Abstract
In this research an expert system for automated detection of abnormality of electrocardiogram (ECG)
signal is developed. For this purpose, an off-line data acquisition system from paper based ECG records is
developed by using image processing techniques for the future development of an Indian standard ECG
database. Binarization is done for conversion of TIFF formatted gray tone image to two-tone binary
image with the help of histogram analysis, which almost removes the background noise (e g gridlines of
ECG papers). The rest of the dotted noises are removed by runlength smearing technique. Thinning is
also done to avoid the repetition of co-ordinate information in the dataset. The pixel-to-pixel co-ordinate
information is extracted and calibrated using prior information and converted into ASCII data-file. The
present database contains 100 normal and 100 diseased subjects. Out of the diseased 100 patients, 55
patients have Myocardial Infarction (MI) and the remaining patients have Myocardial Ischemia. This work
is reported in the paper I of original paper list which is given in page (iv) of this thesis. The 2 nd chapter is
prepared on the basis of this topic.
The developed ECG database is further processed for removal of six different types of noises that can
corrupt the ECG signals. All the noises are simulated and removed by digital FIR filters with the help of a
software package Cool Edit Pro offered by Syntrillium Software Corporation. This is done to get a realistic
situation for the algorithm and also to get greater accuracy in time-plane features detection.
In the next step, different time-plane features of ECG signals, which are important for ECG interpretation
and classification, are extracted. At first, the R-R intervals (QRS complex) are detected using square
derivative curve of ECG signal. Base-line is also detected for accurate detection of P wave and ST
segments. Standard assumptions are used for detection of baseline, T waves and ST segments region. P
wave is detected from the first derivative of the samples. Depending on the zero-crossings and shape of
the signal, a syntactic approach is used for detection of P QRS and T waves. The accuracy level for QRS
was 99.4%, for T waves was 96.7% and for P waves 92.2%. The QRS or R-R interval detection part is
reported in paper IV whereas the noise removal and the other features extraction methods are reported
in VI and IX. The 3rd chapter of the thesis is prepared on the basis of these papers.
Frequency plane analysis is also done for extracting frequency plane features, which plays an important
role for heart disease categorization. Both amplitude and phase properties are achieved which are
established into ‘if-then’ rule-base for disease classification. This part of the work is described in the 4th
chapter of the thesis and is reported in the papers II, III, V and VIII.
A rule based rough-set decision system is also developed from time-plane features using the most
popular rough-set software toolbox ROSETTA. An accuracy of 100% for normal and MI patients for both
trained and untrained samples are achieved where as for ischemic patients 100% and 95.8% accuracy
obtained for trained and untrained samples. This portion is reported in papers VII and X. 5 th chapter of
the thesis is prepared on the basis of this work.
i
Preface
The dissertation is submitted to the University of Calcutta as a partial
fulfillment of the requirements for the degree of Doctor of Philosophy
in Technology. The research was carried out at the Computer Vision
and Pattern Recognition Unit of Indian Statistical Institute, Kolkata,
India as well as in the Department of Applied Physics, University
College of Technology, University of Calcutta, Kolkata, India during the
years 2000-2004.
ii
Acknowledgement
I would like to express my gratitude to Dr Madhuchhanda Mitra of Department of Applied Physics,
University of Calcutta for giving me the opportunity to do this work and for her proficient supervision
and guidance during the course of study.
I am deeply grateful to Professor B B Chaudhuri of Computer Vision and Pattern Recognition (CVPR)
Unit of Indian Statistical Institute for his patient supervision and advice on the research work. I am
indebted to him for his help in writing the reports, and revising the language of the thesis.
The financial support granted by Council of Scientific and Industrial Research (CSIR), Government of
India is gratefully acknowledged.
I wish to thank Dr Arup Dasbiswas, Associate Professor of Institute of Post Graduate Medical
Education and Research (IPGMER), University of Calcutta [S S K M Hospital, Kolkata] and Dr Ajoy
Sarkar of Peerless Hospital and B K Roy Research Center, Kolkata for their expert opinion and
collaboration regarding the medical aspects of this work.
I would like to thank Dr. S.B. Bhattachariya of Department of Cardiology, S.S.K.M. Hospital(IPGMER),
Dr. B. Hulder, Asst. Prof., Dept. of Cardiology, Burdwan Medical College Hospital, Dr. T. Goswami,
Medical consultant, Dr. B. Bala, Dr. S. K. Roy and Dr. A. K. Das of Uttarpara General Hospital, Dr.
S. Chatterjee of Barakpur General Hospital, Dr. Kanak Kumar Mitra, Asst. Prof. , Dept of Cardiology,
R. G. Kar Medical College, Dr. Basujit Gangopadhyaya of Apollo Clinic, Dr. P. C. Mandal of
Anandalok Groups of Hospital, Dr. D. Ghosh Roy, Dr. T. K. Saha, Dr. Kaushik Mitra and Dr. D.
Bhattachariya for sharing their experience and opinion with us and also for their cooperation and help.
I also wish to thank Mr. U. Garain of CVPR Unit and Dr. A. Raychowdhuri of Jadavpur University for
their help in different occasions.
I thank Goutam Dhar, a final year B. Tech. student for his help to develop the computer program of P
and T wave detection as a part of his final year project
I thank all the other faculty and staff members of CVPR Unit, Indian Statistical Institute and Department
of Applied Physics, University of Calcutta for help and cooperation extended to me during the course of
my research.
Finally, I would like to thank all my family members specially my husband Mr. Saibal Mitra and my parents
Mr. D. L. Sarkar and Mrs. Sipra Sarkar for their unstinted support and inspiration during these years of
my studies.
Kolkata, July 2004
Sucharita Sarkar
iii
List of Original Papers
This thesis is based on the following papers, which are referred to in the abstract by their Roman
numerals. The author is responsible for handling the whole problem, developing the computer
programming for all the modules, doing the experiment, analyzing the data and writing the
manuscripts.
I.
Sucharita Mitra, M Mitra , “An Automated Data Extraction System From 12 Lead ECG
Images”, Computer Methods and Programs in Biomedicine, (Elsevier Science
publication), vol. 71(1), May(2003), pp 33-38.
II. S. Mitra, M. Mitra, “Phase Response Properties of Normal and Diseased ECG Signals
using an Automated Data Acquisition System”, technical journal of The Institution of
Engineers (India), vol. 84, May(2003), pp 14-17.
III. S. Mitra, M. Mitra & B. B. Chaudhuri , “Generation of Digital Time Database from Paper
ECG Records and Fourier Transform Based Analysis for Disease Identification” accepted
for publication in Computers in Biology and Medicine, Elsevier Science publication.
IV. S. Mitra, M. Mitra, “Detection of QRS Complex of ECG Signals from Square-Derivative
Curve”, accepted for publication in the AMSE(France) journal (Advances in Modeling).
V. S. Mitra, M. Mitra, S. B. Bhattachariya & B. B. Chaudhuri , “Generation of Digital Time
Database from Paper ECG Records and Application of Frequency Plane Analysis for
Disease Identification”, Proceedings-CD of International Congress on Biological and
Medical Engineering (ICBME), 4-7 December, 2002 Singapore.
VI. S. Mitra, M. Mitra & B. B. Chaudhuri, "Time-plane Feature Extraction from ECG signals
for Development of Disease Classifier", Proceedings of International Conference on
Communication, Devices and Intelligent Systems (CODIS 2004), 8-10 January, 2004,
Kolkata, INDIA, pp. 653-656.
VII. S. Mitra, M. Mitra, S. Chattopadhyay & B. B. Chaudhuri, “An Approach to a Rough-set
Decision System for Classification of Different Heart Diseases”, Proceedings of
International Conference on Modeling and Simulation, MS’2004, Lyon, France, 5-7 July,
2004.
VIII. Sucharita Mitra, M. Mitra & B. B. Chaudhuri, “Frequency-plane Analysis of Normal and
Pathological ECG Signals in Application of Disease Identification”, accepted in Journal of
Medical Engineering and Technology.
IX. Sucharita Mitra, M. Mitra & B. B. Chaudhuri, “ECG Features Extraction for Analysis with
the Help of an Off-line Data Acquisition Package”, communicated and under review to
IETE Technical Review.
iv
X. Sucharita Mitra, M. Mitra & B. B. Chaudhuri, “A Rough Set Based Inference Engine for
ECG Classification” communicated and under review at IEEE Transactions on
Instrumentation and Measurement.
N. B.
After marriage my surname has been changed from Sarkar to Mitra; and I have
published my papers using my present surname Also note that I had registered as Sucharita
Sarkar at the time of admission to Calcutta University Hence I have continued to use my
maiden name for my Ph D Registration as well as for all official communications with the
University
v
LIST OF FIGURES
Figure1.1 A Schematic Diagram of the Circulatory System
3
Figure 1.2 The Anatomical Structure of Human Heart
4
Figure 1.3 Layers of the Heart Muscle
4
Figure 1.4 Pacemaker potentials and action Potentials in the SA node
6
Figure 1.5 An action potential in a myocardial cell from the ventricles
6
Figure 1.6 An ECG Strip
8
Figure 1.7 The placement of the bipolar leads and the exploratory electrode for the unipolar
chest leads in an electrocardiogram (ECG); (RA = right arm, LA = Left arm, LL = left leg )
9
Figure 1.8 The conduction of electrical impulses in the heart, as indicated by the
electrocardiogram (ECG) The direction of the arrows in (e) indicates that depolarization
of the ventricles occurs from the inside (endocardium) out (to the epicardium), whereas
the arrows in (g) indicate that repolarization of the ventricles occurs in the opposite direction
10
Figure 2.1 Block Schematic of the Developed Off-line Data Acquisition Package
31
Figure 2.2 Gray Level Histogram of An ECG Image
33
Figure 2.3 Runlength Smearing Algorithm
34
Figure2.4 8 neighbors of P1
35
Figure 2.5 Image of original ECG signal from chart record
36
Figure 2.6 Extracted ECG signal before thinning
36
Figure 2.7 Reproduced ECG signal from extracted database after thinning
36
Figure 2.8 Image of original ECG signal from chart record
37
Figure 2.9 Extracted ECG signal before thinning
37
Figure 2.10 Reproduced ECG signal from extracted database after thinning
37
Figure 2.11 Original ECG images from image database and reconstructed signals
from digital time database
37
Figure 2.12 Graphical analysis of standard deviation (for Ordinate)
41
Figure 3.1 Few noisy recorded ECG data
49
Figure 3. 2 (a) ECG Signal with 0% Noise Level
Figure 3.2(b) ECG signal corrupted by 10% white noise
51
51
vi
Figure 3.2(c) ECG signal corrupted by 20% white noise
51
Figure 3.2(d) ECG signal corrupted by 30% white noise
51
Figure 3.3(a) ECG signal corrupted by 10% power line oscillation
52
Figure 3.3(b) ECG signal corrupted by 20% power line oscillation
52
Figure 3.3(c) ECG signal corrupted by 30% power line oscillation
52
Figure 3.4 A screen shot of 50 Hz notch filter working on ECG signal
corrupted by 10% power line oscillation
52
Figure 3.5(a) ECG signal corrupted by 10% base line shift
53
Figure 3.5(b) ECG signal corrupted by 20% base line shift
53
Figure 3.5(c) ECG signal corrupted by 30% base line shift
53
Figure 3.6(a) ECG signal corrupted by 10% abrupt baseline shift
53
Figure 3.6(b) ECG signal corrupted by 20% abrupt baseline shift
53
Figure 3.6(c) ECG signal corrupted by 30% abrupt baseline shift
53
Figure 3.7 ECG signal corrupted with 20% composite noise
54
Figure 3.8(a) Noise free ECG signal
54
Figure 3.8(b) Filtered output of ECG signal
54
Figure 3.9 A screen shot of 0 6 Hz high-pass filter working on ECG signal
corrupted by 20% base line shift
54
Figure 3.10 QRS complex or R-R interval detection
56
Figure 3.11 Filtered output of ECG signal after removal of all simulated noises
57
Figure 3.12 Graphical representation of 2nd order derivative after removal of
all simulated noises
57
Figure3.13 Graphical representation of square of 2nd order derivatives after
removal of all simulated noises
57
Figure 3.14 Reproduced ECG signals with different shapes
58
Figure 3.15 Square Derivative Curve of reproduced ECG signals
58
Figure 3.16 Different time plane features which are extracted
59
Figure 4.1 Time-domain Representation of Sin-wave
68
Figure 4.2 Time-domain Representation of Square Wave
68
vii
Figure 4.3 Spectrum or Frequency-domain Representation of the Square Wave
68
Figure 4.4 Effect in the frequency domain of sampling in the time domain
(a) Spectrum of original signal (b) spectrum of sampling function (c) Spectrum of
sampled signal with fs >2fc (d) Spectrum of sampling function with fs < 2fc
(e) Spectrum of sampled signal with fs < 2fc
70
Figure 4.5 Block Schematic of the Proposed System
73
Figure 4.6 Amplitude diagram of Infarction and Normal patients for lead V4
77
Figure 4.7 Amplitude diagram of Ischemia and Normal patients for lead V6
78
Figure 4.8 Phase diagram of H7 for lead AVL for (a) Infarction Patients,
(b) Normal subjects
82
Figure 5.1 Block Schematic of the Proposed System
90
Figure 5.2 A Screenshot of Generated Rule-set
96
Figure 5.3 Confusion Matrix Output for Standard Voting Classifier
96
viii
LIST OF TABLES
Table 2.1 QRS Portion Of Extracted Database Of The Image Shown In Figure 2 5
39
Table 2.2 QRS Portion Of Extracted Database Of Another Image
40
Table 3.1 QRS Detection Accuracy In Different Noise Levels
61
Table 3.2 T Wave Detection Accuracy In Different Noise Levels
62
Table 3.3 P Wave Detection Accuracy In Different Noise Levels
62
Table 3.4 A Portion Of Extracted Time Plane Features
62
Table 4.1 A Portion Of Numerical Values Of Amplitude And Phase Of ECG After DFS For Lead V6
75
Table 4.2 A Portion Of Numerical Values Of Amplitude And Phase Of ECG After DFS For Lead V4
76
Table 4.3 A Portion Of List Of Sum Of First Five Harmonics
79
Table 4.4 A Portion Of List Of Sum Of First Two Harmonics
79
Table 4.5 Phase Properties Of Normal Vs Myocardial Infarction
80
Table 4.6 Phase Properties Of Normal Vs Myocardial Ischemia
80
Table 4.7 Result Obtaine From Rule Based Classifier
81
Table 5.1 A Potion Of Decision Table
93
Table 5. 2 Result Obtained From Rule Based Rough Set Decision System
95
ix
CONTENTS
Abstract
i
Preface
ii
Acknowledgement
iii
List
of
Original
Papers
iv
List of Figures
vi
List of Tables
ix
Contents
x
1. General Introduction
1
The Heart
1
Anatomy
2
Electrical Activity of the Heart
5
Clinical Observation of Heart
6
Electrocardiography (ECG)
Cardiac Diseases Interpreted by ECG
7
10
Computer Analysis of ECG and Historical Background
13
Thesis Overview
17
Literature Survey
20
ECG Data Extraction and Signal Analysis
21
ECG Feature Extraction and Time-plane Analysis
23
Frequency-plane Analysis of ECG Signals
25
ECG Classification and Abnormality Detection
27
2. Automated Data Extraction from 12 Lead ECG Images
29
Introduction
29
Materials and Detailed Methods
31
Binarization of the Input Images
31
Removal of Grid Lines from Graphical Papers
33
Thinning of the Input Signal
34
Raw Data Extraction
35
Data Sorting and Regeneration of the ECG Signal
36
Results
38
Discussions
41
x
3. Noise Removal and Time-plane Features Extraction from ECG Signal
43
Introduction
43
Noise Removal from Extracted ECG Signals
45
Noise Characteristics
45
Different Filters and their Usage
47
Noise Simulation and Removal from ECG Signals
49
A Few Important Time-plane Features
54
Time-plane Features Extraction
56
QRS Complex Detection
56
Base Line Detection
58
P, ST Segment and T Wave Detection
59
Detailed Algorithm
60
Results
61
Discussion
64
4. Frequency Plane Features Extraction from ECG Signals
65
Introduction
65
Frequency Plane Transformation of Digitized ECG Data
66
The Theory of Analysis of Digitized ECG Wave
71
Experimental Procedure
72
Result
73
Discussions
82
5. A Rough Set Based Inference Engine from Different Time-plane Features for ECG Classification 85
Introduction
85
Rough Set- A Tool for Representing and Reasoning about Imprecise or Uncertain Information 87
Proposed Method
90
Knowledge Base Development
90
Development of Inference Engine
92
Result
93
Discussion
97
6. Conclusion and Future Scopes
99
References
105
Appendix A
119
Appendix B
120
xi
Annexure 1 (Publication List)
121
Annexure 2 (Different Modules)
123
Vita
144
Original Papers
145
xii