See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/336312280

Dynamic Feature Extraction of ECG Signal

Conference Paper · May 2014

CITATIONS READS
0 13

4 authors, including:

Rajveer K Shastri Deosarkar Shankar

Vidya Pratishthan's Kamalnayan Bajaj Institute of Engineering & Technology, Baram… Dr.Babasaheb Ambedkar Technological University , Lonere I dia
38 PUBLICATIONS   114 CITATIONS    58 PUBLICATIONS   144 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Multi band Antenna for Wireless Applications View project

Gaussianity and Stationarity analysis of Underwater Ambient Noise Data View project

All content following this page was uploaded by Rajveer K Shastri on 07 October 2019.

The user has requested enhancement of the downloaded file.

 210 IWCEM, TS-5: Students Contributory, Paper No. C60, May 2014

Dynamic Feature Extraction of ECG Signal

Juie Peshave1 , Rajveer Shastri2 , Arnab Das3 , Shankar Deosarkar4
1,2,4 VPCOE, Baramati,

University of Pune, India

3 MILT, Pune, India
1 juiepeshave@gmail.com, 2 rajveer shastri@yahoo.com, 3 arnab1972@hotmail.com, 4 sbdeosarkar@yahoo.com

Abstract—Electrocardiogram (ECG) is one the important Previously time domain characteristics of the ECG signals
biomedical signal. One heartbeat of ECG consists of different were analyzed but this does gives all information about features
segments such as QRS complex, ST segment and PR segment. therefore frequency analysis is required. As analysis of ECG
Features of an ECG signal are nothing but these segments and signal provides information about the cardiac diseases, it is
intervals between fiducial points such as RR interval, amplitude important to extract features from signal accurately.
of P, R and T wave. Several techniques are discovered and
are still developing for analyzing ECG signal. Some of them In the past, there was much work on heartbeat feature
are Continuous Wavelet Transform, Discrete Wavelet Transform extraction. These works gives different characteristics of the
and Pan Tompkins Algorithm. In this paper, with the help heartbeats using several methods such as morphological analy-
of extracted dynamic feature heart rate is determined using sis in which inflection points in the ECG signal are determined
wavelet transform. This system is validated on standard MIT- using first and second derivative method [3]- [4], hermite
BIH arrhythmia database and it yields about 80% of sensitivity. function in which coefficients of the hermite polynomials
provides information about the width of the QRS complexes
[5]- [6], higher order statistics explores symmetry property
Keywords—Electrocardiograph, Feature extraction, Discrete using skewness and kurtosis [6], continuous wavelet transform
Wavelet Transform(DWT). gives position of QRS complex but its performance depends
on provided scale and pan tompkins algorithm [7]. These all
algorithms are made valid on MIT-BIH arrhythmia database
I. I NTRODUCTION available on site physionet [8].
The paper is organized as follows: in section II basic
ECG can be recorded by noninvasive technique with the framework of overall procedure is explained; section III in-
help of placing surface electrodes on limbs or chest. With the forms dataset; section IV explains theoretical background and
application of some advance signal processing techniques we algorithm; results are validated in section V and section VI is
can predict different diseases of cardiac systems [1]. the conclusion.
An ECG signal consists of P-QRS-T components which are
having specific magnitude and intervals. In feature extraction
method amplitude and intervals between these components are II. F RAMEWORK
calculated to get abnormality in the heart rhythm [2]. Arrhyth- Following Fig. 2 shows basic framework of arrhythmia
mia is one of the life-threatening disease can be monitored monitoring.
by extraction of features. Arrhythmia is basically categorized
into 5 classes; here bradycardia, tachycardia and premature
ventricular contraction are detected.

Fig. 2. Basic framework of arrhythmia monitoring

A. Recording of an ECG
ECG signal is recorded by placing surface electrodes on
human chest using leads V1,V2,V3,V4,V5 and V6[2].In some
cases, because of coughing lead positions gets displace and
Fig. 1. Features of ECG signal recoded signal become noisy hence modified limb leads are
used which are place on hip or shoulder.
 IWCEM, TS-5: Students Contributory, Paper No. C60, May 2014 211

B. Preprocessing are reorganized into 5 classes namely N (normal beats), S

The preprocessing of raw ECG is necessary to reduce (Supraventricular ectopic beat), V (Ventricular ectopic beat),
various types of noise in order to improve signal-to-noise F (Fusion beat), Q (Unknown beat).
ratio. General sources of noise are baseline wander, power line
interference due to electrode movement etc [1]. IV. M ETHODOLOGY
This section provides brief information of the QRS feature
C. Heartbeat Segmentation detection, gives detail information of this process and its
To segment complete heartbeat, fixed size segment are theoretical background. Section II gives all basic steps for
considered. The advantage using of fixed segment size is, it making decision of whether the recorded heartbeat is normal
can avoid detection of fiducial points related to other waveform or abnormal. Ectopic heartbeats (Arrhythmia) are detected on
components (P wave or T wave) which are very sensitive to the basis of distorted waveform shape or some missing im-
noise. The limitation of using fixed size segment comes in portant components from heartbeat. Since, cardiac arrhythmia
case if heart is beating faster, the interval between two beats is associated with variability and irregularity in the heartbeats
become smaller and segment may contain information about it is beneficial to utilize combination of morphological and
the neighboring beats, which gives wrong results. So adapting dynamic features of the heartbeat. Morphological features
heartbeat segment size is more preferable [2]. include components inside the heartbeat whereas dynamic
features include the components in between the heartbeats. In
this paper, to detect arrhythmia disease wavelet transform is
D. QRS Detection
applied and dynamic features are extracted from corresponding
Segmentation follows feature extraction which mainly in- output which is explained below.
cludes QRS detection and thereby arrhythmia is detected.

III. DATABASE A. Wavelet Transform

Standard database is available on site called PhysioNet. ECG has characteristics which changes over time and po-
PhysioNet provides free access to large collections of recorded sition. Therefore it is necessary to use the technique which is
physiologic signals (PhysioBank) and related open-source soft- applicable to non-stationary signal (ECG).Wavelet transform
ware (PhysioToolkit). Here data from the MIT-BIH arrhythmia has capability to analyze non-stationary signal, such as ECG
(PhysioBankATM) were used, which includes recordings of signals [1].
many common and life-threatening arrhythmias. Signals have Before application of wavelet on raw signal, selection of
been recorded from different leads such as V1, V2, V4, V5 and proper wavelet is important. Three different parameters are
MLII. The AAMI recommended practice was used to combine estimated:
the MIT-BIH heartbeat types into five heartbeat classes which 1 Energy of coefficients
were used in all processing. 2 Noise in coefficients
3 Reconstruction error
Wavelet Used 100mV5 111mMLII 106mVI
Haar 99.9822 99.9838 99.9962
db4 99.9910 99.9949 99.9977
db8 99.9906 99.9949 99.9976
sym3 99.9870 99.9923 99.9968
sym7 99.9906 99.9949 99.9973

Table I shows energy of approximate and detail coefficients

which must be as high as possible. For extracting ECG
components detailed coefficients are preferred.
Wavelet Used 100mV5 111mMLII 106mVI
Haar 2.0967 2.0967 3.1450
db4 1.2741 1.4710 1.5085
db8 1.1905 1.2331 1.2157
sym3 1.4260 1.5952 1.5894
sym7 1.1957 1.2999 1.2711

After application of wavelet, original signal is decomposed

into specified levels. Coefficients in each level has different fre-
quency content based on which features are detected. Therefore
Fig. 3. Dataset
it is important to have less noise in coefficients.
From Table I, II and III daubechies of order 4 performs
Arrhythmia is categorized into 16 different heartbeat classes. better in all respect, so Db4 was selected to extract ECG
According to ANSI-AAMI standard these original 16 classes features.
 212 IWCEM, TS-5: Students Contributory, Paper No. C60, May 2014

Therefore, in this paper daubechies wavelet having order

4 is selected and original signal is decomposed. Signal can 60
(1)
be decomposed into single level or multilevel. In case of RRinterval
multilevel decomposition, signal decomposition is iterative and
signal is fragmented into lower resolution components. Fig.
4 shows multilevel decomposition in which approximates are
broken down into many levels.

Fig. 4. Multilevel DWT

Fig. 6. Positions of R wave

The sampling frequency of the ECG signal is 360 Hz, this Above signal is from MIT-BIH database which is taken
shows that maximum frequency present in signal is 180Hz. over 10 seconds of period. As, for 60 seconds there should
Most energy of the QRS complex lies in about 40 Hz hence be presence 60 to 100 beats therefore for 10 seconds number
QRS complexes are detected in second level detail coefficients. of beats should be in between 11 to 16. According to this
For orthogonal wavelets, using reconstruction coefficients, concept above signal is normal as it gives heart rate about 14.
coefficients are transformed into wavelet details. The details
have same length as original signal so the locations of R from
original signal can be determined easily [2], [9], [10]. V. E XPERIMENTAL R ESULTS
Fig. 5 shows original signal and detected positions of QRS From the Fig. 6 the average program sensitivity is calcu-
complexes. lated based on detection of QRS complexes and its position
Following to this, these positions are plotted on original detection. Table IV shows sensitivity result.
signal with the help of which heart rate is determined. This is Sensitivity can be calculated using True Positive (TP) and
shown in Fig. 6 False Negative (FN) values.
TABLE I. SENSITIVITY OF DETECTED QRS COMPLEX
Method Used TP FN Sensitivity
Positions of QRS complexes Wavelet 28 7 80%

VI. C ONCLUSION AND F UTURE S COPE

In this paper, we presented the proper way of QRS complex
detection. Novel method of wavelet transform determines
dynamic features and thereby heart rate. With the use of MIT-
BIH arrhythmia database sensitivity is calculated which yields
80%.
Currently, QRS detection process is performed using
daubechies 4 wavelet and thresholding method. Further new
wavelet can be designed which is able to reconstruct original
ECG signal perfectly.

R EFERENCES
Fig. 5. Original signal and detected positions of QRS complexes
[1] M. Rangayyan ,“ Biomedical Signal Analysis: A case study approach
IEEE Press Series in Biomedical Engineering. John Wiely and Sons,
Heart rate can be expressed as Singapore, 2002.
 IWCEM, TS-5: Students Contributory, Paper No. C60, May 2014 213

[2] C. Ye, B. V. K. Vijaya Kumar, “Heartbeat Classification Using Mor- Shankar Deosarkar received his BE Degree in
phological and Dynamic Features of ECG Signals,IEEE transactions on Electronics in 1988 from Amravati University and
biomedical engineering, vol. 59, no. 10, October 2012. his both M. Tech and Doctorate Degrees in the area
[3] P. D. Chazal, “ Automatic classication of heartbeats using ECG mor- of Microwave Communication in 1990 and 2004
phology and heartbeat interval features, IEEE Trans.Biomed. Eng., vol. respectively from S.G.G.S. Institute of Engineer-
51, no. 7, pp. 11961206, Jul. 2004. ing and Technology, Nanded. He has to his credit
about 35 research publications at the National and
[4] P. D. Chazal and R. B. Reilly, “A patient-adapting heartbeat classifier
International level. At present he is guiding three
using ECG morphology and heartbeat interval features, IEEE Trans.
Research Scholars in the area of EMI / EMC and
Image Process., vol. 16, no. 1, pp. 188-197, Jan. 2007.
Microstrip Antenna Design. He was invited to deliver
[5] M. Lagerholm, C. Peterson, G. Braccini, L. Edenbrandt, and L. Sornmo, scholarly lectures at McGill University, Montreal,
“Clustering ECG complexes using Hermite functions and self-organizing Canada, Electromagnetic Research Center Ottawa, Government of Canada and
maps, IEEE Trans. Biomed. Eng., vol. 47, no. 7, pp. 838848, Jul.2000. Princeton University, New Jersey, USA and UGC / AICTE Refresher Courses.
[6] S. Osowski, L. T. Hoa, and T. Markiewic, “Support vector machine-
based expert system for reliable heartbeat recognition, IEEE Trans.
Biomed.Eng., vol. 51,no. 4, pp. 582589, Apr. 2004.
[7] C. Meyer, J. F. Gavela, and M. Harris, “Combining Algorithms in Auto-
matic Detection of QRS Complexes in ECG Signals, IEEE transactions
on information technology in biomedicine, vol. 10, no. 3,July 2006.
[8] MIT-BIH(Massachusetts Institute of Technology-Beth Israel Hospital)
database.
[9] M. Llamedo and J. P. Martnez, “Heartbeat Classification Using Feature
Selection Driven by Database Generalization Criteria, IEEE transactions
on biomedical engineering, vol. 58, no. 3, march 2011.
[10] J. P. Martnez, R. Almeida, S. Olmos, A. P. Rocha, and P. Laguna,
“Wavelet-Based ECG Delineator: Evaluation on Standard Databases.

Juie Peshave received the B.E. Electronics and

Telecommunication in 2011 from Pune University,
Pune and she is pursuing M.E. Degree in Electronics
with specialization in Digital System from Vidya
Pratishthan’s college of Engineering, Baramati, In-
dia. She has presented paper in IEEE ICCSP con-
ference at Melmaruvathur. Her area of research is
Biomedical signal Processing.

Rajveer Shastri received the B.E. Electronics and

Telecommunication in 2000 from Dr. Babasaheb
Ambedkar Marathawada University, Aurangabad, the
M.E. Degree in Electronics with specialization in
Computer Technology from Shri Guru Govind Singh
Engineering and Technology, Nanded, He is now
pursuing Ph.D. degree, in Electronics from Swami
Ramanand Tirth Marathawada Univeristy Nanded.
He worked as lecturer from 2002 to 2008 and from
2008 to till date he is working as assistant professor
in Vidya Pratishthan’s college of Engineering, Bara-
mati, India. He is life member of IEEE, ISTE and IETE. He has published
10 international journal, conference papers and 5 national conference papers.

Arnab Das received B. Tech degree in Electrical

Engineering from Naval College of Engineering,
Lonavala, India in 1995. He obtained M. Tech degree
in Radar and Communication with specialization
in Underwater Signal Processing in 2002and PhD
degree in 2009 from Indian Institute of Technol-
ogy, Delhi, India. He was commissioned into the
Indian Navy in 1994. He worked as Project Officer
at CARE, IIT Delhi for R & D in Underwater
Signal processing and hardware design in 2003-
2007. Currently he is instructor at MILIT Pune,
India. His research interests are Underwater Passive Surveillance, Underwater
Communication and Pattern recognition.

View publication stats