelet
220 2048 0 00 0.00 100 02 00 0.09 100
denoising”, Computer methods and programs in
biomedicine, vol. 82, pp. 187–195, 2006.
221 2427 0 04 0.16 99.8 04 06 0.41 99.7
[8] Yan Lu, Jingyu Yan, and Yeung Yam, “Model Based
222 2484 0 20 0.81 99.2 06 00 0.24 100 ECG denoising using empirical mode
223 2605 0 00 0.00 100 01 00 0.04 100 decomposition,” IEEE International Conference on
Total 2448.7 1.05 4.88 0.01 99.84 3.9 1.29 0.21 99.94 Bioinformatics and Biomedicine, pp. 191-196, 2009.
[9] Yun-Chi Yeh and Wen-June Wang, “QRS complex
IV.CONCLUSION detection for ECG signal, The Difference Operation
Method”, Computer Methods and Programs in
Our sole objective of this paper is to develop a method Biomedicine, vol. 9, pp. 245-254, 2008.
for efficient analysis of ECG signal. In this piece of work, we [10] D.Benitez, “The use of Hilbert Transform in ECG
have proposed a novel method of enhancement of ECG signal Signal Analysis”, Comput. Biol .Med, vol. 31, pp.
using empirical mode decomposition (EMD). Deviating from 399- 406, 2001.
other approaches of using EMD, we have proposed the use of [11] A.Barros, A.Mansour and N.Ohnishi,“Removing
low-pass Butterworth filters for efficient noise removal and Artifacts from ECG signals using independent
implemented continuous wavelet transform for R-peak component analysis”, Neuro-computing, vol. 22, pp-
detection. We have found that the efficiency in case of CWT 173-186, 1998.
[12] G.D. Clifford and L. Tarassenko, “One-pass training
of optimal architecture auto-associative neural
network for detecting ectopic beats,” Electron Letter,
vol. 37, no. 18, pp. 1126-1127, Aug. 2001.
[13] P.M. Agante and J.P.M de Sa, “ECG noise filtering
using wavelets using soft thresholding methods”, In
Proc. Comput. Cardiology , pp. 535-542, 1999.
[14] Jing- tian Tang,Xiao –li Yang, “The Algorithm of R
peak detection in ECG based on empirical Mode
Decomposition”, IEEE, 4th International Conference
on Natural Computation.
[15] Yun-Chi Yeh, Wen-Jun Wang, and Che Wun Chiou,
“Cardiac arrhythmia diagnosis method using linear
discriminant analysis on ECG signals”, Measurement
vol. 42, pp. 778-789, 2009.