ArticlePDF Available

Origins of ECG and Evolution of Automated DSP Techniques: A Review


Abstract and Figures

Over the years, researchers have studied the evolution of Electrocardiogram (ECG) and the complex classification of cardiovascular diseases. This review focuses on the evolution of the ECG and covers the most recent signal processing schemes with milestones over the last 150 years systematically. Development phases of ECG, ECG leads, portable ECG monitors, Signal Processing schemes and Complex Transformations are discussed. This paper summarizes the development of ECG features detection for cardiac anomalies and the history of the development of ECG monitors, beginning from String Galvanometer. It also discusses the automated detections on ECG, beginning from 1960 to the most recent signal processing techniques. Additionally, this paper provides recommendations for future research directions.
Content may be subject to copyright.
Received September 25, 2021, accepted September 30, 2021, date of publication October 13, 2021,
date of current version October 21, 2021.
Digital Object Identifier 10.1109/ACCESS.2021.3119630
Origins of ECG and Evolution of Automated
DSP Techniques: A Review
VLSI and Embedded Systems Research Laboratory, Dhirubhai Ambani Institute of Information and Communication Technology, Gandhinagar 382007, India
Corresponding author: Biswajit Mishra (
This work was supported by the Science and Engineering Research Board (SERB) under the Department of Science and
Technology (DST), Government of India, under Grant CRG/2019/004747.
ABSTRACT Over the years, researchers have studied the evolution of Electrocardiogram (ECG) and the
complex classification of cardiovascular diseases. This review focuses on the evolution of the ECG and
covers the most recent signal processing schemes with milestones over the last 150 years systematically.
Development phases of ECG, ECG leads, portable ECG monitors, Signal Processing schemes and Complex
Transformations are discussed. This paper summarizes the development of ECG features detection for
cardiac anomalies and the history of the development of ECG monitors, beginning from String Galvanometer.
It also discusses the automated detections on ECG, beginning from 1960 to the most recent signal processing
techniques. Additionally, this paper provides recommendations for future research directions.
INDEX TERMS Automatic, classification, databases, ECG, Einthoven, evolution, signal processing.
In 1902, Dutch scientist Willem Einthoven invented the
String Galvanometer to measure the electrical cardiac activ-
ity that has become one of the century’s most significant
contributions. This invention of ElectroCardioGram (ECG)
or EleKtrocardioGram (EKG) revolutionizes the diagnoses
of cardiovascular anomalies. The history behind the subject
covers similar developments occurring with oscillographs
and electrometers. Therefore, researchers are interested in
knowing the developments of ECG and its origin that led to
the global acceptance of one of the most influential research
of the century.
The ECG signal’s evolution is well studied and reported
in [1]–[3]. In [1], the authors discuss ECG background and
the development of physiological instrumentation, oscillo-
graphs, String Galvanometer and the efforts by Cambridge
Scientific Instruments (CSI) company to make ECG monitors
practically usable in hospital settings. In [2], the authors
discuss Thomas Lewis’s role for his efforts to make ECG
globally accepted, the development of the String Galvanome-
ter, Cambridge electrocardiograph machine, including the
changes in the design for medical use. In [3] authors discuss
the exciting history of ECG origin, electrophysiology practice
in 19th century, the contribution of Willem Einthoven, String
The associate editor coordinating the review of this manuscript and
approving it for publication was Sudipta Roy .
Galvanometer and early definitions of cardiac arrhythmias.
It also discusses the American observations of the ECG,
the role of Thomas Lewis and the development of Electrocar-
diography, ECG and Myocardial Infarction (MI), precordial
leads, and augmented limb leads, Vectorcardiogram in clin-
ical physiology and the challenges of Electrocardiography.
While these reviews focus on the historical development
of the ECG, a recent review [4] focuses on various ECG
signal processing research development that has occurred
during 2000-2020. In [5], authors discuss several automatic
detection methods for Myocardial Infarctions in their review.
Authors in [6] discussed ECG based detection and prediction
models leading to sudden death due to cardiac failure.
This review focuses on the developments in the cardiac
health monitoring schemes starting from ECG origin to the
most recent ECG signal processing techniques. We have
started with the work presented by Willem Einthoven and
studied the history behind his work to cover the roadmap of
ECG development.
Though the history and evolution of ECG are reported
in [1]–[3] to the best of authors’ knowledge, no holistic
review has been reported in the literature that covers the
ECG Evolution, early signal processing techniques, to the
most recent techniques in a systematic manner. This review
discusses the precursors of ECG, ECG origination, feature
detection for cardiac health monitoring, development of elec-
trodes and leads, and the development of bedside monitors
VOLUME 9, 2021
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.
For more information, see 140853
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
from the early string galvanometers. We also discuss the
signal processing techniques during 60’s -90’s, various stan-
dard databases, electrodes, different cardiac health monitor-
ing schemes, and recent trends in these processing methods.
The review also discusses the limitations of existing machine
learning and neural network based classifiers. The trade-
off between the complexity of signal processing techniques
and hardware and software system requirements for real-
time systems is also discussed. The shortcomings of utilizing
the standard databases, in terms of different demographics,
changing definitions of diseases over the period, are some of
the critical findings of this review.
We have studied the papers presented by William
Einthoven and the pioneering research in this field for the
article selection. Further, we have selected the most cited
papers from Journals and conferences to follow the timeline
of biomedical engineering development for cardiac health
monitoring schemes. During the years 2000-2020, abundance
of papers can be found in this stream related to front end
development for ECG signal acquisition, ECG sensors, ECG
electrodes, automatic signal processing with various com-
pression techniques, etc. In this review, we have mainly con-
sidered the work related to the automatic signal processing
domain. To the best of the authors’ knowledge, a broad
review in this domain is not presented in the literature cov-
ering the history of automatic signal processing development
schemes. We have not included the commercial systems
available for cardiac health monitoring in the market in this
review as the main focus of the work is to determine recent
trends in automatic signal processing techniques. Addition-
ally, the hardware of wearable devices is not the focus of
the review. Interested readers can find the wearable devices
review in [7].
Advances in oscillographs were significant for the develop-
ment of ECG as they provided varying means for recording
the alternating voltage. The first oscillograph that enabled
the recording of electrical variations by Blondel in 1893 is
assumed to be the first predecessor of ECG [8]. The elec-
tromechanical oscilloscope consisted of a moving part to pro-
vide oscillations to detect electrical current passing through
it. He offered three probable solutions for the recording of
electrical variations. The first approach was based on the
moving magnet principle, the second on the moving coil, and
the last was to adapt the telephones for recording purposes.
He chose the moving magnet approach to record the elec-
trical variations until this time; moving coil galvanometers
were exceptional and not used to record the electrical signal
The moving coil principle provided by the Blondel
was the second generation of ECG’s precursors. Dudell,
in 1897 [9] replaced the conventional moving magnets with
Phosphor Bronze Strips that utilize the moving coil princi-
ple along with a mounted mirror that reflected a beam of
light. The reflected beam fell on the photographic plate and
FIGURE 1. Dudells oscillograph (Picture credit: [9]) A is the brass oil bath
in which two vibrators are fixed B, core of electromagnet which is excited
by two coils C, one of the two coils D, two pairs of terminals for
connecting the two coils E, bolts that hold the oil bath in position
between the poles of the magnet F, two of three leveling screens (one is
hidden behind the oil bath) G, terminals of one vibrator H, fuse K,
thermometer with bulb in center of oil bath.
provided a magnified recording of the movement of phosphor
bronze strips. Dudell’s oscillograph is shown in Fig. 1.
In 1897, Clement Ader developed a Galvanometer [10],
this significant invention aimed to boost the telegraphic trans-
mission speed on long cables. The transmission principle
was based on fine metal wires with 20 µmdiameter, vibrat-
ing between the poles of large magnets. This galvanometer
by Ader is perceived to be the first string galvanometer.
When Willem Einthoven started experiments on recording the
heart’s electrical activity in 1902, by his String Galvanometer,
he was unaware that Clement Ader had already developed a
similar system. Notably, the principle of operation for both
the galvanometers was the same: a String was employed to
record the electrical variation between large poles of magnets.
Einthoven’s experiment was successful in its own right as
later it was observed that Ader’s Galvanometer’s sensitivity
was lower than Einthoven’s String Galvanometer. And this
would not have been adequate for recording the physiolog-
ical signals from the human body during the experiments.
When Einthoven learned about Ader’s String Galvanome-
ter, he credited Ader and the researchers involved in these
researches in one of his early papers [11].
By now, several groups were already working on recording
the heart’s electrical activity both from the signal and physi-
ological point of view. Fig. 2shows a brief overview of such
events during the late 1800s to 1900.
The first known successful event of recording the electrical
activity of the heart was performed by Alexander Muirhead
between 1869 - 1870 at St. Bartholomew’s Hospital in Lon-
don using the Thomson Siphon Recorder [12], [13] as shown
in Fig. 3. It was developed by William Thomson, a Telegraph
Engineer. Muirhead recorded the ECG signal only once and
after that, he never followed the research on physiological
Much before the Galvanometer experiments, Marey
demonstrated theoretically in 1861 that it is possible to use
a Capillary Electrometer to measure the electrical activity
of the human heart. It was never investigated practically
on humans, but the electrical activity of the heart of the
140854 VOLUME 9, 2021
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
FIGURE 2. Precursor’s of ECG and events of measuring electrical activity of heart till the string
galvanometer development for ECG measurement.
FIGURE 3. Syphon recorder used by Alexander Muirhead to measure the
cardiac activity (Picture credit: [13]).
frog was demonstrated using the Capillary Electrometer
[14], [15]. The electrometer was invented by Gabriel Lip-
mann in 1872 while working in G. R. Kirchoff’s laboratory
in Berlin. It was used to detect minor potential differences
applied to the thick end filled with mercury and the thin end
with the sulphuric acid solution (see Fig.4a).
Around the same period, Augustus D. Waller used Marey’s
technique and applied it to the exposed hearts of mammals.
It led to the successful event of recording the electrical
activity of the heart in 1887 using Lipmann Capillary Elec-
trometer [16], [17]. The ECG wave obtained by Waller [18]
and the experimental setup at Royal Society [19] is shown
in Fig. 4. This event was widely publicized in local London
Newspapers. Interestingly after this, he was investigated and
tried under the ‘Animal Cruelty Act’ for his experiments
involving his pet dog, Jimmie. Also, Waller himself was not
convinced that it could be used widely in the biomedical
domain. He went on further and stated, ‘‘I do not imagine that
electrocardiography is likely to find any very extensive use in
the hospital. . . It can at most be of rare and occasional use
to afford a record of some rare anomaly of cardiac action.
Waller was not the first to use the term ‘‘Electrocardiogram,
and could not perhaps foresee this as ‘‘the future’’ in Medical
Technology and this is one of the reasons that his contribu-
tions are not acknowledged widely.
By this time, the most notable research findings were coming
out from Willem Einthoven’s String Galvanometer experi-
ments [20], [21] that measured cardiac potentials in 1902.
His experiments demonstrated that String Galvanometer was
easier to use, free from damping, and more sensitive than the
Capillary Electrometer. The String Galvanometer is shown
in Fig. 5. His contributions revolutionized the biomedical car-
diac signal acquisition forever. Einthoven was rightly entitled
as the ‘‘Father of Modern Electrocardiography’’.
The first deflections of the cardiac activity were named A,
B, C, D by Einthoven et al. [22] shown in Fig. 6a. A mathe-
matically corrected version of deflections named as P, Q, R, S,
T was superimposed on the former ones shown in Fig. 6a [23].
The naming conventions P, Q, R, S and T are still used to
represent an ECG signal as shown in Fig. 6b. The reason
for changing the cardiac deflection’s name from A, B, C,
D to P, Q, R, S and T is still unclear but most likely it may
have been done to include the successive points in the ECG
tracings [24].
According to Einthoven, in [20], [21], the specifications
for the String Galvanometer (Fig. 5a) is described as ‘The
String Galvanometer is essentially composed of a thin silver-
coated quartz filament (about 3µm thick), which is attached
like a string in a strong magnetic field. When an electric
current is conducted through this quartz filament, the filament
reveals a movement that can be observed and photographed
using considerable magnification, this movement is similar to
the movement of the capillary electrometer. It is possible to
regulate the sensitivity of the galvanometer very accurately
within broad limits by tightening or loosening the string’.
The original apparatus weighed around 600 lbs and
required approximately two rooms for placing it. Provision
for cooling the electromagnets was provided with continuous
water flow. Saline water in buckets was used as electrodes on
the left leg, left arm and right arm locations shown in Fig. 5b.
Einthoven demonstrated significant differences in normal
and abnormal ECG waveforms in 1906 [25] and 1908 [26].
In the experiments to follow, in 1912 [27], Einthoven inves-
tigated and found out that the heart creates a potential differ-
ence at different locations and the magnitude and direction of
VOLUME 9, 2021 140855
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
FIGURE 4. Waller’s experimentation to measure the electrical activity of
the current changes at different locations of the heart can be
represented by the Einthoven Triangle is shown in Fig. 7.
He demonstrated that Lead I had advantages for judging the
T waves, in Lead II peaks were usually larger and Lead III was
most suited for the diagnosis of Ventricular hypertrophy1of
left and right ventricles. He also observed a linear relationship
between the three leads that yielded Lead II Lead I =
Lead III to obtain any lead by combining the other two leads.
By then, Europe had accepted electrocardiograms and the
rest of the world followed. The ECG research mainly evolved
into three groups (see Fig. 8): the first group was working
to categorize the ECG signals for cardiac conditions: mainly
consisted of Physicians; the second group was working on
optimizing electrodes and leads, and the third group was
working on optimizing the size of string galvanometer and
designing portable bedside monitors, mostly technological
The need to extend the ECG device to bedside monitors
was a deriving factor for Einthoven. In 1903, he approached
Cambridge Scientific Instruments Company Limited (CSI) to
reduce the String Galvanometer’s size to a more practicable
form. CSI became interested in the idea and produced their
first String Galvanometer around 1905 (Fig. 5b) with 10%
royalty fees going to the inventor. The first instrument was
sold to MacDonald’s laboratory in Sheffield in 1905, the sec-
ond to J.C. Bose at Presidency College, Calcutta in 1906 and
the third to Keith Lucas in Cambridge in 1907. Later, Dudell
made some modifications to the original design to reduce
the instrument’s size, which resulted in a reduced royalty to
Einthoven. Edward Schafer of the University of Edinburgh
bought the advanced version of the String Galvanometer
and was the first to buy the string galvanometer for clinical
use in 1908. Subsequently, after world war- I, this ECG
machine evolved meant to be placed by the bedside. After
a few modifications to the design, Harold Segall designed the
instrument [1] that could be carried in two wooden cases, each
weighing around 50 lbs. It was the beginning of portable ECG
The use of vacuum tubes for a reduced form factor and
to amplify the electrocardiogram instead of the mechanical
amplification by the string galvanometer is said to be used
by 1925, reported by both Fye [3] and Parker et al. [28].
Subsequent advances in electronic components resulted in the
first portable ECG machine by 1928, that was powered by a
vehicle battery. This was until the invention of increasingly
smaller transistor electronics [3]. More recently, microchips
allowed for developing the 12 lead ECG that we are familiar
with today. Around 1935, Sunborn Company designed the
ECG machine that could be kept in a wooden box weighing
1Ventricular hypertrophy is thickening of the walls of a ventricle (lower
chamber) of the heart.
140856 VOLUME 9, 2021
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
FIGURE 5. Einthoven’s string galvanometer to measure the electrical activity of the heart.
around 25 lbs [1]. The invention of transistors by the 1960s
made the ECG machines portable for use in hospitals and the
invention of Holter2recorders in 1961 [29] paved these for
usage in out of hospital settings.
Following Einthoven’s invention, scientists became inter-
ested in this emerging field and started categorizing the sig-
nals based on their interpretations. Fig. 9shows various ECG
features such as R-R interval, PR interval, QT interval and ST
interval based on P wave, QRS complex, T wave, J point and
Baseline level of the signal.
Einthoven, in 1906 categorized normal and abnormal
ECGs that were translated by Cardiologist Henry Black-
burn [30]. He discussed the first electrocardiographic tracings
of atrial fibrillation,3premature ventricular contractions,4
ventricular bigeminy,5atrial flutter.6The beginning of ECG
2Named on the scientist Norman Jeff Holter who developed it.
3Cardiac anomaly where R-R interval is abnormal and P wave is missing
at instances.
4Abnormal heartbeat where contractions begin in the ventricles, instead
of Sinoatrial node of heart.
5Arrhythmia where there is a pattern of irregular heartbeat and regular
heartbeat occurrence.
6Type of arrhythmia, where the heart’s upper chambers (atria) beat too
related research was also demonstrated in an experimental
setup that induced heart block in a dog, as shown in Fig.10.
By then, Thomas Lewis, a physician, was convinced about
the significance of Einthoven’s contribution to determine var-
ious heart anomalies. Independently, he concluded that atrial
fibrillation is a common cause of arrhythmia and termed as
a ‘‘clinical condition’’ [31]. Fig. 11 shows the ECG signal
of Atrial Fibrillation case and ECG waves of mother and
fetal studied by Lewis [2]. Six major categories of anoma-
lies were also coined by Lewis, namely: sinus arrhythmia,
heart block, premature contractions, proximal tachycardia,
auricular fibrillation and alteration of the pulse. The ECG
machine used by Lewis during 1930 to diagnose the patients
is shown in Fig. 12 [32]. In addition, he also explained the
terms such as sino auricular node, pacemaker, premature
contractions, proximal tachycardia and auricular fibrillation.
The contribution of Lewis’ research to bridge Einthoven’s
research was vital and can be understandable by Einthoven’s
statements after he was awarded the Nobel Prize in 1924.
In his Nobel lecture, Einthoven stated about Lewis, ‘I owe
you so much. Without your steady and excellent work to which
you have devoted a great part of your life there would have
been in all probability no question of a Nobel prize for me.
You have given to Medicine at least as much as I have’’ [33].
Lewis continued working as the world’s leading electrocar-
diographer while Einthoven investigated the theoretical bases
for Electrocardiography.
VOLUME 9, 2021 140857
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
FIGURE 6. First ECG tracings provided by Einthoven (Picture credit: [20]).
FIGURE 7. Einthoven’s triangle to obtain the relationship between lead I,
II and III (Picture credit: [27]).
The clinical features of Myocardial Infarction were first
published in 1910 by Russian Physicians W P. Obrastzow
and N. D. Straschesko. They reported two main findings:
FIGURE 8. Research categorization in the following years after EKG origin.
prolonged chest discomfort and persistent dyspnea7however,
these features were not based on ECG tracings. Further,
the electrocardiographic features for various diseases were
observed as listed below.
1917- Electrocardiographic features for acute myocar-
dial infarction were first published by Oppenheimer and
Marcus Rothschild [34].
1920- Harold Pardee reported ST-segment elevation in
Lead II, Lead III and T wave inversion features of
Myocardial Infarction [35].
1924-Woldemar Morbitz found out two different types
of second-order Atrioventricular (AV) blocks that are
named after him, known as Morbitz type I and type
2 blocks [36].
1930- WPW syndrome was described by scientists
Wolf, Parkinson and White named after the scientists in
which Bundle branch block with short PR interval was
discussed [37].
1931-1932- Charles Wolferth and Francis Wood
reported the electrocardiographic features for the
Angina Pectoris8after moderate exercise [38], [39].
1935- Sylvester McGinn and Paul D. White find
features for the cardiac condition, acute pulmonary
1939- Richard Langendorf obtained the ECG features
for Atrial Infarction [41].
1942- Arthur Master, Friedman Rudolph and Dack
Simon standardized the two step exercise test also
known as the Master two-step for cardiac function [42].
1944- Young and Koenig reported the PR Segment devi-
ations for Atrial Infarction condition [43]
The development of such Electrocardiographic features is
till date continued.
Developmental activities were also going to find out the
best materials for electrodes. Electrodes are connected to
7Shortness of breath or difficulty in breathing.
8Angina pectoris is the medical term for chest pain or discomfort due to
coronary heart disease. It occurs due to narrowing of arteries due to blockage
and also known as ischemia.
9It is a blockage of an artery in the lungs.
140858 VOLUME 9, 2021
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
FIGURE 9. Various features of an ECG signal based on P wave, QRS complex, T wave
and J point.
FIGURE 10. ECG tracings of experimentally induced heart block in a dog
by Einthoven to demonstrate role of ECG to determine cardiac anomalies
(Picture credit [30]).
various specific locations on the human body. Electrodes are
conductive pads that enable the recording of the electrical
activity of the human heart. An ‘ECG lead’ can be obtained by
analyzing the various electrode signals and considering dif-
ferent electrodes’ positions. It provides different viewpoints
to measure the heart’s electrical activity, and it is similar to
clicking a picture of the heart from different angles to get a
better understanding by the physicians.
Leads may be unipolar or bipolar in nature. In the unipolar
leads, the potential difference between any specific electrode
and ground terminal is considered. In Bipolar leads, the dif-
ference between two electrodes’ signal is considered with
reference to the ground terminal. Unipolar leads provide the
horizontal view of the heart, and bipolar leads provide the
heart’s frontal view.
FIGURE 11. ECG signals studied by Thomas Lewis (Picture credit: [2]).
In 1893, Einthoven first used the term EKG and studied
the graphs using the capillary electrometer. Later he built a
string galvanometer based on a 3-electrode EKG machine
VOLUME 9, 2021 140859
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
FIGURE 12. ECG Machine of Thomas Lewis for checking the patients in
the year 1930. Picture from medical exhibits in the science museum,
London (Picture credit: [32]).
TABLE 1. ECG and VCG leads and electrodes.
in 1902. In the year 1912, Einthoven mathematically reported
the Einthoven’s triangle [27]. This became the basis for future
EKG, Vectorcardiography (VCG) and development of Elec-
trodes and Leads for ECG acquisition (see Fig.13) that is
being used till date. Table 1shows the corresponding number
of electrodes and leads for ECG and VCG schemes.
In 1934, Frank Wilson defined an ‘indifferent electrode’
that was later known as ‘Wilson Central Terminal’ by con-
necting the right arm, left leg and left arm with resistances
typically 5K[44]. Wilson Central Terminal is an artifi-
cially constructed reference for electrocardiography, which is
assumed to be at zero potential and steady during the cardiac
cycle so that the reference point for unipolar potential remains
fixed. It worked as a ground terminal for other unipolar leads.
The events that occur during each heartbeat are termed a
cardiac cycle that can be divided into two parts: a period
of relaxation known as diastole and a period of contraction
known as systole. A cardiac cycle on an ECG signal is shown
in Fig. 14.
In 1938, the American Heart Association and the Cardiac
Society of Great Britain published their recommendation for
recording the exploring lead from six sites named V1 through
V6 across the precordium [44].
Later, Emanuel Goldberger extended the Wilson Central
Terminal with Augmented Unipolar Leads (avl,avr and avf)
also known as Goldberger leads for obtaining a detailed
view of the frontal plane [45]. Further, in 1953 the general
theory of heart vector projection was presented by Frank [46]
that provided a mathematical framework where three vec-
tors determined the person’s complete cardiac health that
confirmed the robustness of the methods used then with
a mathematical validation. In the following year, in 1954,
the American Heart Association published their recommen-
dation for standardization of 12-lead Electrocardiogram and
Vectorcardiogram [47]. Till date, 12 Lead ECG and 3-Lead
VCG continues to be the standard of ECG measurement
The theoretical background for Vectorcardiogram (VCG) was
developed by Burger [48] that the heart vector represents
the total cardiac electrical dipole strength and direction. This
concept was later utilized by Frank [46] to provide a complete
electrode configuration known as VCG. It had widespread
clinical use till 1960s but, by the year 1987, VCG had
been discontinued [49]. It may be attributed to diagnostic
performance of ECG was getting better with experienced
cardiologists [50]. However, the statistical programs for diag-
nosis performed better on the VCG signals. The use of VCG
was vanishing until the year 1987 when novel methods of
obtaining the VCG signals from the 12 Lead ECG signals
[51]–[53] were reported. A detailed study of VCG origin and
its significance in medical diagnosis can be found in [49].
Various attempts have been made to utilize VCG for the
disease diagnosis in recent years (2000-2021) [54]–[60].
In [54], a triggering system that uses the spatial information
of the VCG to minimize the effects of magnetic resonance
related noise and rejection of arrhythmic premature ventric-
ular depolarizations had also been demonstrated. In [55],
detection of various cardiac diseases had been discussed
based on VCG signals. Reconstruction of the equivalent heart
vector for the QRS complex from limb lead voltages and
the VCG parametric space derived from the frontal plane
VCG has been used to classify different ECG abnormalities.
In [56], authors compared the diagnostic utility of various
planar QRS-T angles (ECG signals) to the spatial QRS-T
angle (VCG signals) in detecting various cardiac anomalies.
Authors derived the VCG signals from ECG signals and con-
cluded that diagnostic accuracy of spatial QRS-T angles was
better for detecting coronary artery disease, hypertrophic car-
diomyopathy, or left ventricular systolic dysfunction. In [57],
authors discussed detection of cardiac ischemia based on
VCG signals with a neural network based classifier. Indepen-
dent Components Analysis and Principal Component Analy-
sis techniques are used for feature dimensionality reduction.
It was concluded that VCG is an efficient diagnostic tool for
detection of ischemia where ECG signals failed to categorize.
In [58], a typical VCG signal on the thorax and its reliability
of the detection of the PR-time are analyzed for detecting
various cardiac anomalies. In [59], [60], authors discussed
techniques to detect posterior myocardial infarction using
VCG signals as standard 12 Lead ECG signals do not provide
140860 VOLUME 9, 2021
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
FIGURE 13. Addition of various leads to ECG signals to get a complete 3-D View of heart and
development of VCG over the years.
FIGURE 14. Representation of a cardiac cycle.
this information. Weighted support vector machine [59] and
discriminative multiscale eigenfeatures [60] from the sta-
tionary wavelet transform subband matrices is utilized for
classification purpose.
VCG signals have lesser information content than the stan-
dard 12 Lead ECG signals [49]. However, it has shown to
be beneficial for detecting cardiac conditions by automated
statistical methods. In certain typical cardiac conditions, for
example, posterior myocardial infarction VCG signals pro-
vide more diagnostic features than standard 12 Lead ECG
signals that are inefficient. Still, 12 lead ECG signals are
significantly more diverse compared to VCG signals for
automated detection schemes. We can even utilize the same
electrode configuration to obtain VCG by utilizing various
conversion matrices [51]–[53]. It could be interesting to look
into various aspects of utilizing VCG signals efficiently.
ECG electrodes are the conductors to obtain the electrical
signals of heart activity and are connected to specific body
locations. During the invention of the String Galvanometer,
Einthoven utilized the saline water buckets as Electrodes.
Nowadays, Silver - Silver Chloride (Ag- AgCl) are primarily
used to obtain ECG signals. These electrodes are connected
with the electrolyte gel on the skin to increase conductivity,
hence known as wet electrodes. An electrolyte is usually
composed of a salt solution gel material. Ag- AgCl electrodes
are widely used in conventional schemes as these types of
electrodes provide a high signal to noise ratio, but there are
some disadvantages associated with these types of electrodes.
Some patients seem to be allergic to these gels, and the hairs
on the skin make it difficult to apply for some cases. Also,
the wet electrodes are not comfortable for long term ECG
monitoring [61].
Electrodes can be categorized as active and passive elec-
trodes. In the active electrodes, the pre-amplification module
is immediately after the conductive material between the skin
and the electrode and is present to enhance the signal-to-
noise ratio. The passive electrodes provide a direct connection
between the metal layer and the processing unit. Various
other types of electrodes are also reported in the litera-
ture and are compared with the conventional wet Ag-AgCl
VOLUME 9, 2021 140861
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
electrodes [62]–[73]. A list of other types of electrodes
recently reported in the literature is given- below:
Gel less ECG Electrode [62]: This work discussed the grip-
style dry electrodes for ECG measurement during physical
activity and presented an innovative design for a portable
ECG amplifier that mitigates some of the pre-identified issues
of electrodes.
Capacitive Electrodes for noncontact ECG monitoring
[63], [64]: The noncontact capacitive electrodes can obtain
ECG signals through clothes and implemented with the real-
time denoising algorithm.
Carbon Nanotube (CNT)/Polydimethylsiloxane (PDMS)
composite-based dry electrodes [65]: In this work, the CNT/
PDMS composite-based dry ECG electrodes were readily
connected to the conventional ECG devices, and showed its
long-term wearable monitoring capability and robustness to
motion and sweat.
Esophageal Electrodes for long term monitoring based on
Titanium Nitride (TiN) and iridium oxide (IrOx) [66]: This
work discussed the advantages of esophageal electrodes over
wet and dry electrodes for long term monitoring without the
need of electrolyte gels. The TiN and IrOx are identified as
suitable materials for esophageal electrodes that are superior
to the standardized surface skin-electrode concerning signal
distortions, and thus, it might help prolong conventional ECG
recordings maintaining high-quality signals.
Underwater Electrodes based on Carbon Black Pow-
der (CB) and Polydimethylsiloxane (PDMS) [67]: In this
work, hydrophobic electrodes that provide all morphological
waveforms without distortion of an ECG signal for both dry
and water-immersed conditions was discussed.
Dry Metal Electrodes [68]: The wearable multi-lead elec-
trocardiogram (ECG) recorder suggested that dry metal elec-
trodes provide a comfortable sensation, less skin-irritating,
easy clean surfaces, reusable capability and more durability
compared to conventional ECG electrodes.
Dry textile-based electrodes, needle array electrodes and
silver-coated surface electrodes [69]: The three types of dry
electrodes, viz. textile electrodes, needle array electrodes and
Silver coated surface electrodes, were fabricated and tested
for acquisition of ECG signals. The dry electrodes were
fabricated as active electrodes and recording from the dry
electrodes are compared to that of the wet electrodes.
Carbon Based Electrodes for Wearable Applications [70]:
The flexible dry electrodes for long-term biosignal monitor-
ing were designed by mixing carbon nanofibers (CNFs) in
biocompatible-elastomer (MED6015).
Silver Nanowire based Dry Electrode [71]: The silver
nanowire (AgNW)-based dry electrodes were fabricated for
noninvasive and wearable ECG sensing.
Garment type electrode [72]: The multi-channel telemeter
and garment-type electrodes were developed that exhibited a
sufficient R-wave detection rate in four positions
Poly(3,4-ethylenedioxythiophene) Polystyrene Sulfonate
(PEDOT:PSS) and PDMS coated cotton fabric electrode [73]:
A flexible electroconductive textile material was developed
by coating PEDOT: PSS/PDMS on cotton fabric via flat
screen printing. The coated fabric was utilized as ECG elec-
trodes and compared with the conventional electrodes.
ECG data compression techniques reduce the computational
cost for any system by removing redundant information
and retaining essential parameters of the signal. Efficient
compression techniques significantly reduce the storage and
transmission requirement for any portable system and can
optimize its performance [74].
ECG compression can be classified into lossy, lossless,
direct, transformation-based, prediction-based, 1D and 2D
techniques. Lossy techniques are capable of achieving higher
compression ratios at the cost of reconstruction errors. On the
other hand, lossless compression offers lower compression
ratios but allows for nearly perfect signal reconstruction.
In biomedical signal processing, lossless compression tech-
niques are preferred over the lossy ones [74] because, for
clinical applications, loss of information could prove fatal.
Additionally, compression techniques can also be classi-
fied into direct and transformation based approaches. In direct
compression, the time domain signals are compressed, while
in the transformation methods, the signal is converted to
the frequency domain using Fourier transform or the time-
frequency domain using WT. Various Prediction based
models [75], [76] are reported in the literature that pro-
vide comparatively higher compression ratios. Various 2-D
ECG compression techniques [77]–[79] are also discussed for
achieving the optimum results.
Various compression techniques are developed in the last
few years [82]–[89]. In [82], an ECG compression tech-
nique using unsupervised dictionary learning titled CULT
was reported. The algorithm expanded its dictionary upon
the arrival of any unseen pattern and used discrete cosine
transformation to make it immune to incoming noise. In [83],
a sparse encoding algorithm consisting of two subcategories
based on geometry-based methods and WT based itera-
tive thresholding was reported. In [84] an energy-efficient
novel block-sparsity-based multichannel ECG compression
scheme that utilizes spatiotemporal correlation and multi-
scale information of the signal using wavelet transform of
the signal was reported. In [85], a deep learning technique
based on a convolutional auto-encoder is applied to achieve
ECG signal compression without any independent encod-
ing method. In [86], a linear method based on the sparsity
of the ECG signal and compressed sensing was used to
achieve compression in real time. In the recovery phase,
authors utilized an efficient method known as the Kronecker
technique. The system implementation was based on full-
adder/subtractor (FAS) and shift registers, without using any
external processor or training algorithm. References [87] dis-
cusses a lossy compression algorithm based on fast wavelet
transformation that provided insignificant delay for the com-
pression at low level distortion of the signal. In [88], empir-
ical mode decomposition and wavelet transformation to
140862 VOLUME 9, 2021
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
compress the ECG signal was discussed. References [89]
utilize the 2D Discrete Cosine Transform coefficient and
iterative JPEG2000 encoding for compression purposes that
proved efficient.
Interesting readers can find a detailed review of lossless
ECG compression techniques in [80] and wavelet based ECG
compression techniques in [81].
Other streams of cardiac health monitoring that are also of
interest to researchers are PhonoCardioGraphy (PCG), Bal-
listoCardioGraphy (BCG), ApexCardioGraphy (ACG), Seis-
moCardioGraphy (SCG) and KinetoCardioGraphy (KCG).
Phonocardiogram is a plot of high fidelity recording of the
sounds made by the heart. Monitoring and recording equip-
ment for PCG was developed during 1930- 1940s and was
standardized around 1950 [90]. PCG originated in an attempt
to time the occurrence of heart sounds in a cardiac cycle. The
acquisition system of PCG, similar to ECG, consists of low
noise amplifiers and filters. As this review focuses only on
the ECG and related signal processing schemes, the interested
reader may refer to more details regarding PCG in [90]–[92].
Ballistocardiogram measures the Ballistic forces (Mechani-
cal Forces) generated by the heart and is a noninvasive method
based on the measurement of the body motion generated by
the blood’s ejection at each cardiac cycle [93]. Apexcardio-
gram was first described by Marey [94]. The curves of the
apexcardiogram display all consecutive phases of the cardiac
cycle; contraction-and-emptying and relaxation-and-filling.
The apex cardiogram’s waveform is caused primarily by
movements of the left ventricle against the chest wall. Thus,
it is a translation of the sequence of hemodynamic events
occurring in the underlying left ventricle [95]. Seismocardio-
gram (SCG) is the recording of body vibrations induced by
the heartbeat. SCG contains information on cardiac mechan-
ics, in particular, heart sounds and cardiac outputs [96]. Kine-
tocardiography (KCG) records indicate movements as the
result of the motions of the heart and utilize only the low-
frequency motions (0–30 Hz) [97].
Though these methods by themselves are interesting, none
evolved to match the process of ECG methods.
YEARS [60’S–90’S]
By 1961, Norman Jeff Holter had designed the Holter [29]
for continuous ECG monitoring in hospitals. Holter designed
a backpack recorder that weighed 75 lbs and it was able to
record and transmit ECG signals to the hospitals for further
evaluation. It became a landmark invention for the automatic
detection and transmission of biomedical signals, specifically
ECG signals.
A new era had ushered for automated computerized
detection. The first automated classification on 20 clinically
normal individuals on magnetic tape recorders to store
a 1-minute recording of each subject [98] were utilized.
In 1965, average transient computing based on average
response computing techniques was presented to extract the
ECG signal from the noisy records such as exercise stress
test results in [99]. In 1966, the system was developed for
online computer monitoring of critically ill patients [100].
This system provided continuous monitoring for various
health parameters such as ECG, systolic and diastolic blood
pressure, pulse rate, temperature readings at various parts
of the body, manual inputs from the user, etc., within the
hospital setting due to the high form factor of the system.
The system was used at the California Shock Research Unit
for clinical management of seriously ill patients. The sys-
tem provided online acquisition, processing and display of
the data. In 1967, Vector Cardio Graph (VCG) was used to
separate normal and Left Ventricular Hypertrophy (LVH) on
subjects [101]. Four different techniques were utilized for cat-
egorization, such as sum of amplitude measurement, vector
differences, weighted vector differences and class separating
differences. Two hundred subjects’ samples were utilized for
obtaining the results out of which 100 were LVH Samples and
100 were normal subjects (case-control study). In a similar
year, another work [102] was presented to analyze the normal
subjects’ frontal leads to expand the conventional amplitude
and time base factor.
By 1968 advanced mathematical concepts from signals
and systems viz Contour Analysis were used to catego-
rize the normal and abnormal ECG Rhythms [103]. Around
2000 samples were subjected to an automated contour plot
and the results were compared with the physician’s results to
determine the accuracy of the technique.
By 1970, the online Real Time algorithms for ECG wave-
forms were started to developed [104], that provided the
information whenever the values were exceeded beyond the
predefined limits for a single lead ECG signal.
In 1971, both 12 lead ECG and 3 Lead VCG tech-
niques were considered standard methods of monitoring and
around 1100 patients for the automated detections were
obtained [105], [106]. This study suggested no appreciable
differences from automated techniques obtained from 3 Lead
VCG and 12 Lead ECG data inputs. It also concluded that
3 Lead VCG data was sufficient for automated detections as it
saved the computing time and emphasized VCG as preferable
for automated detections. Although VCG became famous for
automated detections, it never became popular for physicians
due to the nonstandardized lead configuration, and the other
reason is medical doctors are accustomed to using 12 lead
ECG in clinical applications [107]. VCG can be obtained by
12 lead ECG signals but this field also requires more attention
from the researchers.
In 1973, a research based on clustering techniques [108]
for pattern recognition. The categorization was based on the
clusters defined by the human operators. If the values fall
outside the boundary limits of clusters the human invention
was requested. The system was semiautomated and classified
the QRS complexes of the ECG signal. In 1974, another
research work [109] was published for QRS and premature
VOLUME 9, 2021 140863
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
FIGURE 15. Milestones of ECG signal processing in the initial years.
ventricular beat detection for a continuous real time ECG
signal. However, the authors were not convinced of its
widespread use and specified that it was limited for specific
research purposes.
Around the similar year, µprocessors were made available
for automatic analysis in biomedical engineering. The first
reported use of µprocessor for Ambulatory ECG monitor-
ing is believed to be around 1976 [110], [111]. These were
µprocessors powered bedside ECG monitors and became an
important milestone in automated ECG methods due to trans-
formation in the research methodology. Various papers were
reported [112]–[117] during the same period that provided the
µprocessors based systems for Ambulatory ECG Monitoring.
Until 1980’s, all these automated detection algorithms and
methods were based on separate databases. So, there was a
need for a standard database as the results were invariably
data-dependent. Literature [118]–[120] suggests that various
attempts to obtain the Gold Standard database for comparison
begun. Around 1983, MIT and Beth Israel Hospital Arrhyth-
mia Laboratory released the data obtained from Holter tapes
patients between 1975 and 1979 [119]. It later became a stan-
dard database and is used till date. The databases [121]–[141]
available on Physiobank for automated processing schemes
comparisons are shown in Table 2. PhysioBank is a large
and growing archive of well-characterized digital recordings
of physiologic signals and related data for the biomedical
research community. In this paper, only ECG and relevant
signals databases are considered. The databases provide the
annotations for quantification of ECG waves (Truth values for
P, QRS and T waves to be compared with automated gener-
ated results) and the diagnosis of various diseases confirmed
by cardiologists. Except this, some of the databases also
provide information for compression tests and signal to noise
ratio information of the signal. These standard databases
continue to be the gold standard of ECG processing research.
Research related to Holter Tapes’ automated analysis with
µcomputers and Automated Holter Scanning were published
during 1983 [142]. The system consisted of two µcomputers
to detect QRS durations for arrhythmias of 24 hours recorded
on Holter tapes. It determined the heart rate variability and
PVC counts, a method used till date. Around the same year,
another method of QRS complex detection was presented
in [143]. In this method, the QRS complex was represented
by a single positive pulse along with onset and end of it, by a
dynamic threshold technique that utilized the time domain
features. But, the results of the method was provided on the
simulated ECG data as a software based technique.
A portable µcomputer based Arrhythmia Monitor was
designed [117] for storing 16 seconds arrhythmia intervals.
The major difference of this system with Holter tapes was
that it did not store any normal rhythm data and was advanta-
geous in terms of memory utilization. The system was able
to provide continuous and long term monitoring for high-
risk patients. In 1987, filter design was illustrated in [144]
for biomedical signal processing techniques. The filters were
implemented for the ECG signals and quantization of filter
coefficients was used to design various filters with its imple-
mentation on 8 bit µprocessor.
These µprocessor-based systems became the interest of
researchers with databases available to them and the evolution
of data compression techniques. The processing of various
data compression techniques also evolved rapidly as pro-
cessing the 12 Lead ECG data for automated methods was
computationally expensive. The compression techniques fur-
ther minimized redundant information present in the original
signal and helped the system practically feasible. These meth-
ods are believed to have early beginnings during 1968 and
continued in later years (1984- 1992) as cited in various
works [145]–[149].
Various data compression algorithms as Turning Point
Algorithms [115], FAN algorithm [150], AZTEC Algo-
rithm [148], CORTES algorithm [146], Fast Walsh Trans-
form [151] and SLOPE algorithm [152] are discussed. One
of the main concerns in biomedical data reconstruction was
the clinical acceptability of these signals. During the data
compression, the requirement for lossless information and the
140864 VOLUME 9, 2021
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
TABLE 2. Some of the ECG databases available on Physionet website.
removal of repeated or redundant signals proved a challenge.
The data reduction techniques provided viable options for
storing or processing large amounts of data with lower storage
requirements and became successful.
Following this, Pan and Tompkins proposed the seminal
algorithm for QRS detection [153] for normal and abnor-
mal waveforms. This algorithm provided accuracy of more
than 99% for QRS detection and revolutionized the means
for arrhythmia monitoring. The algorithm also provided the
ideal means for heart rate variability measurements in real-
time processing and reporting various cardiac conditions
and diseases. In 1987, another research [154] provided the
preliminary heart rate variability (HRV) analysis by using
the autoregressive modeling techniques and power spectral
density estimates. For the QRS detection, it followed the clas-
sical technique by obtaining the derivative of the ECG signal
followed by adaptive thresholding. After obtaining the R-R
interval information, it discriminated the normal and patho-
logical subjects by utilizing the autoregressive modeling and
power spectral density estimates. In 1988, two methods for
detecting the QRS complexes were discussed in [155] based
on the length transformation and energy transformation of the
signal. In both the methods QRS complexes of the signals
were enhanced and other components of the signal were sup-
pressed significantly and detection accuracy for QRS com-
plexes was found out be over 99%.
Around 1989, research on connectionist systems, better
known as neural networks for diagnostic purposes, was
proposed. Neural networks were first used for ECG sig-
nal processing during 1990 for diagnostics [156]–[162],
categorization and QRS detections and proved interesting.
The application of neural networks also proved to be
VOLUME 9, 2021 140865
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
advantageous in classifications and detections with extended
computations. Over the years, such artificial intelligence
algorithms were extended towards categorizing normal and
abnormal waveforms and pattern matching. During 1992,
detection of QRS complexes for very noisy signals was
demonstrated using neural networks [160]. This work used a
multilayer perceptron neural network as an adaptive whiten-
ing filter instead of a typical linear filter.
Another work on the QRS template matching was updated
by ANN recognition algorithm was discussed in [161].
Several hidden layers in multilayer perceptron with eigen-
value decomposition method to classify the signals avail-
able in MIT/BIH database were provided in [162] and the
technique was also patient adaptable. Although the neural
networks provided better detection accuracy than the con-
ventional classification (based on thresholding and empirical
values), the computations requirements for such systems
were high and often difficult to realize on customized
In [163], authors discussed the utilization of Wavelet
Transform (WT) for ECG analysis and compression tech-
niques. The research presented a preliminary investigation
into its application to the study of both ECG and heart
rate variability data. Further, WTs were also studied that
provided time and frequency analysis for the ECG signals
discussed in [164]–[168]. The authors suggested that wavelet
transforms’ efficiency measures were comparatively higher
than conventional methods [164], [165]. The timeline for the
signal processing is shown in Fig. 15.
ECG signal processing is classified into four major subcat-
egories: acquisition of ECG signal, preprocessing, feature
points selection and classifier selection. Fig. 16 shows various
steps of standard ECG signal processing steps.
The ECG signal is obtained through publicly available
databases or various ECG acquisition methods. The number
of leads for ECG data can be chosen according to the appli-
cation of the system. So, the ECG input to the system may be
a single lead ECG signal, Bipolar three lead signal, standard
12 Lead ECG signal or the 3 Lead VCG signal.
The preprocessing step is the initial step before the feature
selection process. The preprocessing stages usually denoise
the ECG signal affected by various kinds of noises such as
baseline wandering, power line interference, electromyogram
noise etc. [169]. In some of these cases, filters are employed
at this level to consider a particular band of frequencies.
The introduction of the preprocessing stage before feature
selection leads to more accurate results. However, we found
out that the signal quality assurance before the preprocessing
stage is mostly missing from most of the researches. Various
noise removal or preprocessing techniques for the ECG signal
are discussed in [170], interested readers may have a look at it.
In the subsequent step, the signal features, namely tem-
poral, spectral or time-frequency are selected. The original
ECG signal is in the time domain, and for converting it to
other domains, various transforms such as Fourier transform
or wavelet transforms are needed. In some of the cases,
statistical information regarding the signal is also considered.
After the feature selection, classifiers are implemented to
categorize the signals. These classifiers may be empirical,
thresholding, machine learning or a deep neural network.
Based on the results of classifiers, detection of the disease
or diagnosis is usually done.
In literature, time domain [171]–[182], frequency domain
methods [183]–[196] or time frequency domain meth-
ods [197]–[213] are classified with the empirical, thresh-
olding based, machine learning or deep neural network
approaches. In this report, the papers are classified according
to feature point selection or the classifiers scheme discussed
in the work.
The signal processing schemes can also be classified into
hardware based, software based or hardware-software based
approaches. In the software processing schemes, the compu-
tational load does not pose a constraint on the system. There-
fore, complex processing schemes are often used to obtain
results that provide comparatively more accurate results.
However, in the Real-Time system for resource-constrained
regions, the computational load does matter and poses a
real challenge for the battery-driven systems, deployment
and relevance. Hence, for automated detections, there must
always be a tradeoff between performance and resource
requirements. The performance matrices of the system is
defined in terms of True Positives (TP), False Positives (FP),
True Negatives (TN), False Negatives (FN), Accuracy(Acc),
False Detection Rate (FDR), Sensitivity (Se), Specificity
(Sp), Positive Predictivity (PPV), Error Rate (ER), Efficiency
of Recognition (EOR), Classification Rate (CR), Classifica-
tion Error Rate (CER), False Acceptance Rate (FAR), False
Rejection Rate (FRR) and Detection Error Rate (DER) etc.
Various research works are classified in the following sub-
sections based on their feature point selection strategy or the
classification scheme.
Feature point selection is a crucial stage as it requires the
tradeoff between the system complexity, accuracy and bat-
tery requirements for the system. Processing techniques may
vary from the simple morphological features capturing to
complex transformations. Features may be from the time
domain, frequency domain, or a mix of both as in the time-
frequency domain. Researches related to the field are listed
and compared in the following subsections.
The time domain processing or the temporal domain feature
extraction allows the processing in discrete samples. With
the evolution of µprocessors around 1974, the conversion
of continuous signals to discrete ones became usual, so the
signals were also usable by µprocessors. It was easier to
convert the discrete signals again back to analog [145].
140866 VOLUME 9, 2021
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
FIGURE 16. Recent ECG signal processing trends categorized into ECG signal acquisition,
preprocessing, feature point selection and classifiers sections.
TABLE 3. Literature comparison based on time domain features of
ECG signals.
Table 3shows the literature based on various time domain
methods, in [171], authors discussed two algorithms for QRS
detection; the first algorithm detects the current beat while
the second algorithm has an RR interval analysis component.
Authors in [172] discussed noise filtering, QRS detection,
Wave Delineation and Data Compression of the ECG signal
in time domain.
Authors utilized the difference operation method for QRS
complex detection in two stages in [173]. The first stage was
to find the R point by applying the difference equation to an
ECG signal and the second stage looked at Q and S points
based on the R point to find the QRS complex. By doing
so, Q and S point detection efficiency was dependent on the
R peak detection algorithm. In [174], authors utilized the
squared double-difference signal of ECG signals to detect
the R peaks and the results are provided only for the normal
cases. In [175], authors utilized a time-domain morphol-
ogy and gradient-based approach based on a combination
of extrema detection and slope information, using adaptive
thresholding for ECG features extraction. The main limitation
of this algorithm was that its robustness not tested against
baseline wandering variations. In [176], [177], authors deter-
mined the P and T waves of ECG signal using two mov-
ing average filters that provided the dynamic event-related
thresholding by utilizing the signal’s QRS information. Then
the results were compared against the annotations pro-
vided by the cardiologists on the MITDB database. Further,
in [178]–[180] authors extracted the ECG features and clas-
sified various ECG anomalies based on the 12 Lead ECG
signal. The advantage of this algorithm was forward search
processing; it lowers the memory requirement in real-time
processing. Similarly, in [181], [182] authors detected various
ECG features required for the classification of ECG anoma-
lies. The results of real time system were not reported in the
VOLUME 9, 2021 140867
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
TABLE 4. Literature comparison based on frequency domain features of ECG signals.
In conclusion, temporal features such as RR, PR, QT inter-
val, ST deviations were utilized to provide the diagnos-
tic results. Various filtering methods such as differentiator,
moving average filter, low pass, high pass etc. have been
utilized to obtain ECG signals’ feature points. The time
domain methods are of prime importance as the cardiol-
ogist detects the heart’s anomalies using 12- Lead ECG
signal in the time domain that provides synergistic efforts
between doctors and engineers towards automated detection
Research on the frequency domain processing of ECG sig-
nals started around the 1970s. During the initial years,
the researchers focused on obtaining the frequency band
amplitude information related to ECG signals such as the
band of QRS complex, P wave and T wave [183], [184] etc.
The features required for diagnosis are primarily temporal
and statistical information, as developed by cardiologists.
Hence, ECG signal processing based on only spectral infor-
mation is rare compared to the time domain and time-
frequency methods. The frequency domain methods are
shown in Table 4. As can be seen in [185], [187], [191]–[193]
research, results were not available in terms of standard
performance matrices, which makes it quite difficult for
comparison. In [186], the ECG signals were classified in
different categories namely ventricular fibrillation and flutter
(VF), Artifacts (A), series of complexes of aberrant mor-
phology (CAM) and one unknown category by utilizing the
spectral features of the ECG signal. However, the ECG data
used for the purpose consisted of only 55 ECG signals from
an unknown database. In [188] authors detected the QRS,
P and T waves of the ECG signals by application of discrete
Fourier transform on the ECG signals, acquired locally and
from MITDB. The delineated components were evaluated
visually and by computing the normalized mean square error
between the original and recreated signal. In [189] authors
detected the duration of ventricular fibrillation in swine and
humans using the frequency domain features of ECG signal.
In [190] determination of countershock (Medical Procedure)
success was done by obtaining the Fourier transform of ECG
signal and it was concluded that the median frequency, dom-
inant frequency and amplitude of the signal could predict the
success or failure of the procedure.
In [194], authors determined the power line interference,
a common source of noise in the ECG signal that leads to
imprecise measurements of the ECG wave durations and
amplitudes. In [195] authors determined the heart rate vari-
ability using spectral analysis. However, the results were
not validated on any standard databases. In [196], authors
detected anomalies of ECG signal by utilizing the frequency
domain features and the results were compared with the
temporal methods.
Frequency domain features are essential to obtain the com-
plete analysis for the signal. However, the disease diagnosis
requires the utilization of temporal features as most of the fea-
tures for cardiac anomalies were developed by cardiologists
and physicians accustomed to time domain signals.
140868 VOLUME 9, 2021
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
TABLE 5. Literature comparison based on time-frequency domain features of ECG signals.
Time-frequency domain methods utilize the time as well
as frequency space for obtaining the features simultane-
ously. The most popular technique in this domain is Wavelet
Transform (WT) and the ECG signal processing using the
same is discussed in [197]–[214]. The advantage of using
WT over Fourier transform is that Fourier Transform of
a signal provides the information regarding the frequency
and its magnitude; however, it cannot provide frequency
information for a localized signal in time. To overcome
the poor time resolution of the Fourier transform, Short
Time Fourier Transform (STFT) has been developed. It pro-
vides the time-frequency representation of the signal. In the
Wavelet Transform signal’s different frequency components
are analyzed at different time resolutions, also known as
multiresolution analysis [215]. The multiresolution analysis
capability of WT makes it suitable for biomedical signal
processing schemes [214]. Various researches for WT and
other time-frequency schemes are given in Table 5. Denois-
ing techniques for an ECG signal using new wavelet- and
wavelet packet-based schemes were discussed with simu-
lated noises in [216]. Various WT techniques such as Cross
Wavelet Transform (XWT) [205], Continuous Wavelet Trans-
form (CWT) [197] and Discrete Wavelet Transform (DWT)
[199]–[203], [205], [207], [210]–[212] are employed in
In [197], authors report the detection of the arrhythmias
in the ECG signals of pigs. WT was utilized for the com-
pression of ECG signal in [198]–[200]. The main reason
for utilizing WT for the compression is because it trans-
forms the signal’s energy into fewer transform coefficients.
So, most of the transform coefficients with lower energy can
be discarded [217]. It is an efficient and flexible scheme
for compression [218]. Authors in [201], [202], [207], [209]
detected various features such as R peaks, QRS complex
using WT. In [203]–[205], [208], [210]–[212] authors clas-
sified the ECG signals by utilizing the wavelet coefficients
as features. In [206] two algorithms were presented based on
wavelets for real-time ECG signal compression. The authors
achieved the correct diagnosis (CD) values up to 100% for
VOLUME 9, 2021 140869
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
various compression ratios. In [213], authors have discussed
a method for ECG signal classification and analysis that
utilizes wavelet based information as the input. This method
followed a three-stage procedure for analyzing ECG signals
starting with noise suppression and then wavelet-based fea-
ture extraction and classification stages.
The Wavelet Transform has certain disadvantages as
it becomes computationally intensive for finer resolution.
Discrete Wavelet Transform (DWT) offers fast computations
due to the discretization of wavelets as the minimum energy
transform coefficients are discarded at the cost of efficiency
of the system [215].
The time domain, frequency domain and time-frequency
domain methods are used for various specific applications.
The temporal or time domain features such as R-R interval,
QT interval, ST-T interval and PR duration are of prime
importance because the cardiologists are accustomed to such
features from standard 12 Lead ECG signal. However, apply-
ing preprocessing methods on the ECG signals for automated
detections may alter certain features such as ST segment
elevation and depression of the ECG signal that could lead
to false outcomes. Hence, the time domain methods require
careful preprocessing to improve the accuracy.
The frequency domain features such as power spectral den-
sity and frequency bands for a particular section of the signal
helps in implementing the filters for delineation. However,
the classification that is based on frequency domain features
is infrequent and requires more attention.
WT is the most common and widely used for the time-
frequency methods. References [197]–[213] discuss auto-
mated classification and compression methods based on WT.
Continuous WT based methods are computationally expen-
sive for portable systems. Preferred discrete WT offers faster
computations at the cost of efficiency.
It is to be noted that the system is dependent on the clas-
sifier following the feature point selection. Therefore, it is
difficult to compare these methods based on feature selection
methods only.
The classification stage considers the extracted feature points
from the previous stage and classifies the signal into dif-
ferent categories by using empirical, adaptive and constant
threshold-based, machine learning based and deep neural
networks based classifiers. The conventional empirical classi-
fiers are generally based on the medical observations for the
particular field. Thresholding based or decision logic-based
approach is based on defined logical rules, e.g. R-R interval,
ST interval etc.
The machine learning based approaches based on mul-
tivariate statistical pattern recognition have a widespread
utilization in biomedical signal processing. These methods
utilize correlation analysis, regression techniques and tem-
plate matching to identify abnormal patterns or a particular
class of signals [219], [220]. However, as these statistical
methods move towards greater accuracy, the computational
cost for the system also increases. The most popular tech-
niques are deep neural networks also known as ANN con-
sists of multiple hidden layers between the input and output
layers [221]. Each layer consists of neurons with different
weights and biases. The neurons can pass the information to
other neurons in other layers. The backpropagation technique
provides feedback and updates the weight associated with
neurons offering supervised and unsupervised learning. The
deep learning technique offers more accuracy to the system
at the cost of increased system complexity, which may be a
serious challenge for battery operated portable systems.
Support Vector Machine (SVM) is also widely used for
classification of different types of signals. It provides a super-
vised learning method to optimize the gap between two dif-
ferent categories of training sets. Another type of classifier
is Convolutional Neural Networks (CNN) is a particular type
of feed-forward neural network. It represents the input data
in the form of multidimensional arrays and consists of three
layers: convolution, pooling, and fully connected.
Recent developments in deep neural networks are
widespread, with the latest techniques discussed in [221]
are Recurrent Neural Networks (RNN), Convolution Neu-
ral Networks (CNN) and other generative models such as
Autoencoders and Generative Adversarial Network (GAN).
In the following subsection, we have selected the literature
mainly focusing on classification schemes based on machine
learning and neural network approaches for ECG signal
processing. The reason for selecting only the classifier based
on machine learning of deep learning approaches is the
widespread utilization of these techniques compared to the
other techniques.
This section categorizes (shown in Table 6and Table 7) vari-
ous researches based on machine learning and deep learning
techniques for classification.
In [222], authors provided the customized ECG classifier
with patient-specific data based on an unsupervised learning
technique. The method’s limitation was that it required the
development of a local classifier for each patient with patient-
specific data. In [223], authors detected the QRS complexes
of 12 Lead ECG signals available in CSE dataset -3 with
supervised learning of ANN. The backpropagation algorithm
has been used to train the system and to update the weight
and biases of neural networks.
Authors in [224], utilized the ANN for arrhythmia clas-
sification, ischemia detection, and recognition of chronic
myocardial diseases. It used both static and recurrent ANN
with preprocessing and postprocessing that defined the
dimensions of input features for neural networks.
140870 VOLUME 9, 2021
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
TABLE 6. Literature comparison for machine learning and deep learning techniques based ECG classification.
Authors in [225] utilized an unsupervised learning clus-
tering scheme for the classification using Hermite functions
based features of QRS complexes. The limitation was that it
did not provide signal quality information in the input vector’s
self-organizing maps. Authors in [226], used a beat recogni-
tion and classifier based on a supervised learning scheme that
VOLUME 9, 2021 140871
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
TABLE 7. Literature comparison for machine learning and deep learning techniques based ECG classification in last 2-3 years.
utilized fuzzy hybrid neural network and higher-order statis-
tics features as inputs. In [227], authors utilized Hermite basis
function expansion of the QRS complexes of ECG waveforms
and modified Takagi-Sugeno-Kang neuro-fuzzy network for
heartbeat recognition and classification based on a supervised
learning scheme. In [228], authors utilized a popular super-
vised machine learning approach known as Support Vector
Machine (SVM) for the recognition purpose. The input fea-
tures in the method were obtained by two methods, namely
higher-order statistics (HOS) and Hermite characterization of
the QRS complex. In [229], authors classified the ECG data
into three categories, namely normal beat, ventricular ectopic
beat (VEB), supraventricular ectopic beat (SVEB). The clas-
sification was based on a statistical classifier model utilizing a
supervised learning scheme. The limitation of the technique
was that heartbeat fiducial points were manually annotated.
In [230], authors provided supervised learning based on a
decision tree based classifier algorithm to be implemented
on a personal digital assistant (PDA). But, the algorithm was
not implemented in a real time environment as the PDA was
meant only for a system demonstration. In [231], authors used
features such as ST segment area, R-S interval, ST-slope,
R-T interval, QRS area, Q-T interval, R-wave amplitude,
heart beat rate and four statistical features QRS energy, mean
of the power spectral density, auto-correlation coefficient and
signal histogram were applied to signal stage and two stage
feed forward neural networks for the anomalies detection.
In [232], authors used supervised learning that required
block based neural networks as classifiers. It utilizes Hermite
coefficients and R-R intervals as input features to classify
Authors in [233] utilized the supervised particle swarm
optimization (PSO) with the SVM classifier on the auto-
matically detected features. In [234], authors compressed
the ECG signal using local extreme extraction, adaptive
hysteretic filtering and Lempel-Ziv-Welch (LZW) coding.
The reconstructed waveform was verified with frequent nor-
mal and pathological cardiac beats using a multilayer per-
ceptron neural network trained with original cardiac patterns
and tested with reconstructed ones. In [235], authors did the
screen apnea screening using the time domain and frequency
domain features. It used two approaches, namely K-nearest
neighbor (KNN) (clustering or unsupervised technique) and
neural networks that offer the supervised learning scheme.
The limitation of the method was that it was unable to detect
isolated apneas and other physiological and pathological
events during sleep, such as cyclic alternating pattern and
periodic leg movements that could affect the classifier’s effi-
cacy. In [236] authors used local fractal dimension (LFD) of
neighboring sample points of ECG signal segments are used
as the features. Two different methods used for estimating the
LFD, namely power spectral density based fractal dimension
estimator (PSDFE) and variance based fractal dimension esti-
mator (VFE). In [237], authors obtained the ECG features
set with sequential forward floating search algorithm based
on linear discriminants. The most suitable subset was again
evaluated with a multilayer perceptron Neural network as a
supervised learning scheme. The selection of suitable features
led to complexity reduction for the system. In [238], mobile-
cloud based ECG monitoring scheme was compared with the
mobile-based systems. The system utilized the supervised
ANN for classification. In [239], Particle Swarm Optimiza-
tion (PSO) based wavelets are applied to the SVM for cate-
gorizing various ECG signals. In [240], three neural network
classifiers, Back Propagation Network (BPN), Feed Forward
Network (FFN) and Multilayered Perceptron (MLP) were uti-
lized for ECG anomalies detection. In [241], authors classi-
fied the ECG signals into SVEB and VEB with the supervised
1-D Convolutional Neural networks and the patient-specific
data. The method’s limitation was that the dedicated CNN
was trained for an individual patient and often posed a
challenge in the Real-time environment. In [242], authors
140872 VOLUME 9, 2021
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
designed a personalized heartbeat classification model for
long ECG Signals and implemented the system with parallel
general regression neural network (GRNN). Further to this,
they implemented an online-learning module into the parallel
GRNN for the real-time personalized automatic classifica-
tion on the Holter ECG data. In [243], authors implemented
various machine learning schemes such as end-to-end CNN,
KNN, linear SVM, Gaussian kernel SVM, and Multilayer
perceptron classifiers for categorizing the paroxysmal atrial
fibrillation cases and concluded that integration of convolu-
tion neural network as a feature extractor with other conven-
tional neural network-based classification methods provided
better results. In [244], bidirectional long-short term memory
networks based wavelet sequences called DBLSTM were
used for categorizing various arrhythmic signals such as Nor-
mal Sinus Rhythm (NSR), Ventricular Premature Contraction
(VPC), Paced Beat (PB), Left Bundle Branch Block (LBBB),
and Right Bundle Branch Block (RBBB) available in the
MITDB Database. However, the work did not validate the
results under the noisy input conditions. In [245], authors
used KNN and Principal Component Analysis (PCA) clas-
sifier techniques with autoregressive modeling schemes to
categorize Atrial Tachycardia, Premature Atrial Contractions
and Sinus Arrhythmia.
In [246], authors converted the time domain ECG signals
into time-frequency domain spectrogram by utilizing Short-
time Fourier transform and the spectrogram was utilized as
the input to 2-D and 1-D convolution neural network for the
arrhythmia classification on the MITDB database. In [247],
authors integrated a long short-term memory based auto-
encoder (LTSM-AE) network for features learning with an
SVM for ECG arrhythmias classification and followed a
supervised learning scheme. In [248], authors developed a
system to categorize the severity stages of Myocardial Infarc-
tion condition and utilized an attention-based recurrent neural
network for automated diagnosis of the three MI severity
stages by processing the 12 Lead ECG data. The work’s
limitation was that it addressed only the classification of MI
severity stages, not the diagnosis of the conditions.
An arrhythmia detection from a 12-lead varied-length
ECG using Attention-based Time-Incremental Convolutional
Neural Network was presented in [249]. The method was
validated on the China Physiological Signal Challenge
Database [250], which consists of atrial fibrillation, first-
degree atrioventricular block, left bundle branch block,
right bundle branch block, premature atrial contraction, pre-
mature ventricular contraction, ST-segment depression and
ST-segment elevation signals with varied length. Refer-
ences [251] discussed arrhythmia detection using a deep
genetic ensemble of classifiers. For determining the ECG
features, the spectral power density was estimated based on
Welch’s method and discrete Fourier transform to obtain
other useful features. References [252] presented a method
for automatic diagnosis of the 12-lead ECG using a deep
neural network. The performance for the system was vali-
dated on a large dataset consisting of more than 2 million
subjects [253]. 98% data has been utilized to train the system,
and the remaining 2% has been used to test the system. Refer-
ences [254] identified the sleep disorders with only ECG sig-
nals by utilizing optimal antisymmetric biorthogonal wavelet
filter banks. It has categorized the signals using supervised
machine learning approaches and outperformed the other
methods based on other physiological signals. In [255], ECG
arrhythmia classification has been done by using Convolu-
tional Neural Networks and recurrence plots. The 1-D ECG
data has been converted to 2-D recurrence plots and further
utilized the classification of arrhythmias that has been val-
idated using publicly available databases. In [256], authors
detected the P wave, QRS complex and T waves of an ECG
signal using the Long Short Term Memory (LSTM) com-
bined with the Convolutional Neural Networks by utilizing
the temporal features. In [257], a hybrid ECG arrhythmia
classification algorithm termed Manta Ray Foraging Opti-
mization with SVM is proposed to automatically determine
the relevant features of Local Binary Patterns, Higher Order
Statistics, wavelet and magnitude values for categorizing the
ECG signals.
As can be seen in the works mentioned above, most of
the methods utilize the databases as the input signal and
are processed as a software-based approach. Most systems
offered good efficiency with the increment in computational
load. The methods generally adopted the supervised learning
schemes for the classification on the databases that added
constraint on the system level implementation because for the
real time data, the system needs to be trained on real time
ECG data.
This review provides insight into the invention of ECG, global
acceptance of ECG, the evolution of leads, and the system’s
transformation from the huge String galvanometer to portable
monitors in hospital settings.
Additionally, it offers a view on the earlier ECG signal
processing schemes to the most recent ones. The signal pro-
cessing can be categorized mainly into four steps: acquir-
ing the ECG signal, preprocessing of ECG signal, feature
extraction in different domains, and classification schemes.
The research challenges of current ECG processing trends are
discussed below.
1. The machine learning, deep learning techniques and
wavelet transforms are designed on the software processing
domains that pose a constraint for the real time systems.
2. The researchers utilized the databases available on Phy-
sionet for data acquisition that are old dated. For example,
the widely used MIT-BIH arrhythmia database was recorded
during 1975-1980. Over the last 40 years, various standard
definitions for the diseases changed according to various
standards viz., American Heart Association(AHA), European
Society of Cardiology (ESC) etc. Therefore, the use of the
annotations provided in the database for comparing the results
VOLUME 9, 2021 140873
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
with the automatic means remains slightly questionable.
Improvements of annotations with the recent standards and
inclusion of most recent databases may lead to more accuracy.
3. Lead uniformity is missing with the automatic methods
discussed. For example, the MIT-BIH arrhythmia database
provides the MLII lead and V5 leads and other databases such
as PTB provides the 15 Lead data. Some of the methods are
tested on the 12 Lead data and others are on 2 Lead data.
4. The literature does not utilize patients information
(age, gender, previous medical history etc.) provided with the
dataset that can add additional variables for the classifiers and
can lead to more efficient systems, according to the European
society of cardiology.
5. Utilization of supervised learning classifiers is suitable
to obtain results on the databases. For system implementation,
the data needs to be trained and tested based on the actual
data and that is a challenge in the case of supervised learning
6. The demographics conditions of the countries or regions
may vary; hence the algorithm developed to diagnose for a
specific region of people must be adaptable to these condi-
tions, which can be validated particularly with the locally
obtained database. The standard/existing databases do not
address this.
This review also concludes that researchers could explore
the possible improvements in the system’s quality for field
deployment. Fig. 17 shows possible directions (shaded) that
needs more attention.
FIGURE 17. Future directions for researchers in the ECG signal processing
domain. Gray areas represent the dimensions where future researchers
can look into to fill the Gaps in the current research trends.
1. The validation of the system with the locally developed
database may provide better accuracy in real time environ-
ment as cardiac diseases are highly dependent on individual
lifestyle and eating habits. Researchers’ focus is needed to
develop the databases according to the demographics where
we need to implement the system.
2. Signal quality assurance is the way forward for the
researchers. However, some methods employ the preprocess-
ing or denoising techniques to remove the random noises
existing in the ECG signal, but the effects of such techniques
on the ECG features are not discussed and these techniques
are employed without assuring the quality of the signal as
an application of such techniques may alter the essential
parameters present in the signal. Hence, we propose that
the signal quality assurance stage after the signal acquisition
stage may be beneficial as if the signal quality is passed,
then there is no need to preprocess the signal and if it
fails, there must be a preprocessing stage to autocorrect the
3. Research alternatives for the application of supervised
learning techniques to portable systems must be studied.
In this review, we found out that recent ECG signal processing
techniques mainly utilize the machine learning approaches
based on supervised learning techniques on the available
databases by utilizing the time domain, frequency domain,
time-frequency domain, and statistical features. The problem
associated with the methods is that it requires the annotations
or the truth values for testing the systems that create the
problem for the implementation of the system in the real time
scenario when the ECG signal is obtained locally with the
sensors as the system must be trained with the similar ECG
Computational loads in Hardware settings need to be stud-
ied for the computationally exhaustive deep learning tech-
niques to make them compatible with real-time systems.
4. The significance of VCG, needs to be studied fur-
ther for automatic detections. Some of the earlier researches
[5], [105], [106] shows the significance of VCG over the ECG
signal, but it is still not a popular choice among researchers.
The reason may be the doctors are accustomed to the 12 Lead
ECG signals. Comparative studies of ECG and VCG based
methods may open the doors for a new dimension in this
5. Further to this, adding the parameters such as demo-
graphics and other user specific parameters could lead to
more efficient and accurate systems.
The authors would like to thank the Dhirubhai Ambani Insti-
tute of Information and Communication Technology for the
research support.
[1] J. Burnett, ‘‘The origins of the electrocardiograph as a clinical instru-
ment,’Med. Hist., vol. 29, no. S5, pp. 53–76, 1985.
[2] S. L. Barron, ‘‘Development of the electrocardiograph in great Britain,’
Brit. Med. J., vol. 1, no. 4655, p. 720, 1950.
[3] W. B. Fye, ‘‘A history of the origin, evolution, and impact of elec-
trocardiography,’’ Amer. J. Cardiol., vol. 73, no. 13, pp. 937–949,
May 1994.
[4] M. Wasimuddin, K. Elleithy, A. Abuzneid, M. Faezipour, and
O. Abuzaghleh, ‘‘Stages-based ecg signal analysis from traditional
signal processing to machine learning approaches: A survey,’’ IEEE
Access, vol. 8, pp. 177782–177803, 2020.
140874 VOLUME 9, 2021
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
[5] S. Ansari, N. Farzaneh, M. Duda, K. Horan, H. B. Andersson,
Z. D. Goldberger, B. K. Nallamothu, and K. Najarian, ‘‘A review of auto-
mated methods for detection of myocardial ischemia and infarction using
electrocardiogram and electronic health records,’IEEE Rev. Biomed.
Eng., vol. 10, pp. 264–298, 2017.
[6] M. Z. Suboh, R. Jaafar, N. A. Nayan, and N. H. Harun, ‘‘ECG-based
detection and prediction models of sudden cardiac death: Current per-
formances and new perspectives on signal processing techniques,’’ Int. J.
Online Biomed. Eng., vol. 15, no. 15, p. 110, Dec. 2019.
[7] A. Bansal and R. Joshi, ‘‘Portable out-of-hospital electrocardiogra-
phy: A review of current technologies,’’ J. Arrhythmia, vol. 34, no. 2,
pp. 129–138, 2018.
[8] A. Blondel, ‘‘Oscillographes, nouveaux appareils pour l’étude des oscil-
lations électriques lentes,’Comptes rendus de l’Académie des Sciences,
vol. 116, pp. 502–506, 1893.
[9] S. L. Baron, The Development of the Duddell Oscillograph, With Some
Biographical Notes on William du Bois Duddell, no. 2. London, U.K.:
Cambridge Instrument Company, 1950.
[10] C. Ader, ‘‘Sur un nouvel appareil enregistreur pour cables sous-marins,’’
CR Acad Sci, Paris, vol. 124, pp. 1440–1442, 1897.
[11] T. N. Raju, ‘‘The Nobel chronicles. 1924: Willem Einthoven (1860-
1927),’Lancet, vol. 352, no. 9139, 1998, p. 1560.
[12] G. Green and J. T. Lloyd, Kelvin’s Instruments and the Kelvin Museum.
Glasgow, U.K.: Univ. of Glasgow, 1970.
[13] R. Kennedy,Electrical Installations of Electric Light, Power, Tractionand
Industrial Electrical Machinery, vol. 4. Edinburgh, U.K.: Jack Publisher,
[14] E. J. Marey, ‘‘Des variations électriques des muscles et du coeur en parti-
culier, étudiées au moyen de l’électromètre de M. Lippmann,’’ Acadamie
des Sci., vol. 82, pp. 975–977, 1876.
[15] J. Timon-David, ‘‘Premiere étude des insectes fossiles oligocenes du
basin de Marseille,’Comp. Rendus Hebdomadaires des Séances de
l’Acadamie des Sciences, vol. 216, pp. 455–457, 1943.
[16] A. D. Waller, ‘‘A demonstration on man of electromotive changes
accompanying the Heart’s beat,’’ J. Physiol., vol. 8, no. 5, pp. 229–234,
Oct. 1887.
[17] Capillary Electrometer. Accessed: Jun. 25, 2021. [Online]. Avail-
[18] E. Besterman and R. Creese, ‘‘Waller–pioneer of electrocardiography,’
Brit. Heart J., vol. 42, no. 1, p. 61, 1979.
[19] N. Herring, ‘‘ECG diagnosis of acute ischaemia and infarction: Past,
present and future,’QJM, vol. 99, no. 4, pp. 219–230, Feb. 2006.
[20] W. Einthoven, ‘‘Die galvanometrische Registrirung des menschlichen
Elektrokardiogramms, zugleich eine Beurtheilung der Anwendung des
Capillar-Elektrometers in der Physiologie,’Pflüger, Archiv für die
Gesammte Physiologie des Menschen und der Thiere, vol. 99, nos. 9–10,
pp. 472–480, Nov. 1903.
[21] W. Einthoven, ‘‘Ein neues galvanometer,’’ Annalen der Physik, vol. 317,
no. 13, pp. 1059–1071, 1903.
[22] W. Einthoven, A. S. Jaffe, P. Venge, and B. Lindahl, ‘‘Galvanometrische
registratie van het menschelijk electrocardiogram,’’ in Herinneringsbun-
del Professor SS Rosenstein. Leiden, The Netherlands: Eduard Ijdo, 1902,
pp. 101–107.
[23] W. Einthoven, ‘‘Uber die from des menschlichen Elektrokadiogramms,’’
Pflugers Arch fd ges des Physiol, vol. 60, no. 3, pp. 101–123, 1895.
[24] J. W. Hurst, ‘‘Naming of the waves in the ECG, with a brief account of
their genesis,’Circulation, vol. 98, no. 18, pp. 1937–1942, Nov. 1998.
[25] W.Einthoven, ‘‘Weitere Mitteilungen über das Saitengalvanometer.Anal-
yse der saitengalvanometrischen Kurven. Masse und Spannung des
Quarzfadens und Widerstand gegen die Fadenbewegung,’Annalen der
Physik, vol. 326, no. 14, pp. 665–700, 1906.
[26] W. Einthoven and B. Vaandbager, ‘‘Weiteres über das Elektrokardio-
gramm,’Ges. Physiol., vol. 122, no. 12, pp. 517–584, 1908.
[27] W. Einthoven, ‘‘The different forms of the human electrocardiogram and
their signification,’Lancet, vol. 179, no. 4622, pp. 853–861, 1912.
[28] S. Parker, A. Black, P. Parker, S. Regan and M. Weeks, Medicine:
The Definitive Illustrated History. London, U.K.: DK, 2016.
[29] N. J. Holter, ‘‘New method for heart studies: Continuous electrocardio-
graphy of active subjects over long periods is now practical,’Science,
vol. 134, no. 3486, pp. 1214–1220, Oct. 1961.
[30] W. Einthoven, ‘‘Le tl cardiogramme,’’ Arch. Internat Physiol., vol. 4,
pp. 132–164, 1906.
[31] T. Lewis, ‘‘Report CXIX. Auricular fibrillation: A common clinical con-
dition,’Brit. Med. J., vol. 2, no. 2552, p. 1528, 1909.
[32] Jonathan Cardy. Thomas Lewis Electrocardiograph Machine at Science
Museum in London. Accessed: Jun. 25, 2021. [Online]. Available:
[33] W. Einthoven, ‘‘The string galvanometer and the measurement of the
action currents of the heart,’Nobel Lecture, vol. 11, pp. 94–111,
Dec. 1925.
[34] B. S. Oppenheimer and M. A. Rothschild, ‘‘Electrocardiographic changes
associated with myocardial involvement: With special reference to prog-
nosis,’J. Amer. Med. Assoc., vol. 69, no. 6, pp. 429–431, 1917.
[35] H. E. Pardee, ‘‘An electrocardiographic sign of coronary artery obstruc-
tion,’Arch. Internal Med., vol. 26, no. 2, pp. 244–257, 1920.
[36] W. Mobitz, ‘‘Über die unvollständige störung der erregungs-überleitung
zwischen vorhof und kammer des menschlichen herzens,’Zeitschrift für
die Gesamte Experimentelle Medizin, vol. 41, no. 1, pp. 180–237, 1924.
[37] L. Wolff, J. Parkinson, and P. D. White, ‘‘Bundle branch block with short
PR interval in healthy young people prone to paroxysmal tachycardia,’’
Amer. Heart J., vol. 5, no. 6, p. 685, 1930.
[38] F. C. Wood and C. C. Wolferth, ‘‘Angina pectoris: The clinical and
electrocardiographic phenomena of the attack and their comparison with
the effects of experimental temporary coronary occlusion,’’ Arch. Internal
Med., vol. 47, no. 3, pp. 339–365, 1931.
[39] S. Goldhammer, ‘‘Electrokardiographische untersuchungen bei Kranken
mit Angina Pectoris,’Z. Klin. Med., vol. 122, p. 134, 1932.
[40] S. McGinn and P. D. White, ‘‘Acute cor pulmonale resulting from pul-
monary embolism: Its clinical recognition,’’ in J. Amer. Med. Assoc.,
vol. 104, no. 17, pp. 1473–1480, 1935.
[41] R. Langendorf, ‘‘Elektrokardiogramm bei Vorhof-Infarkt,’Acta Medica
Scandinavica, vol. 100, nos. 1–2, pp. 136–149, Apr. 2009.
[42] A. M. Master, R. Friedman, and S. Dack, ‘‘The electrocardiogram after
standard exercise as a functional test of the heart,’’ Amer. Heart J., vol. 24,
no. 6, pp. 777–793, Dec. 1942.
[43] E. W. Young and A. Koenig, ‘‘Auricular infarction,’Amer. Heart J.,
vol. 28, no. 3, pp. 287–294, Sep. 1944.
[44] M. Alghatrif and J. Lindsay, ‘‘A brief review: History to understand
fundamentals of electrocardiography,’’ J. Community Hospital Internal
Med. Perspect., vol. 2, no. 1, p. 14383, Apr. 2012.
[45] E. Goldberger, ‘‘A simple, indifferent, electrocardiographic electrode of
zero potential and a technique of obtaining augmented, unipolar, extrem-
ity leads,’Amer. Heart J., vol. 23, no. 4, pp. 483–492, 1942.
[46] E. Frank, ‘‘General theory of heart-vector projection,’’ Circulat. Res.,
vol. 2, no. 3, pp. 258–270, May 1954.
[47] F. N. Wilson, C. E. Kossmann, G. E. Burch, E. Goldberger, A. Graybiel,
H. H. Hecht, F. D. Johnston, E. Lepeschkin, and G. B. Myers, ‘‘Recom-
mendations for standardization of electrocardiographic and vectorcardio-
graphic leads,’Circulation, vol. 10, no. 4, pp. 564–573, Oct. 1954.
[48] G. van Herpen, ‘‘Professor Herman Burger (1893-1965), eminent teacher
and scientist, who laid the theoretical foundations of vectorcardiography–
and electrocardiography,’’ J. Electrocardiol., vol. 47, no. 2, pp. 168–174,
Mar. 2014.
[49] S. Man, A. C. Maan, M. J. Schalij, and C. A. Swenne, ‘‘Vectorcardio-
graphic diagnostic & prognostic information derived from the 12-lead
electrocardiogram: Historical review and clinical perspective,’’ J. Elec-
trocardiol., vol. 48, no. 4, pp. 463–475, Jul. 2015.
[50] E. Simonson, ‘‘Tuna N, Okamoto N. Toshima H. Diagnostic accuracy of
the vectorcardiogram and electrocardiogram,’Amer. J. Cardiol., vol. 17,
no. 6, pp. 829–878, 1966.
[51] C. L. Levkov, ‘‘Orthogonal electrocardiogram derived from the limb and
chest electrodes of the conventional 12-lead system,’’ Med. Biol. Eng.
Comput., vol. 25, no. 2, pp. 155–164, Mar. 1987.
[52] L. Edenbrandt and O. Pahlm, ‘‘Vectorcardiogram synthesized from a 12-
lead ECG: Superiority of the inverse dower matrix,’’ J. Electrocardiol.,
vol. 21, no. 4, pp. 361–367, Jan. 1988.
[53] J. A. Kors, G. Van Herpen, A. C. Sittig, and J. H. Van Bemmel, ‘‘Recon-
struction of the Frank vectorcardiogram from standard electrocardio-
graphic leads: Diagnostic comparison of different methods,’Eur. Heart
J., vol. 11, no. 12, pp. 1083–1092, Dec. 1990.
[54] J. M. Chia, S. E. Fischer, S. A. Wickline, and C. H. Lorenz, ‘‘Performance
of QRS detection for cardiac magnetic resonance imaging with a novel
vectorcardiographic triggering method,’J. Mag. Reson. Imag., vol. 12,
no. 5, pp. 678–688, Nov. 2000.
VOLUME 9, 2021 140875
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
[55] D. N. Ghista, U. R. Acharya, and T. Nagenthiran, ‘‘Frontal plane vector-
cardiograms: Theory and graphics visualization of cardiac health status,’
J. Med. Syst., vol. 34, no. 4, pp. 445–458, Aug. 2010.
[56] R. A. Brown and T. T. Schlegel, ‘‘Diagnostic utility of the spatial versus
individual planar QRS-T angles in cardiac disease detection,’J. Electro-
cardiology, vol. 44, no. 4, pp. 404–409, Jul. 2011.
[57] A. R. M. Dehnavi, I. Farahabadi, H. Rabbani, A. Farahabadi,
M. P. Mahjoob, and N. R. Dehnavi, ‘‘Detection and classification of car-
diac ischemia using vectorcardiogram signal via neural network,’J. Res.
Med. Sci., Off. J. Isfahan Univ. Med. Sci., vol. 16, no. 2, p. 136, 2011.
[58] L. Mignanelli and C. Rembe, ‘‘VCG signals on the thorax and detection
of the PR-interval,’’ in Laser Doppler Vibrometry for Non-Contact Diag-
nostics. Cham, Switzerland: Springer, 2020, pp. 155–166, .
[59] E. Prabhakararao and S. Dandapat, ‘‘A weighted SVM based approach
for automatic detection of posterior myocardial infarction using VCG
signals,’’ in Proc. Nat. Conf. Commun. (NCC), Feb. 2019, pp. 1–6.
[60] E. Prabhakararao and S. Dandapat, ‘‘Automated detection of posterior
myocardial infarction from VCG signals using stationary wavelet trans-
form based features,’IEEE Sensors Lett., vol. 4, no. 6, pp. 1–4, Jun. 2020.
[61] S. Majumder, L. Chen, O. Marinov, C.-H. Chen, T. Mondal, and
M. J. Deen, ‘‘Noncontact wearable wireless ECG systems for long-term
monitoring,’IEEE Rev. Biomed. Eng., vol. 11, pp. 306–321, 2018.
[62] E. Richard and A. D. C. Chan, ‘‘Design of a gel-less two-electrode ECG
monitor,’’ in Proc. IEEE Int. Workshop Med. Meas. Appl., Apr. 2010,
pp. 92–96.
[63] Y. M. Chi and G. Cauwenberghs, ‘‘Wireless non-contact EEG/ECG elec-
trodes for body sensor networks,’’ in Proc. Int. Conf. Body Sensor Netw.,
Jun. 2010, pp. 297–301.
[64] S. Peng, S. Bao, and W. Chen, ‘‘Capacitive coupled electrodes based
non-contact ECG measurement system with real-time wavelet denoising
algorithm,’’ in Proc. 41st Annu. Int. Conf. IEEE Eng. Med. Biol. Soc.
(EMBC), Jul. 2019, pp. 6587–6590.
[65] H. C. Jung, J. H. Moon, D. H. Baek, J. H. Lee, Y. Y. Choi, J. S. Hong,
and S. H. Lee, ‘‘CNT/PDMS composite flexible dry electrodes for long-
term ECG monitoring,’IEEE Trans. Biomed. Eng., vol. 59, no. 5,
pp. 1472–1479, May 2012.
[66] T. Niederhauser, A. Haeberlin, M. Jungo, J. Goette, M. Jacomet,
R. Abächerli, and R. Vogel, ‘‘Electrodes for long-term esophageal
electrocardiography,’’ IEEE Trans. Biomed. Eng., vol. 60, no. 9,
pp. 2576–2584, Sep. 2013.
[67] B. A. Reyes, H. F. Posada-Quintero, J. R. Bales, A. L. Clement,
G. D. Pins, A. Swiston, J. Riistama, J. P. Florian, B. Shykoff, M. Qin, and
K. H. Chon, ‘‘Novel electrodes for underwater ECG monitoring,’IEEE
Trans. Biomed. Eng., vol. 61, no. 6, pp. 1863–1876, Jun. 2014.
[68] T.-R. Hsu, T.-Y. Hsu, S.-Y. Yang, W.-H. Huang, Z.-C. Ou, and C.-C. Fan,
‘‘Wearable multi-lead ECG recorder with dry metal electrodes,’’ in Proc.
11th Int. Microsyst., Packag., Assem. Circuits Technol. Conf. (IMPACT),
Oct. 2016, pp. 73–76.
[69] M. G. Srinivasa and P. S. Pandian, ‘‘Dry electrodes for bio-potential
measurement in wearable systems,’’ in Proc. 2nd IEEE Int. Conf.
Recent Trends Electron., Inf. Commun. Technol. (RTEICT), May 2017,
pp. 270–276.
[70] H. Jung, D. Kwon, S. Lee, J. Ahn, A.-H. Kim, Y.-J. Kim, and J. Moon,
‘‘Carbon based electrode for wearable biosignal monitoring patch,’’ in
Proc. Int. Flexible Electron. Technol. Conf. (IFETC), Aug. 2018, pp. 1–2.
[71] Q. Qin, J. Li, S. Yao,C. Liu, H. Huang, and Y. Zhu, ‘‘Electrocardiogram of
a silver nanowire based dry electrode: Quantitative comparison with the
standard Ag/AgCl gel electrode,’IEEE Access, vol. 7, pp. 20789–20800,
[72] Y. Chihara and T. Yamakawa, ‘‘Evaluation of a developed multichannel
R-R interval telemeter and garment-type electrode,’’ in Proc. 41st Annu.
Int. Conf. IEEE Eng. Med. Biol. Soc. (EMBC), Jul. 2019, pp. 4173–4176.
[73] G. B. Tseghai, B. Malengier, K. A. Fante, A. B. Nigusse, B. B. Etana,
and L. Van Langenhove, ‘‘PEDOT:PSS/PDMS-coated cotton fabric for
ECG electrode,’’ in Proc. IEEE Int. Conf. Flexible Printable Sensors Syst.
(FLEPS), Aug. 2020, pp. 1–4.
[74] P. Kligfield, L. S. Gettes, J. J. Bailey, R. Childers, B. J. Deal,
E. W. Hancock, G. Van Herpen, J. A. Kors, P. Macfarlane, D. M. Mirvis,
and O. Pahlm, ‘‘Recommendations for the standardization and interpreta-
tion of the electrocardiogram: Part I: The electrocardiogram and its tech-
nology a scientific statement from the American heart association electro-
cardiography and arrhythmias committee, council on clinical cardiology;
The American college of cardiology foundation; and the heart Rhythm
society endorsed by the international society for computerized electro-
cardiology,’’ J. Amer. College Cardiol., vol. 49, no. 10, pp. 1109–1127,
Mar. 2007.
[75] G. Nave and A. Cohen, ‘‘ECG compression using long-term prediction,’’
IEEE Trans. Biomed. Eng., vol. 40, no. 9, pp. 877–885, Sep. 1993.
[76] C.-K. Tseng, L.-J. Kau, and W.-Y. Cheng, ‘‘A Takagi-Sugeno fuzzy
neural network-based predictive coding scheme for lossless compression
of ECG signals,’’ in Proc. IEEE Region Conf. (TENCON), Nov. 2017,
pp. 1646–1660.
[77] H. Lee and K. M. Buckley, ‘‘ECG data compression using cut and align
beats approach and 2-D transforms,’IEEE Trans. Biomed. Eng., vol. 46,
no. 5, pp. 556–564, May 1999.
[78] S.-C. Tai, C. Sun, and W.-C. Yan, ‘‘A 2-D ECG compression method
based on wavelet transform and modified SPIHT,’’ IEEE Trans. Biomed.
Eng., vol. 52, no. 6, pp. 999–1008, Jan. 2005.
[79] B. Huang, Y. Wang, and J. Chen, ‘‘2-D compression of ECG signals using
ROI mask and conditional entropy coding,’’ IEEE Trans. Biomed. Eng.,
vol. 56, no. 4, pp. 1261–1263, Apr. 2009.
[80] A. Tiwari and T. H. Falk, ‘‘Lossless electrocardiogram signal compres-
sion: A review of existing methods,’’ Biomed. Signal Process. Control,
vol. 51, pp. 338–346, May 2019.
[81] M. S. Manikandan and S. Dandapat, ‘‘Wavelet-based electrocardio-
gram signal compression methods and their performances: A prospec-
tive review,’’ Biomed. Signal Process. Control, vol. 14, pp. 73–107,
Nov. 2014.
[82] J. Qian, P. Tiwari. S. P. Gochhayat, and H. M. Pandey, ‘‘A noble double-
dictionary-based ECG compression technique for IoTH,’IEEE Internet
Things J., vol. 7, no. 10, pp. 10160–10170, Feb. 2020.
[83] R.B. Roy, A. Roy, A. Mukherjee, A. Ghosh, S. Bhattacharyya, and
M. K. Naskar, ‘‘Sparse encoding algorithm for real-time ECG compres-
sion,’’ in Recent Trends in Signal and Image Processing. Singapore:
Springer, 2019, pp. 31–38.
[84] S. Kumar, B. Deka, and S. Datta, ‘‘Multichannel ECG compression using
Block-Sparsity-based joint compressive sensing,’’ Circuits, Syst., Signal
Process., vol. 39, no. 12, pp. 6299–6315, Dec. 2020.
[85] F. Wang, Q. Ma, W. Liu, S. Chang, H. Wang, J. He, and Q. Huang,
‘‘A novel ECG signal compression method using spindle convolu-
tional auto-encoder,’’ Comput. Methods Programs Biomed., vol. 175,
pp. 139–150, Jul. 2019.
[86] V. Izadi, P. K. Shahri, and H. Ahani, ‘‘A compressed-sensing-based
compressor for ECG,’Biomed. Eng. Lett., vol. 10, no. 2, pp. 299–307,
May 2020.
[87] L. Rebollo-Neira, ‘‘Effective high compression of ECG signals at low
level distortion,’’ Sci. Rep., vol. 9, no. 1, pp. 1–12, Dec. 2019.
[88] C. K. Jha and M. H. Kolekar, ‘‘Empirical mode decomposition and
wavelet transform based ECG data compression scheme,’IRBM, vol. 42,
no. 1, pp. 65–72, Feb. 2021.
[89] A. Pandey, B. S. Saini, B. Singh, and N. Sood, ‘‘Quality controlled
ECG data compression based on 2D discrete cosine coefficient filtering
and iterative JPEG2000 encoding,’’ Measurement, vol. 152, Feb. 2020,
Art. no. 107252.
[90] H. B. Sprague, ‘‘History and present status of phonocardiography,’’ IRE
Trans. Med. Electron., vol. PGME-9, pp. 2–3, Dec. 1957.
[91] J. Nedoma, M. Fajkus, R. Martinek, S. Kepak, J. Cubik, S. Zabka, and
V. Vasinek, ‘‘Comparison of BCG, PCG and ECG signals in application
of heart rate monitoring of the human body,’’ in Proc. 40th Int. Conf.
Telecommun. Signal Process. (TSP), Jul. 2017, pp. 420–424.
[92] L. S. Stodieck and M. W. Luttges, ‘‘Relationships between the electro-
cardiogram and phonocardiogram: Potential for improved heart monitor-
ing,’ISA Trans., vol. 23, no. 4, pp. 59–65, 1984.
[93] L. Giovangrandi, O. T. Inan, R. M. Wiard, M. Etemadi, and
G. T. A. Kovacs, ‘‘Ballistocardiography—A method worth revisiting,’
in Proc. Annu. Int. Conf. IEEE Eng. Med. Biol. Soc., Aug. 2011,
pp. 4279–4282.
[94] E. J. Marey, La méthode Graphique dans les Sciences Expérimentales et
Particulièrement en Physiologie et en Médecine, vol. 1878, G. Masson,
Ed. G. Paris, France: Masson, 1878.
[95] E. Tafur, L. S. Cohen, and H. D. Levine, ‘‘The normal apex cardiogram:
Its temporal relationship to electrical, acoustic, and mechanical cardiac
events,’’ Circulation, vol. 30, no. 3, pp. 381–391, Sep. 1964.
[96] P. Castiglioni, A. Faini, G. Parati, and M. D. Rienzo, ‘‘Wearable seismo-
cardiography,’’ in Proc. 29th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc.,
Aug. 2007, pp. 3954–3957.
[97] W. H. Bancroft, Jr., ‘‘Kinetocardiography: Past and present,’Bibliotheca
Cardiologica, vol. 37, p. 58, Jan. 1979.
[98] C. A. Caceres, C. A. Steinberg, S. Abraham, W. J. Carbery, J. M. Mcbride,
W. E. Tolles, and A. E. Rikli, ‘‘Computer extraction of electro-
cardiographic parameters,’Circulation, vol. 25, no. 2, pp. 356–362,
Feb. 1962.
140876 VOLUME 9, 2021
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
[99] P. Rautaharju and H. Blackburn, ‘‘The exercise electrocardiogram: Expe-
rience in analysis of ‘noisy’ cardiograms with a small computer,’’ Amer.
Heart J., vol. 69, no. 4, pp. 515–520, 1965.
[100] R. E. Jensen, H. Shubin, P. F. Meagher, and M. H. Weil, ‘‘On-line
computer monitoring of the seriously-ill patient,’Med. Biol. Eng., vol. 4,
no. 3, pp. 265–272, May 1966.
[101] J. Klingeman and H. V. Pipberger, ‘‘Computer classifications of electro-
cardiograms,’Comput. Biomed. Res., vol. 1, no. 1, pp. 1–17, Mar. 1967.
[102] D. A. Brody, R. C. Arzbaecher, M. D. Woolsey, and T. Sato, ‘‘The nor-
mal atrial electrocardiogram: Morphologic and quantitative variability
in bipolar extremity leads,’Amer. Heart J., vol. 74, no. 1, pp. 4–12,
Jul. 1967.
[103] L. Pordy, H. Jaffe, K. Chesky, C. K. Friedberg, L. Fallowes, and
R. E. Bonner, ‘‘Computer diagnosis of electrocardiograms. IV. A com-
puter program for contour analysis with clinical results of rhythm and
contour interpretation,’Comput. Biomed. Res., vol. 1, no. 4, pp. 408–433,
Feb. 1968.
[104] L. J. Haywood, V. K. Murthy, G. A. Harvey, and S. Saltzberg, ‘‘On-
line real time computer algorithm for monitoring the ECG waveform,’
Comput. Biomed. Res., vol. 3, no. 1, pp. 15–25, Feb. 1970.
[105] P. W. Macfarlane, A. R. Lorimer, and T. D. Lawrie, ‘‘3 and 12 lead
electrocardiogram interpretation by computer. A comparison on 1093
patients,’Brit. Heart J., vol. 33, no. 2, p. 266, 1971.
[106] P. W. Macfarlane, A. R. Lorimer, and T. D. V. Lawrie, ‘‘ECG reporting
by computer,’’ Scottish Med. J., vol. 15, no. 10, pp. 363–369, Oct. 1970.
[107] H. Yang, S. T. Bukkapatnam, and R. Komanduri, ‘‘Spatiotemporal rep-
resentation of cardiac vectorcardiogram (VCG) signals,’Biomed. Eng.
online, vol. 11, no. 1, pp. 1–15, 2012.
[108] C. A. Swenne, J. H. van Bemmel, S. J. Hengeveld, and M. Hermans,
‘‘Pattern recognition for ECG-monitoring: An interactive method for the
classification of ventricular complexes,’’ Comput. Biomed. Res., vol. 6,
no. 2, pp. 150–160, Apr. 1973.
[109] F. Yanowitz, P. Kinias, D. Rawling, and H. A. Fozzard, ‘‘Accuracy of a
continuous real-time ECG dysrhythmia monitoring system,’Circulation,
vol. 50, no. 1, pp. 65–72, Jul. 1974.
[110] J. B. Walters,‘‘Microprocessor based ambulatory EKG monitor,’’ inProc.
29th ACEMB, Boston, MA, USA, 1976, p. 255.
[111] P. S. Schluter, ‘‘The design and evaluation of a bedside cardiac arrhythmia
monitor,’’ Ph.D. dissertation, Massachusetts Inst. Technol., Cambridge,
MA, USA, 1981.
[112] J. G. Webster, ‘‘An intelligent monitor for ambulatory ECGS,’’ Biomed.
Sci. Instrum., vol. 14, pp. 55–60, Apr. 1978.
[113] W.J. Tompkins, ‘‘A portable microcomputer-based system for biomedical
applications,’Biomed. Sci. Instrum., vol. 14, pp. 61–66, Apr. 1978.
[114] J. P. Abenstein, ‘‘Algorithms for real-time ambulatory ECG monitoring,’
Biomed. Sci. Instrum., vol. 14, p. 73, Apr. 1978.
[115] W. C. Mueller, ‘‘Arrhythmia detection program for an ambulatory ECG
monitor,’’ Biomed. Sci. Instrum., vol. 14, p. 81, Apr. 1978.
[116] P. Agarwal and R. Priemer, ‘‘Microprocessor based digital signal process-
ing system,’Comput. Biol. Med., vol. 9, no. 2, pp. 87–95, Jan. 1979.
[117] N. V. Thakor, J. G. Webster, and W. J. Tompkins, ‘‘Design, implementa-
tion and evaluation of amicrocomputer-based portable arrhythmia moni-
tor,’’ Med. Biol. Eng. Comput., vol. 22, no. 2, pp. 151–159, Mar. 1984.
[118] R. E. Hermes, D. B. Geselowitz, and G. C. Oliver, ‘‘Development,
distribution, and use of the American Heart Association database for
ventricular arrhythmia detector evaluation,’’ Comput. Cardiol., Jan. 1980,
pp. 263–266.
[119] R. G. Mark, P. S. Schluter, G. Moody, P. Devlin, and D. Chernoff,
‘‘An annotated ECG database for evaluating arrhythmia detectors,’IEEE
Trans. Biomed. Eng., vol. BE-29, no. 8, p. 600, Jan. 1982.
[120] N. V. Thakor, ‘‘Performance measures for computerized arrhythmia mon-
itors: Annotated database in golden standard,’’ in Proc. IEEE/Eng. Med.
Biol. Soc. Annu. Conf., Aug. 1982, pp. 211–215.
[121] G. B. Moody and R. G. Mark, ‘‘The impact of the MIT-BIH arrhythmia
database,’IEEE Eng. Med. Biol. Mag., vol. 20, no. 3, pp. 45–50,
May/Jun. 2001.
[122] L. Marsanova, A. Nemcova, R. Smisek, T. Goldmann, M. Vitek, and
L. Smital, ‘‘Automatic detection of P wave in ECG during ventricular
extrasystoles,’’ in Proc. World Congr. Med. Phys. Biomed. Eng. Singa-
pore: Springer, 2019, pp. 381–385.
[123] G. B. Moody and R. G. Mark, ‘‘A new method for detecting atrial
fibrillation using RR intervals,’Comput. Cardiol., vol. 10, pp. 227–230,
[124] G. B. Moody, R. G. Mark, and A. L. Goldberger, ‘‘Evaluation of the
TRIM ECG data compressor,’’ Comput. Cardiol., vol. 15, pp. 167–170,
Sep. 1988.
[125] A. L. Goldberger, L. A. N. Amaral, L. Glass, J. M. Hausdorff,
P. C. Ivanov, R. G. Mark, J. E. Mietus, G. B. Moody, C.-K. Peng, and
H. E. Stanley, ‘‘PhysioBank, PhysioToolkit, and PhysioNet: Components
of a new research resource for complex physiologic signals,’’ Circulation,
vol. 101, no. 23, pp. e215–e220, Jun. 2000. Accessed: May 26, 2021.
[126] S. D. Greenwald, ‘‘The development and analysis of a ventricular
fibrillation detector,’’ Ph.D. dissertation, Massachusetts Inst. Technol.,
Cambridge, MA, USA, 1986.
[127] G. B. Moody, W. K. Muldrow, and R. G. Mark, ‘‘A noise stress test for
arrhythmia detectors,’Comput. Cardiol., vol. 11, no. 3, pp. 381–384,
[128] Y. Ichimaru and G. B. Moody, ‘‘Development of the polysomnographic
database on CD-ROM,’’ Psychiatry Clin. Neurosci., vol. 53, no. 2,
pp. 175–177, 1999.
[129] P. Albrecht, ‘‘ST segment characterization for long term automated
ECG analysis,’’ Ph.D. dissertation, Dept. Elect. Eng. Comput. Sci., Mas-
sachusetts Inst. Technol., Cambridge, MA, USA, 1983.
[130] S. D. Greenwald, R. S. Patil, and R. G. Mark, Improved Detection
and Classification of Arrhythmias in Noise-Corrupted Electrocardio-
grams Using Contextual Information. New York, NY, USA: IEEE, 1990,
pp. 461–464.
[131] P. Laguna, R. G. Mark, A. Goldberg, and G. B. Moody, ‘‘A database
for evaluation of algorithms for measurement of QT and other wave-
form intervals in the ECG,’’ in Proc. Comput. Cardiol., Sep. 1997,
pp. 673–676.
[132] R. Bousseljot, D. Kreiseler, and A. Schnabel, ‘‘Nutzung der EKG-
signaldatenbank CARDIODAT der PTB über das internet,’’Biomedizinis-
che Technik, Biomed. Eng., vol. 40, no. s1, pp. 317–318, Jul. 2009.
[133] A. Taddei, G. Distante, M. Emdin, P. Pisani, G. B. Moody, C. Zeelenberg,
and C. Marchesi, ‘‘The European ST-T database: Standard for evaluating
systems for the analysis of ST-T changes in ambulatory electrocardiogra-
phy,’’ Eur. Heart J., vol. 13, no. 9, pp. 1164–1172, Sep. 1992.
[134] García-González, A. Argelagós-Palau, M. Fernández-Chimeno, and
J. Ramos-Castro, ‘‘A comparison of heartbeat detectors for the seismo-
cardiogram,’’ in Proc. Comput. Cardiol., Sep. 2013, pp. 461–464.
[135] Q. Zhang, D. Zhou, and X. Zeng, ‘‘A novel framework for motion-
tolerant instantaneous heart rate estimation by phase-domain multiview
dynamic time warping,’IEEE Trans. Biomed. Eng., vol. 64, no. 11,
pp. 2562–2574, Nov. 2017.
[136] N. Iyengar, C. K. Peng, R. Morin, A. L. Goldberger, and L. A. Lipsitz,
‘‘Age-related alterations in the fractal scaling of cardiac interbeat interval
dynamics,’Amer. J. Physiol.-Regulatory, Integrative Comparative Phys-
iol., vol. 271, no. 4, pp. R1078–R1084, Oct. 1996.
[137] T. Penzel, G. B. Moody, R. G. Mark, A. L. Goldberger, and J. H. Peter,
‘‘The apnea-ECG database,’’ in Proc. Comput. Cardiol., vol. 27,
Sep. 2000, pp. 255–258.
[138] G. Moody, A. Goldberger, S. McClennen, and S. Swiryn, ‘‘Predicting the
onset of paroxysmal atrial fibrillation: The computers in cardiology chal-
lenge 2001,’’ in Proc. Comput. Cardiol., vol. 28, Sep. 2001, pp. 113–116.
[139] M. G. Terzano, L. Parrino, A. Sherieri, R. Chervin, S. Chokroverty,
C. Guilleminault, M. Hirshkowitz, M. Mahowald, H. Moldofsky, A. Rosa,
R. Thomas, and A. Walters, ‘‘Atlas, rules, and recording techniques for
the scoring of cyclic alternating pattern (CAP) in human sleep,’Sleep
Med., vol. 2, no. 6, pp. 537–553, Nov. 2001.
[140] S. D. Greenwald, ‘‘The development and analysis of a ventricular fibril-
lation detector,’’ M.S. thesis, Massachusetts Inst. Technol., Cambridge,
MA, USA, 1986.
[141] F. M. Nolle, F. K. Badura, J. M. Catlett, R. W. Bowser, and M. H. Sketch,
‘‘CREI-GARD, a new concept in computerized arrhythmia monitoring
systems,’Comput. Cardiol., vol. 13, pp. 515–518, Oct. 1986.
[142] M. L. Ahlstrom and W. J. Tompkins, ‘‘Automated high-speed analy-
sis of Holter tapes with microcomputers,’IEEE Trans. Biomed. Eng.,
vol. BME-30, no. 10, pp. 651–657, Oct. 1983.
[143] M. E. Nygårds and L. Sörnmo, ‘‘Delineation of the QRS complex using
the envelope of the ECG,’’ Med. Biol. Eng. Comput., vol. 21, no. 5,
pp. 538–547, Feb. 1983.
[144] N. V. Thakor and D. Moreau, ‘‘Design and analysis of quantised coef-
ficient digital filters: Application to biomedical signal processing with
microprocessors,’Med. Biol. Eng. Comput., vol. 25, no. 1, pp. 18–25,
Jan. 1987.
[145] W. J. Tompkins, Biomedical Digital Signal Processing.
Upper Saddle River, NJ, USA: Prentice-Hall, 1993.
[146] J. P. Abenstein and W. J. Tompkins, ‘‘A new data-reduction algorithm for
real-time ECG analysis,’IEEE Trans. Biomed. Eng., vol. BME-29, no. 1,
pp. 43–48, Jan. 1982.
VOLUME 9, 2021 140877
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
[147] L. N. Bohs and R. C. Barr, ‘‘Prototype for real-time adaptive sampling
using the fan algorithm,’Med. Biol. Eng. Comput., vol. 26, no. 6,
pp. 574–583, Nov. 1988.
[148] J. R. Cox, F. M. Nolle, H. A. Fozzard, and G. C. Oliver, ‘‘AZTEC, a
preprocessing program for real-time ECG rhythm analysis,’IEEE Trans.
Biomed. Eng., vol. BME-15, no. 2, pp. 128–129, Apr. 1968.
[149] P. S. Hamilton and W. J. Tompkins, ‘‘Compression of the ambulatory
ECG by average beat subtraction and residual differencing,’’ IEEE Trans.
Biomed. Eng., vol. 38, no. 3, pp. 253–259, Mar. 1991.
[150] L. W. Gardenhire, ‘‘Data compression for biomedical telemetry,’’ in
Biomedical Telemetry. New York, NY, USA: Academic, 1965.
[151] W. S. Kuklinski, ‘‘Fast Walsh transform data-compression algorithm:
ECG applications,’Med. Biol. Eng. Comput., vol. 21, no. 4, pp. 465–472,
[152] S. C. Tai, ‘‘Slope—A real-time ECG data compressor,’Med. Biol. Eng.
Comput., vol. 29, no. 2, pp. 175–179, 1991.
[153] J. Pan and W. J. Tompkins, ‘‘A real-time QRS detection algorithm,’IEEE
Trans. Biomed. Eng., vol. BE-3, no. 3, pp. 230–236, Mar. 1985.
[154] G. Baselli, S. Cerutti, S. Civardi, F. Lombardi, A. Malliani, M. Merri,
M. Pagani, and G. Rizzo, ‘‘Heart rate variability signal processing:
A quantitative approach as an aid to diagnosis in cardiovascular patholo-
gies,’Int. J. Bio-Med. Comput., vol. 20, nos. 1–2, pp. 51–70, Jan. 1987.
[155] F. Gritzali, ‘‘Towards a generalized scheme for QRS detection in ECG
waveforms,’Signal Process., vol. 15, no. 2, pp. 183–192, Sep. 1988.
[156] S. I. Gallant, ‘‘Connectionist expert systems,’’ Commun. ACM, vol. 31,
no. 2, pp. 152–169, Feb. 1988.
[157] Y. Peng and J. A. Reggia, ‘‘A connectionist model for diagnostic problem
solving,’IEEE Trans. Syst., Man, Cybern., vol. 19, no. 2, pp. 285–298,
Mar. 1989.
[158] G. Bortolan, R. Degani, and J. L. Willems, ‘‘Neural networks for ECG
classification,’’ in IEEE Proc. Comput. Cardiol., Sep. 1990, pp. 269–272.
[159] G. Bortolan, R. Degani, and J. L. Willems, ‘‘ECG classification with
neural networks and cluster analysis,’Comput. Cardiol., vol. 20,
pp. 177–180, Oct. 1991.
[160] Q. Xue, Y. H. Hu, and W. J. Tompkins, ‘‘Neural-network-based adaptive
matched filtering for QRS detection,’IEEE Trans. Biomed. Eng., vol. 39,
no. 4, pp. 317–329, Apr. 1992.
[161] Y. Suzuki, ‘‘Self-organizing QRS-wave recognition in ECG using neural
networks,’IEEE Trans. Neural Netw., vol. 6, no. 6, pp. 1469–1477,
Nov. 1995.
[162] J. H. Frenster, ‘‘Neural networks for pattern recognition in medical
diagnosis,’’ in Proc. 12th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc.,
Nov. 1990, pp. 1423–1424.
[163] J. A. Crowe, N. M. Gibson, M. S. Woolfson, and M. G. Somekh,
‘‘Wavelet transform as a potential tool for ECG analysis and compres-
sion,’J. Biomed. Eng., vol. 14, no. 3, pp. 268–272, May 1992.
[164] M. Akay, Y. M. Akay, P. Cheng, and H. H. Szeto, ‘‘Time-frequency
analysis of the electrocortical activity during maturation using wavelet
transform,’Biol. Cybern., vol. 71, no. 2, pp. 169–176, Jun. 1994.
[165] C. Li, C. Zheng, and C. Tai,‘‘Detection of ECG characteristic points using
wavelet transforms,’IEEE Trans. Biomed. Eng., vol. 42, no. 1, pp. 21–28,
Jan. 1995.
[166] J. S. Sahambi, S. N. Tandon, and R. K. P. Bhatt, ‘‘Using wavelet
transforms for ECG characterization. An on-line digital signal process-
ing system,’IEEE Eng. Med. Biol. Mag., vol. 16, no. 1, pp. 77–83,
Jan. 1997.
[167] M. Bahoura, M. Hassani, and M. Hubin, ‘‘DSP implementation of wavelet
transform for real time ECG wave forms detection and heart rate anal-
ysis,’Comput. Methods Programs Biomed., vol. 52, no. 1, pp. 35–44,
Jan. 1997.
[168] J. P. Martínez, R. Almeida, S. Olmos, A. P. Rocha, and P. Laguna,
‘‘A wavelet-based ECG delineator: Evaluation on standard databases,’
IEEE Trans. Biomed. Eng., vol. 51, no. 4, pp. 570–581, Apr. 2004.
[169] J. P. V. Madeiro, P. C. Cortez, J. M. da Silva Monteiro Filho, and
P. R. F. Rodrigues, ‘‘Techniques for noise suppression for ECG signal
processing,’’ in Developments and Applications for ECG Signal Process-
ing. New York, NY, USA: Academic, 2019, pp. 53–87.
[170] S. Chatterjee, R. S. Thakur, R. N. Yadav, L. Gupta, and D. K. Raghuvan-
shi, ‘‘Review of noise removal techniques in ECG signals,’’ IET Signal
Process., vol. 14, no. 9, pp. 569–590, Dec. 2020.
[171] I. I. Christov, ‘‘Real time electrocardiogram QRS detection using com-
bined adaptive threshold,’’ Biomed. Eng.Online, vol. 3, no. 1, p. 28, 2004.
[172] L. Sörnmo and P. Laguna, ‘‘Electrocardiogram(ECG) signal processing,’’
in Wiley Encyclopedia of Biomedical Engineering. Hoboken, NJ, USA:
Wiley, 2006.
[173] Y.-C. Yeh and W.-J. Wang, ‘‘QRS complexes detection for ECG signal:
The difference operation method,’Comput. Methods Programs Biomed.,
vol. 91, no. 3, pp. 245–254, Sep. 2008.
[174] D. Sadhukhan and M. Mitra, ‘‘R-peak detection algorithm for ECG using
double difference and RR interval processing,’’ Proc. Technol., vol. 4,
pp. 873–877, Jan. 2012.
[175] E. B. Mazomenos, T. Chen, A. Acharyya, A. Bhattacharya,
J. Rosengarten, and K. Maharatna, ‘‘A time-domain morphology
and gradient based algorithm for ECG feature extraction,’’ in Proc. IEEE
Int. Conf. Ind. Technol., Mar. 2012, pp. 117–122.
[176] M. Elgendi, M. Meo, and D. Abbott, ‘‘A proof-of-concept study: Simple
and effective detection of P and T waves in arrhythmic ECG signals,’’
Bioengineering, vol. 3, no. 4, p. 26, 2016.
[177] M. Elgendi, B. Eskofier, and D. Abbott, ‘‘Fast T wave detection calibrated
by clinical knowledge with annotation of P and T waves,’’ Sensors,
vol. 15, no. 7, pp. 17693–17714, 2015.
[178] S. K. Jain and B. Bhaumik, ‘‘An ultra low power ECG signal pro-
cessor design for cardiovascular disease detection,’’ in Proc. 37th
Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. (EMBC), Aug. 2015,
pp. 857–860.
[179] S. Jain and B. Bhaumik, ‘‘An energy efficient application specific inte-
grated circuit for electrocardiogram feature detection and its potential for
ambulatory cardiovascular disease detection,’’ Healthcare Technol. Lett.,
vol. 3, pp. 77–84, Mar. 2016.
[180] S. K. Jain and B. Bhaumik, ‘‘An energy efficient ECG signal processor
detecting cardiovascular diseases on smartphone,’’ IEEE Trans. Biomed.
Circuits Syst., vol. 11, no. 2, pp. 314–323, Apr. 2017.
[181] B. Mishra, N. Arora, and Y. Vora, ‘‘An ECG-PPG wearable device for
real time detection of various arrhythmic cardiovascular diseases,’’ in
Proc. 9th Int. Symp. Embedded Comput. Syst. Design (ISED), Dec. 2019,
pp. 1–5.
[182] N. Arora, B. Mishra, and Y. Vora, ‘‘A low power wearable device for
real-time electrocardiogram monitoring and cardiovascular arrhythmia
detection for resource constrained regions,’J. Low Power Electron.,
vol. 15, no. 2, pp. 234–245, Jun. 2019.
[183] D. P. Golden, R. A. Wolthuis, and G. W. Hoffler, ‘‘A spectral analysis
of the normal resting electrocardiogram,’IEEE Trans. Biomed. Eng.,
vol. BME-20, no. 5, pp. 366–372, Sep. 1973.
[184] O. Rompelman and R. J. Janssen, ‘‘Use of phase spectral information
in assessment of frequency contents of ECG Waveforms,’IEE Proc. A,
Physical Sci., Meas. Instrum., Manage. Educ.-Rev., vol. 129, no. 9,
pp. 679–683, Dec. 1982.
[185] G. A. Myers, G. J. Martin, N. M. Magid, P. S. Barnett, J. W. Schaad,
J. S. Weiss, M. Lesch, and D. H. Singer, ‘‘Power spectral analysis of
heart rate varability in sudden cardiac death: Comparison to other meth-
ods,’IEEE Trans. Biomed. Eng., vol. BME-33, no. 12, pp. 1149–1156,
Dec. 1986.
[186] S. Barro, R. Ruiz, D. Cabello, and J. Mira, ‘‘Algorithmic sequential
decision-making in the frequency domain for life threatening ventricular
arrhythmias and imitative artefacts: A diagnostic system,’’ J. Biomed.
Eng., vol. 11, no. 4, pp. 320–328, Jul. 1989.
[187] H. F. Schels, R. Haberl, G. Jilge, P. Steinbigler, and G. Steinbeck,
‘‘Frequency analysis of the electrocardiogram with maximum entropy
method for identification of patients with sustained ventricular tachy-
cardia,’IEEE Trans. Biomed. Eng., vol. 38, no. 9, pp. 821–826,
Sep. 1991.
[188] I. S. N. Murthy and U. C. Niranjan, ‘‘Component wave delineation of
ECG by filtering in the Fourier domain,’Med. Biol. Eng. Comput.,
vol. 30, no. 2, pp. 169–176, Mar. 1992.
[189] C. G. Brown, R. Dzwoncyzk, and D. R. Martin, ‘‘Physiologic measure-
ment of the ventricular fibrillation ECG signal: Estimating the duration of
ventricular fibrillation,’Ann. Emergency Med., vol. 22, no. 1, pp. 70–74,
Jan. 1993.
[190] H.-U. Strohmenger, K. H. Lindner, and C. G. Brown, ‘‘Analysis of the
ventricular fibrillation ECG signal amplitude and frequency parameters
as predictors of countershock success in humans,’Chest, vol. 111, no. 3,
pp. 584–589, Mar. 1997.
[191] I. Romero and L. Serrano, ‘‘ECG frequency domain features extrac-
tion: A new characteristic for arrhythmias classification,’’ in Proc. Conf.
23rd Annu. Int. Conf. IEEE Eng. Med. Biol. Soc., vol. 2, Oct. 2001,
pp. 2006–2008.
[192] M. Milanesi, N. Vanello, V. Positano, M. F. Santarelli, R. Paradiso,
D. D. Rossi, and L. Landini, ‘‘Frequency domain approach to blind source
separation in ECG monitoring by wearable system,’’ in Proc. Comput.
Cardiol., Sep. 2005, pp. 767–770.
140878 VOLUME 9, 2021
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
[193] A. Sethi, S. Arora, and A. Ballaney, ‘‘Frequency domain analysis of
ECG signals using auto-associative neural networks,’’ in Proc. Int. Conf.
Biomed. Pharmaceutical Eng., Dec. 2006, pp. 531–536.
[194] C. Lehmann, J. Reinstadtler, and A. Khawaja, ‘‘Detection of power-line
interference in ECG signals using frequency-domain analysis,’’ in Proc.
Comput. Cardiol., Sep. 2011, pp. 821–824.
[195] T. Sengthipphany, S. Tretriluxana, and K. Chitsakul, ‘‘Analysis of heart
rate variability and breath to breath interval in frequency domain,’’ in
Proc. 7th Biomed. Eng. Int. Conf., Nov. 2014, pp. 1–4.
[196] K.-C.-J. Chen and P.-C. Chien, ‘‘A fast ECG diagnosis using frequency-
based compressive neural network,’’ in Proc. IEEE 6th Global Conf.
Consum. Electron. (GCCE), Oct. 2017, pp. 1–2.
[197] P. S. Addison, J. N. Watson, G. R. Clegg, M. Holzer, F. Sterz, and
C. E. Robertson, ‘‘Evaluating arrhythmias in ECG signals using wavelet
transforms,’IEEE Eng. Med. Biol. Mag., vol. 19, no. 5, pp. 104–109,
Sep. 2000.
[198] Z. Lu, D. Y. Kim, and W. A. Pearlman, ‘‘Wavelet compression of ECG
signals by the set partitioning in hierarchical trees algorithm,’IEEE
Trans. Biomed. Eng., vol. 47, no. 7, pp. 849–856, Jul. 2000.
[199] R. S. H. Istepanian and A. A. Petrosian, ‘‘Optimal zonal wavelet-based
ECG data compression for a mobile telecardiology system,’IEEE Trans.
Inf. Technol. Biomed., vol. 4, no. 3, pp. 200–211, Sep. 2000.
[200] B. A. Rajoub, ‘‘An efficient coding algorithm for the compression of ECG
signals using the wavelet transform,’IEEE Trans. Biomed. Eng., vol. 49,
no. 4, pp. 355–362, Apr. 2002.
[201] J. P. Martínez, R. Almeida, S. Olmos, A. P. Rocha, and P. Laguna,
‘‘A wavelet-based ECG delineator: Evaluation on standard databases,’
IEEE Trans. Biomed. Eng., vol. 51, no. 4, pp. 570–581, Apr. 2004.
[202] S. Z. Mahmoodabadi, A. Ahmadian, M. D. Abolhasani, M. Eslami, and
J. H. Bidgoli, ‘‘ECG feature extraction based on multiresolution wavelet
transform,’’ in Proc. IEEE Eng. Med. Biol. 27th Annu. Conf., Jan. 2006,
pp. 3902–3905.
[203] C.-C. Chiu, C.-M. Chuang, and C.-Y. Hsu, ‘‘A novelpersonal identity ver-
ification approach using a discrete wavelet transform of the ECG signal,’
in Proc. Int. Conf. Multimedia Ubiquitous Eng., 2008, pp. 201–206.
[204] M. S. Manikandan and S. Dandapat, ‘‘Wavelet threshold based TDL and
TDR algorithms for real-time ECG signal compression,’Biomed. Signal
Process. Control, vol. 3, no. 1, pp. 44–66, Jan. 2008.
[205] S. Banerjee and M. Mitra, ‘‘Application of cross wavelet transform for
ECG pattern analysis and classification,’IEEE Trans. Instrum. Meas.,
vol. 63, no. 2, pp. 326–333, Feb. 2014.
[206] S. Sahoo, P. Biswal, T. Das, and S. Sabut, ‘‘De-noising of ECG Signal
and QRS detection using Hilbert transform and adaptive thresholding,’’
Proc. Technol., vol. 25, pp. 68–75, Jan. 2016.
[207] E. B. Mazomenos, D. Biswas, A. Acharyya, T. Chen, K. Maharatna,
J. Rosengarten, J. Morgan, and N. Curzen, ‘‘A low-complexity ECG
feature extraction algorithm for mobile healthcare applications,’IEEE
J. Biomed. Health Inform., vol. 17, no. 2, pp. 459–469, Mar. 2013.
[208] Y.-J. Min, H.-K. Kim, Y.-R. Kang, G.-S. Kim, J. Park, and S.-W. Kim,
‘‘Design of wavelet-based ECG detector for implantable cardiac pace-
makers,’IEEE Trans. Biomed. Circuits Syst., vol. 7, no. 4, pp. 426–436,
Aug. 2013.
[209] Y. Zou, J. Han, S. Xuan, S. Huang, X. Weng, D. Fang, and X. Zeng,
‘‘An energy-efficient design for ECG recording and R-peak detection
based on wavelet transform,’IEEE Trans. Circuits Syst. II, Exp. Briefs,
vol. 62, no. 2, pp. 119–123, Feb. 2015.
[210] Q. Zhang, D. Zhou, and X. Zeng, ‘‘HeartID: A multiresolution convo-
lutional neural network for ECG-based biometric human identification
in smart health applications,’IEEE Access, vol. 5, pp. 11805–11816,
[211] C. Venkatesan, P. Karthigaikumar, A. Paul, S. Satheeskumaran, and
R. Kumar, ‘‘ECG signal preprocessing and SVM classifier-based abnor-
mality detection in remote healthcare applications,’IEEE Access, vol. 6,
pp. 9767–9773, 2018.
[212] R. R. Sharma, M. Kumar, and R. B. Pachori, ‘‘Joint time-frequency
domain-based CAD disease sensing system using ECG signals,’IEEE
Sensors J., vol. 19, no. 10, pp. 3912–3920, May 2019.
[213] A. K. Sangaiah, M. Arumugam, and G.-B. Bian, ‘‘An intelligent learning
approach for improving ECG signal classification and arrhythmia analy-
sis,’Artif. Intell. Med., vol. 103, Mar. 2020, Art. no. 101788.
[214] M. Unser and A. Aldroubi, ‘‘A review of wavelets in biomedical
applications,’Proc. IEEE, vol. 84, no. 4, pp. 626–638,
Apr. 1996.
[215] A. N. Akansu, P. A. Haddad, R. A. Haddad and P. R. Haddad, Mul-
tiresolution Signal Decomposition: Transforms, Subbands, and Wavelets.
New York, NY, USA: Academic, 2001.
[216] P. E. Tikkanen, ‘‘Nonlinear wavelet and wavelet packet denoising of
electrocardiogram signal,’Biol. Cybern., vol. 80, no. 4, pp. 259–267,
Apr. 1999.
[217] A. Ebrahimzadeh and M. Azarbad, ‘‘ECG compression using wavelet
transform and three-level quantization,’’ in Proc. 6th Int. Conf. Digit.
Content, Multimedia Technol. Appl., Aug. 2010, pp. 250–254.
[218] I. Daubechies, ‘‘CBMS-NSF regional conference series in applied math-
ematics,’’ in Ten Lectures on Wavelets. Philadelphia, PA, USA: SIAM,
1992, p. 61.
[219] S. Parvaneh and J. Rubin, ‘‘Electrocardiogram monitoring and interpreta-
tion: From traditional machine learning to deep learning, and their com-
bination,’’ in Proc. Comput. Cardiol. Conf. (CinC), vol. 45, Dec. 2018,
pp. 1–4.
[220] J. Hoffmann, S. Mahmood, P. S. Fogou, N. George, S. Raha, S. Safi,
K. J. Schmailzl, M. Brandalero, and M. Hubner, ‘‘A survey on machine
learning approaches to ECG processing,’’ in Proc. Signal Process., Algo-
rithms, Archit., Arrangements, Appl. (SPA), Sep. 2020, pp. 36–41.
[221] J. R. Annam, S. Kalyanapu, S. Ch, J. Somala, and S. B. Raju, ‘‘Clas-
sification of ECG heartbeat arrhythmia: A review,’Proc. Comput. Sci.,
vol. 171, pp. 679–688, Jan. 2020.
[222] Y. H. Hu, S. Palreddy, and W. J. Tompkins, ‘‘A patient-adaptable ECG
beat classifier using a mixture of experts approach,’IEEE Trans. Biomed.
Eng., vol. 44, no. 9, pp. 891–900, Sep. 1997.
[223] G. Vijaya, V. Kumar, and H. K. Verma, ‘‘ANN-based QRS-complex
analysis of ECG,’J. Med. Eng. Technol., vol. 22, no. 4, pp. 160–167,
[224] R. Silipo and C. Marchesi, ‘‘Artificial neural networks for automatic ECG
analysis,’IEEE Trans. Signal Process., vol. 46, no. 5, pp. 1417–1425,
May 1998.
[225] M. Lagerholm, C. Peterson, G. Braccini, L. Edenbrandt, and L. Sornmo,
‘‘Clustering ECG complexes using Hermite functions and self-organizing
maps,’IEEE Trans. Biomed. Eng., vol. 47, no. 7, pp. 838–848, Jul. 2000.
[226] S. Osowski and T. H. Linh, ‘‘ECG beat recognition using fuzzy
hybrid neural network,’IEEE Trans. Biomed. Eng., vol. 48, no. 11,
pp. 1265–1271, Nov. 2001.
[227] T. H. Linh, S. Osowski, and M. Stodolski, ‘‘On-line heart beat recogni-
tion using Hermite polynomials and neuro-fuzzy network,’IEEE Trans.
Instrum. Meas., vol. 52, no. 4, pp. 1224–1231, Sep. 2003.
[228] S. Osowski, L. T. Hoai, and T. Markiewicz, ‘‘Support vector machine-
based expert system for reliable heartbeat recognition,’IEEE Trans.
Biomed. Eng., vol. 51, no. 4, pp. 582–589, Apr. 2004.
[229] P. de Chazal, M. O’Dwyer, and R. B. Reilly, ‘‘Automatic classification of
heartbeats using ECG morphology and heartbeat interval features,’IEEE
Trans. Biomed. Eng., vol. 51, no. 7, pp. 1196–1206, Jul. 2004.
[230] J. Rodriguez, A. Goni, and A. Illarramendi, ‘‘Real-time classification
of ECGs on a PDA,’’ IEEE Trans. Inf. Technol. Biomed., vol. 9, no. 1,
pp. 23–34, Mar. 2005.
[231] H. G. Hosseini, D. Luo, and K. J. Reynolds, ‘‘The comparison of different
feed forward neural network architectures for ECG signal diagnosis,’
Med. Eng. Phys., vol. 28, no. 4, pp. 372–378, 2006.
[232] W. Jiang and S. G. Kong, ‘‘Block-based neural networks for personalized
ECG signal classification,’IEEE Trans. Neural Netw., vol. 18, no. 6,
pp. 1750–1761, Nov. 2007.
[233] F. Melgani and Y. Bazi, ‘‘Classification of electrocardiogram signals with
support vector machines and particle swarm optimization,’IEEE Trans.
Inf. Technol. Biomed., vol. 12, no. 5, pp. 667–677, Sep. 2008.
[234] C. M. Fira and L. Goras, ‘‘An ECG signals compression method and
its validation using NNs,’IEEE Trans. Biomed. Eng., vol. 55, no. 4,
pp. 1319–1326, Apr. 2008.
[235] M. O. Mendez, A. M. Bianchi, M. Matteucci, S. Cerutti, and T. Penzel,
‘‘Sleep apnea screening by autoregressive models from a single ECG
lead,’IEEE Trans. Biomed. Eng., vol. 56, no. 12, pp. 2838–2850,
Dec. 2009.
[236] A. K. Mishra and S. Raghav, ‘‘Local fractal dimension based ECG
arrhythmia classification,’Biomed. Signal Process. Control, vol. 5, no. 2,
pp. 114–123, 2010.
[237] T. Mar, S. Zaunseder, J. P. Martínez, M. Llamedo, and R. Poll, ‘‘Optimiza-
tion of ECG classification by means of feature selection,’IEEE Trans.
Biomed. Eng., vol. 58, no. 8, pp. 2168–2177, Aug. 2011.
[238] X. Wang, Q. Gui, B. Liu, Z. Jin, and Y. Chen, ‘‘Enabling smart person-
alized healthcare: A hybrid mobile-cloud approach for ECG telemoni-
toring,’IEEE J. Biomed. Health Informat., vol. 18, no. 3, pp. 739–745,
Oct. 2013.
[239] A. Daamouche, L. Hamami, N. Alajlan, and F. Melgani, ‘‘A wavelet
optimization approach for ECG signal classification,’Biomed. Signal
Process. Control, vol. 7, no. 4, pp. 342–349, Jul. 2012.
VOLUME 9, 2021 140879
N. Arora, B. Mishra: Origins of ECG and Evolution of Automated DSP Techniques: A Review
[240] H. M. Rai, A. Trivedi, and S. Shukla, ‘‘ECG signal processing for abnor-
malities detection using multi-resolution wavelet transform and Artificial
Neural Network classifier,’’ Measurement, vol. 46, no. 9, pp. 3238–3246,
[241] S. Kiranyaz, T. Ince, and M. Gabbouj, ‘‘Real-time patient-specific
ECG classification by 1-D convolutional neural networks,’’ IEEE Trans.
Biomed. Eng., vol. 63, no. 3, pp. 664–675, Aug. 2015.
[242] P. Li, Y. Wang, J. He, L. Wang, Y. Tian, T. S. Zhou, T. Li, and J. S. Li,
‘‘High-performance personalized heartbeat classification model for long-
term ECG signal,’IEEE Trans. Biomed. Eng., vol. 64, no. 1, pp. 78–86,
Jan. 2017.
[243] B. Pourbabaee, M. J. Roshtkhari, and K. Khorasani, ‘‘Deep convolu-
tional neural networks and learning ECG features for screening parox-
ysmal atrial fibrillation patients,’IEEE Trans. Syst., Man, Cybern., Syst.,
vol. 48, no. 12, pp. 2095–2104, Dec. 2018.
[244] Ö. Yildirim, ‘‘A novel wavelet sequence based on deep bidirectional
LSTM network model for ECG signal classification,’Comput. Biol.
Med., vol. 96, pp. 189–202, May 2018.
[245] V. Gupta and M. Mittal, ‘‘KNN and PCA classifier with autoregressive
modelling during different ECG signal interpretation,’Proc. Comput.
Sci., vol. 125, pp. 18–24, Jan. 2018.
[246] J. Huang, B. Chen, B. Yao, and W. He, ‘‘ECG arrhythmia classification
using STFT-based spectrogram and convolutional neural network,’’ IEEE
Access, vol. 7, pp. 92871–92880, 2019.
[247] B. Hou, J. Yang, P. Wang, and R. Yan, ‘‘LSTM-based auto-encoder model
for ECG arrhythmias classification,’IEEE Trans. Instrum. Meas., vol. 69,
no. 4, pp. 1232–1240, Apr. 2020.
[248] E. Prabhakararao and S. Dandapat, ‘‘Myocardialinfarction severity stages
classification from ECG signals using attentional recurrent neural net-
work,’IEEE Sensors J., vol. 20, no. 15, pp. 8711–8720, Aug. 2020.
[249] Q. Yao, R. Wang, X. Fan, J. Liu, and Y. Li, ‘‘Multi-class arrhyth-
mia detection from 12-lead varied-length ECG using attention-based
time-incremental convolutional neural network,’’ Inf. Fusion, vol. 53,
pp. 174–182, Jan. 2020.
[250] F. Liu, C. Liu, L. Zhao, X. Zhang, X. Wu, X. Xu, Y. Liu, C. Ma, S. Wei,
Z. He, and J. Li, ‘‘An open access database for evaluating the algorithms
of electrocardiogram rhythm and morphology abnormality detection,’
J. Med. Imag. Health Informat., vol. 8, no. 7, pp. 1368–1373, Jul. 2018.
[251] P. Pławiak and U. R. Acharya, ‘‘Novel deep genetic ensemble of classi-
fiers for arrhythmia detection using ECG signals,’Neural Comput. Appl.,
vol. 32, no. 15, pp. 11137–11161, Aug. 2020.
[252] A. H. Ribeiro, M. H. Ribeiro, G. M. M. Paixão, D. M. Oliveira,
P. R. Gomes, J. A. Canazart, M. P. S. Ferreira, C. R. Andersson,
P. W. Macfarlane, W. Meira, T. B. Schön, and A. L. P. Ribeiro, ‘‘Auto-
matic diagnosis of the 12-lead ECG using a deep neural network,’Nature
Commun., vol. 11, no. 1, pp. 1–9, Dec. 2020.
[253] CODE Training Dataset. Accessed: Jun. 25, 2021. [Online]. Available:, doi: 10.5281/
[254] M. Sharma, H. S. Dhiman, and U. R. Acharya, ‘‘Automatic identifi-
cation of insomnia using optimal antisymmetric biorthogonal wavelet
filter bank with ECG signals,’Comput. Biol. Med., vol. 131, Apr. 2021,
Art. no. 104246.
[255] B. M. Mathunjwa, Y. T. Lin, C. H. Lin, M. F. Abbod, and J. S. Shieh,
‘‘ECG arrhythmia classification by using a recurrence plot and con-
volutional neural network,’’ Biomed. Signal Process. Control, vol. 64,
Feb. 2021, Art. no. 102262.
[256] A. Peimankar and S. Puthusserypady, ‘‘DENS-ECG: A deep learning
approach for ECG signal delineation,’Expert Syst. Appl., vol. 165,
Mar. 2021, Art. no. 113911.
[257] E. H. Houssein, I. E. Ibrahim, N. Neggaz, M. Hassaballah, and
Y. M. Wazery,‘‘An efficientECG arrhythmia classification method based
on manta ray foraging optimization,’Expert Syst. Appl., vol. 181,
Nov. 2021, Art. no. 115131.
NEHA ARORA received the B.E. degree in elec-
tronics and communication engineering from the
Rajasthan College of Engineering for Women,
Jaipur, Rajasthan, in 2007, and the M.Tech. degree
in VLSI design from the Mody University of
Science and Technology (MITS), Lakshmangarh,
Rajasthan, in 2010. She is currently pursuing the
Ph.D. degree with the Dhirubhai Ambani Institute
of Information and Communication Technology,
Gandhinagar, Gujarat. She worked as an Assistant
Professor at MITS, till 2012. Her research interests include embedded hard-
ware design, VLSI design, and biomedical signal processing.
BISWAJIT MISHRA received the bachelor’s
degree in ECE from the National Institute of Tech-
nology (formerly MREC), Jaipur, in 1996, and the
M.S. and Ph.D. degrees in electronic engineer-
ing from the University of Southampton, U.K.,
in 2003 and 2007, respectively. He joined as a
Research Fellow at the University of Southampton,
working on ultra-low power design methodology
and subthreshold circuits. In 2010, he joined as a
Senior Scientist at ESPLAB-EPFL, Switzerland,
working on energy harvesting electronics, ultra-low power management
units, and battery-less electronics. In 2013, he joined DAIICT, India. From
1996 to 2002, he has worked as a Research and Development Engineer at
Philips Semiconductors, India, U.K., and Switzerland; and Cadence Design
Systems, U.K. His research interests include ultra-low power and low voltage
circuits, sub-threshold design methodology, and geometric algebra hardware.
140880 VOLUME 9, 2021
... The ECG signal can be used to diagnose whether the test subject has ventricular atrial hypertrophy, myocardial ischemia and arrhythmias [12], while the ECG signal is easier to be detected than other bioelectrical signals. Therefore, ECG signals became one of the first biological signals studied by humans and applied in medical clinics [13]. Many studies have been conducted on the recognition of heart health status based on ECG signals [14][15][16][17][18]. The waves in the ECG wave cluster are named in alphabetical order and are P, Q, R, S, T, and U. ...
Full-text available
Electrocardiogram (ECG) signals are often used to diagnose cardiac status. However, most of the existing ECG diagnostic methods only use the time-domain information, resulting in some obviously lesion information in frequency-domain of ECG signals are not being fully utilized. Therefore, we propose a method to fuse the time and frequency domain information in ECG signals by convolutional neural network (CNN). First, we adapt multi-scale wavelet decomposition to filter the ECG signal; Then, R-wave localization is used to segment each individual heartbeat cycle; And then, the frequency domain information of this heartbeat cycle is extracted via fast Fourier transform. Finally, the temporal information is spliced with the frequency domain information and input to the neural network for classification. The experimental results show that the proposed method has the highest recognition accuracy (99.43%) of ECG singles compared with state-of-the-art methods.
The disadvantage of conventional signal processing methods used for Power Quality (PQ) classification is that their mathematical calculations are complex. Therefore, it is necessary to develop simple, effective, and fast signal processing tools to analyze the PQ signal. In this study, a new signal processing method with simple mathematical operations based on the Integer Factor (IF) down sampling/Approximation Derivatives (AD) was developed to analyze PQ signals. IF was employed to analyze the signals at various sampling frequencies, while the AD method was employed to obtain different degrees of detail and approximation coefficients for the signals. In order to prove that the IF/AD signal processing method can perform a fast and detailed signal processing analysis, the IF/AD method and the Discrete Wavelet Transforms (DWT)/Multiresolution method, which is one of the basic signal processing methods, were compared. It was determined that the developed signal processing method has analyzed PQ disturbances in a shorter time and in more detail than DWT/Multiresolution method. After it was determined that the IF/AD method was effective in the analysis of PQ signals, a new classification method based on the IF-AD signal processing approach, the Slime Mould feature selection algorithm, and Support Vector Machine was developed. The classification method was applied to synthetic signals with noise, real-time data generated in the laboratory and PQ online data. The results proved that the developed classification method was successful. Therefore, this study will provide a different approach to PQ classification methods and non-stationary signal analysis studies.
Full-text available
Sleep is a fundamental human physiological activity required for adequate working of the human body. Sleep disorders such as sleep movement disorders, nocturnal front lobe epilepsy, insomnia, and narcolepsy are caused due to low sleep quality. Insomnia is one such sleep disorder where a person has difficulty in getting quality sleep. There is no definitive test to identify insomnia; hence it is essential to develop an automated system to identify it accurately. A few automated methods have been proposed to identify insomnia using either polysomnogram (PSG) or electroencephalogram (EEG) signals. To the best of our knowledge, we are the first to automatically detect insomnia using only electrocardiogram (ECG) signals without combining them with any other physiological signals. In the proposed study, an optimal antisymmetric biorthogonal wavelet filter bank (ABWFB) has been used, which is designed to minimize the joint duration-bandwidth localization (JDBL) of the underlying filters. The -norm feature is computed from the various wavelet sub-bands coefficients of ECG signals. The norm features are fed to various supervised machine learning classifiers for the automated detection of insomnia. In this work, ECG recordings of seven insomnia patients and six normal subjects are used from the publicly available cyclic alternating pattern (CAP) sleep database. We created ten different subsets of ECG signals based on annotations of sleep-stages, namely wake (W), S1, S2, S3, S4, rapid eye moment (REM), light sleep stage (LSS), slow-wave sleep (SWS), non-rapid eye movement (NREM) and W+S1+S2+S3+S4+REM for the automated identification of insomnia. Our proposed ECG-based system obtained the highest classification accuracy of 97.87%, F1-score of 97.39%, and Cohen’s kappa value of 0.9559 for K-nearest neighbour (KNN) with the ten-fold cross-validation strategy using ECG signals corresponding to the REM sleep stage. The support vector machine (SVM) yielded the highest value of 0.99 for area under the curve with the ten fold cross-validation corresponding to REM sleep stage.
Conference Paper
Full-text available
Electrocardiogram (ECG) signals convey a substantial amount of information that can be used for detecting and predicting the occurrence of several diseases and conditions. Approaches to ECG analysis were traditionally based on Signal Processing (SP), but several recent work have managed to substantially increase the quality of the analyses by using Machine Learning (ML) techniques. Still, while ML offers the potential to extract a substantially more information and predict diseases with better accuracy, it is also intrinsically more computationally expensive. Given the importance of this field and recent advances, we present a survey on ML approaches to ECG processing, focusing on particular diseases and conditions that can be detected and the different algorithms used for that. Moreover, we also discuss recent implementations of such algorithms on low-power wearable devices. We identify an opportunity for the development of novel embedded architectures that could enable the continuous monitoring of ECG signals and identify emerging technologies that could help in paving the way towards that.
Full-text available
Electrocardiogram (ECG) gives essential information about different cardiac conditions of the human heart. Its analysis has been the main objective among the research community to detect and prevent life threatening cardiac circumstances. Traditional signal processing methods, machine learning and its subbranches, such as deep learning, are popular techniques for analyzing and classifying the ECG signal and mainly to develop applications for early detection and treatment of cardiac conditions and arrhythmias. A detailed literature survey regarding ECG signal analysis is presented in this article. We first introduce a stages-based model for ECG signal analysis where a survey of ECG analysis related work is then presented in the form of this stage-based process model. The model describes both traditional time/frequency-domain and advanced machine learning techniques reported in the published literature at every stage of analysis, starting from ECG data acquisition to its classification for both simulations and real-time monitoring systems. We present a comprehensive literature review of real-time ECG signal acquisition, prerecorded clinical ECG data, ECG signal processing and denoising, detection of ECG fiducial points based on feature engineering and ECG signal classification along with comparative discussions among the reviewed studies. This study also presents a detailed literature review of ECG signal analysis and feature engineering for ECG-based body sensor networks in portable and wearable ECG devices for real-time cardiac status monitoring. Additionally, challenges and limitations are discussed and tools for research in this field as well as suggestions for future work are outlined. INDEX TERMS ECG analysis, cardiac arrhythmias, QRS and ST detection, ECG classification, deep learning.
Conference Paper
Full-text available
For electrocardiography (ECG) applications, gel dependant metallic electrodes such as Ag/AgCl are typically used, but these cause skin irritation and become dehydrated over time. To overcome these problems, a flexible electro-conductive textile material with a surface resistance of 332.5 Ohm/sq and resistivity of 6.6 - has been developed by coating PEDOT:PSS/PDMS on cotton fabric via flat screen printing. The coated fabric has been used to construct ECG electrodes and was compared with standard Ag/AgCl electrodes. An ECG waveform (with peaks P = 0.14 mV, QRS = 0.96 mV and T = 0.36 mV) has been collected with the textile-based electrodes during 3 minutes of static ECG measurement. The signal quality was comparable with the Ag/AgCl standard electrodes (P = 0.15 mV, QRS = 0.98 mV and T = 0.48 mV). The textile-based dry electrodes could potentially replace the gelled standard biopotential electrodes and avoid associated problems, especially for prolonged monitoring.
Full-text available
An electrocardiogram (ECG) records the electrical signal from the heart to check for different heart conditions, but it is susceptible to noises. ECG signal denoising is a major pre-processing step which attenuates the noises and accentuates the typical waves in ECG signals. Researchers over time have proposed numerous methods to correctly detect morphological anomalies. This study discusses the workflow, and design principles followed by these methods, and classify the state-of-the-art methods into different categories for mutual comparison, and development of modern methods to denoise ECG. The performance of these methods is analysed on some benchmark metrics, viz., root-mean-square error, percentage-root-mean-square difference, and signal-to-noise ratio improvement, thus comparing various ECG denoising techniques on MIT-BIH databases, PTB, QT, and other databases. It is observed that Wavelet-VBE, EMD-MAF, GAN2, GSSSA, new MP-EKF, DLSR, and AKF are most suitable for additive white Gaussian noise removal. For muscle artefacts removal, GAN1, new MP-EKF, DLSR, and AKF perform comparatively well. For base-line wander, and electrode motion artefacts removal, GAN1 is the best denoising option. For power-line interference removal, DLSR and EWT perform well. Finally, FCN-based DAE, DWT (Sym6) soft, MABWT (soft), CPSD sparsity, and UWT are promising ECG denoising methods for composite noise removal.
Full-text available
Wireless body area networks (WBANs) are increasingly used for remote healthcare surveillance in recent times, where electrocardiogram (ECG) signals are continuously acquired and transmitted to a base station or remote hospital for their storage and subsequent analysis. Multichannel ECG (MECG) is preferred over single-channel ECG as it provides more information from diagnostic point of view. One of the biggest challenges is to minimize the energy required for the WBAN network for continuous transmission of MECG data, which in turn demands for efficient data compression. Compressive sensing is an efficient signal processing tool for simultaneous compression and reconstruction of MECG data without visibly no or minimum loss of diagnostic information. In this paper, we propose an energy-efficient novel block-sparsity-based MECG compression scheme, which exploits both spatiotemporal correlation and multi-scale information of MECG data in the wavelet domain, effectively. Experimental results show that the proposed method outperforms other recently developed methods for MECG compression both qualitatively and quantitatively.
The Electrocardiogram (ECG) arrhythmia classification has become an interesting research area for researchers and developers as it plays a vital role in early prevention and diagnosis of cardiovascular diseases. In ECG signal classification, the feature extraction and selection processes are critical steps. Thus, in this paper, different ECG signal descriptors based on one-dimensional local binary pattern (LBP), wavelet, higher-order statistical (HOS), and morphological information are introduced for feature extraction. For feature selection and classification processes,a new hybrid ECG arrhythmia classification approach called MRFO-SVM that combines a metaheuristic algorithm termed Manta ray foraging optimization (MRFO) with support vector machine (SVM)is proposed to automatically determine the relevance features of LBP, HOS, wavelet and magnitude values. In MRFO-SVM approach,the MRFO is utilized to optimize the parameters of SVM and to select the significant features subset that provides the best classification performance, meanwhile SVM is used for classification purposes.The proposed MRFO-SVM approach is trained on the MIT-BIH Arrhythmia database containing four abnormal and one normal heartbeats. The experimental results of ECG arrhythmia classification using the proposed MRFO-SVM revealed with evidence its superiority with overall classification accuracy of 98.26% over seven well-known metaheuristic algorithms.
Cardiovascular diseases affect approximately 50 million people worldwide; thus, heart disease prevention is one of the most important tasks of any health care system. Despite the high popularity electrocardiography, superior automatic electrocardiography (ECG) signal analysis methods are required. The aim of this research was to design a new deep learning method for effectively classifying arrhythmia by using 2-second segments of 2D recurrence plot images of ECG signals. In the first stage, the noise and ventricular fibrillation (VF) categories were distinguished. In the second stage, the atrial fibrillation (AF), normal, premature AF, and premature VF categories were distinguished. Models were trained and tested using ECG databases publicly available at the website of PhysioNet. The MIT-BIH Arrhythmia Database, Creighton University Ventricular Tachyarrhythmia Database, MIT-BIH Atrial Fibrillation Database, and MIT-BIH Malignant Ventricular Ectopy Database were used to compare six types of arrhythmia. Testing accuracies of up to 95.3 % ± 1.27 % and 98.41 % ± 0.11 % were achieved for arrhythmia detection in the first and second stage, respectively, after five-fold cross-validation. In conclusion, this study provides clinicians with an advanced methodology for detecting and discriminating between different arrhythmia types.
Objectives With the technological advancements in the field of tele-health monitoring, it is now possible to gather huge amount of electro-physiological signals such as the electrocardiogram (ECG). It is therefore necessary to develop models/algorithms that are capable of analysing these massive amount of data in real-time. This paper proposes a deep learning model for real-time segmentation of heartbeats. Methods The proposed DENS-ECG algorithm, combines convolutional neural network (CNN) and long short-term memory (LSTM) model to detect onset, peak, and offset of different heartbeat waveforms such as the P-waves, QRS complexes, T-waves, and No waves (NW). Using ECG as the inputs, the model learns to extract high level features through the training process, which, unlike other classical machine learning based methods, eliminates the feature engineering step. Results The proposed DENS-ECG model was trained and validated on a dataset with 105 ECG records of length 15 min each and achieved an average sensitivity and precision of 97.95% and 95.68%, respectively, using a stratified 5-fold cross validation. Additionally, the model was evaluated on an unseen dataset to examine its robustness in QRS detection, which resulted in a sensitivity of 99.61% and precision of 99.52%. Conclusion The empirical results show the flexibility and accuracy of the combined CNN-LSTM model for ECG signal delineation. Significance This paper proposes an efficient and easy to use approach using deep learning for heartbeat segmentation, which could potentially be used in real-time tele-health monitoring systems.