St variability assessment based on complexity factor using independent component analysis, hilbert transform and regression analysis

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 04 Issue: 10 | Oct-2015, Available @ http://www.ijret.org 467
ST VARIABILITY ASSESSMENT BASED ON COMPLEXITY FACTOR
USING INDEPENDENT COMPONENT ANALYSIS, HILBERT
TRANSFORM AND REGRESSION ANALYSIS
S. Thulasi Prasad1
, S. Varadarajan2
1
Assoc. Professor, Dept. of ECE, Sree Vidyanikethan Engineering College, Tirupati, India,stprasad123@yahoo.co.in
2
Professor, Dept. of ECE, S. V. University College of Engineering, Tirupati, India, varadasouri@gmail.com
Abstract
In recent days the computerized ECG has become the most effective and convenient diagnostic tool to identify cardiac diseases
such as Myocardial Ischemia (MI). Among the Cardio vascular diseases (CVDs) the Myocardial Ischemia (MI) is one of the
leading causes of heart attacks. The Myocardial Ischemia (MI) occurs due to the difficulties in the flow of the electrical impulses
from SA node to bundle branches because of the abnormalities in the conduction system. Normally the ECG is used as a main
diagnostic tool to identify the cardiac diseases. In order to obtain accurate information from ECG it is necessary to remove all the
artifacts and extract the pure ECG from noise background. In this paper the removal of the artifacts is achieved with linear
filtering and the extraction of the clean ECG signal is performed using Independent Component Analysis (ICA). After
preprocessing and ECG extraction, the QRS complex of each beat is detected by using Hilbert Transform and simple threshold
detection algorithm. Next the Instantaneous Heart Rate (IHR) from RR interval and Complexity Factor (CF) from time series ST
segment are computed for each beat to form desired feature sets. Later a linear regression model is designed using Instantaneous
heart rate (IHR) and ST segment Complexity Factors (STCFs) based on Linear Regression analysis. The proposed ICA-STCFR
model is used to identify the ischemic beats from the test feature sets of ECG signal to assess the ST-Segment Variability (STV).
The ECG data sets obtained from a local hospital were used to design and test the model. The evaluation parameters, Ischemic
Intensity Factor (IIF), Ischemic Activity Factors (IAF) and Peak to Average Value (PAV) were used to evaluate the proposed
method and compared with Wavelet Transform based method. The proposed ICA-STCFR was found to be yielding better results
than WT-ST method.
Key Words: Myocardial Ischemia, ICA, HT, QRS Complex, RR interval, ST segments, IHR, STCF, Scatter-plot
--------------------------------------------------------------------***----------------------------------------------------------------------
1. INTRODUCTION
In present situation, the bioengineering is becoming a vital
course in engineering courses as it can be used to solve a
number of difficulties which are being faced by the
researchers and doctors in diagnosing ailments. The Digital
Signal Processing (DSP) in bioengineering deals with the
analysis of signals such as filtering, smoothing, digitization,
feature extraction, prediction, in either discrete or
continuous time, to dig out useful information from those
signals etc.
The Electro Cardio Gram (ECG) is the deflection produced
by the electric charge picked up by a set electrodes placed
on the chest near the heart and limbs of the human body.
This gives a waveform representation of the electrical
activity of human heart. The deviations in the normal
electrical patterns indicate various cardiac disorders. During
the cardiac cycle of human heart the bioelectrical action
potentials are transmitted through the various parts of the
heart to perform contractions and relaxations of cardiac
muscles in a rhythmic way. These electrical potentials from
surface of the chest near the heart are recorded as an ECG
[1], [2]. It is used as a basic investigative and major
diagnostic tool by the cardiologists in cardiology to detect
and analyze the Heart Rate Variability, Auricular and
Ventricular Hypertrophy, Myocardial Ischemia and
Infarction (heart attack), Arrhythmias and Pericarditis [3],
[4]. These imperfections are reflected by the variation of the
intervals and magnitudes of the P wave, QRS complex, T
wave, PR segment, QT interval, ST interval, PR segment,
and ST segment of the ECG signal. The analysis of these
can help the doctors to diagnose human heart diseases
accurately. Fig-1 shows the ECG wave containing plenty of
information pertaining to the patient‟s heart. The ST
segment is the portion of the ECG wave that lies between
the J-point and the K-point [5]. This is normally flat in a
healthy subject. In case of defective subject ST segment is
not always flat instead it is elevated or depressed.
Fig-1: ECG with characteristic features

_______________________________________________________________________________________
The measurement of this deviation of ST segment by using
normally existing detection techniques is not always easy
and accurate as the non Gaussian noise is mixed with ECG.
The most commonly used method in the real world
applications are Pan-Tompkins QRS detection algorithm
and its modification introduced by Hamilton and Tompkins
[6], [7]. Pan-Tompkins and its modified Hamilton-
Tompkins algorithms are based on patient specific detection
threshold. These algorithms are well suitable to enhance
only the positions of QRS complexes and not the locations
of onset and off set of ST segment. This difficulty is
overcome by the Hilbert Transform and Independent
Component Analysis (ICA).
2. LITERATURE REVIEW
Shook et al (1989) performed analysis on HR and ST
variability to ECG beats for detecting ischemia. In this work
the maximum and mean ST deviation was measured by
taking deviation of ST segment as the deviation of ECG
signal from at a point that lies 80 ms after J from the
isoelectric line and also the average ST slope computed for
every 30 s of the ECG signal. Badilini et al (1992)
performed analysis on the ST segment in the frequency
domain and created a system to detect ischemia. Taddei et al
(1995) employed a geometric algorithm for estimation of the
ST segment from two lead ECG recordings by developing a
rule-based system. In this method it is assumed that an
ischemic episode contains only ischemic beats and the
information only from the ST segment was used. In their
analysis it was observed that the ischemic beats are rich in
lower frequencies than the normal ones.
Andreao RV et al (2004) proposed an ECG analysis system
based on Markov Model for beat segmentation and ischemia
detection. Ali Zifan et al. (2008) developed an automated
approach for annotating ECG‟s. This approach involves
segmentation of ECG using Adaptive piecewise constant
approximation (APCA) and Piecewise derivative dynamic
time warping (PDDTW). Single lead segmentation method
was used to segment ECG to reduce annotation time of
ECG‟s greatly instead of manual annotation.
Roshan Joy Martis et al (2009) used Gaussian mixture
model (GMM) to suggest an automatic classifier for
detecting cardiac irregularities from ECG. In this suggestion
GMM has been used for classification after estimating linear
prediction (LP) model, calculating residual error signal, and
registration of linearly independent ECG features using
Principle Component Analysis (PCA). Mohamed Ben
Messaoud et al. (2009) designed an algorithm to detect with
Atrial Fibrillation (AF) and ventricular tachycardia based on
P wave characteristics. This method involves recognition of
fiducial point such as R peaks by means of the Pan-
tompkins algorithm, segmentation of the P waves and
utilization of the correlation and regression to compute the
similarity factors between a test P wave segments and the
reference segments.
Mishra (2010) developed a classifier to classify ECG in to
arrhythmia and normal ECG using a local fractal dimension
based on nearest neighbor classification. Dusit Thanapatay
(2010) proposed a method using Principle Components
Analysis (PCA) and Support Vector Machines (SVM)
classifier to explore abnormalities from ECG in heart
functioning. Gu-Young Jeong (2010) proposed an algorithm
based on the polynomial approximation to identify the ST
level changes and then to classify them according to the ST
shape. In this research paper the ST shape is used as an
important diagnostic parameter for detecting myocardial
ischemia.
M. K. Islam et al. (2012) discussed in their published paper
the various digital signal processing techniques to analyze
ECG signals by means of MATLAB tools. They focused on
the study of simulation, acquisition, filtering and feature
extraction of real time ECG data. T Yingthawornsuk et al.
(2012) presented a paper on Classification of Cardiac
Arrhythmia based on SVM. In this paper Support Vector
Machine (SVM) classifier utilized to separate multi - class
arrhythmia dataset. Md. Ashfanoor Kabir et al. (2012)
carried out denoising of ECG signal using Empirical Mode
Decomposition (EMD) and Discrete Wavelet Transform
(DWT). The ECG signal was denoised in EMD analysis by
disposing the number of intrinsic mode functions (IMFs),
and denoised in wavelet transform domain by setting the
appropriate threshold value to decomposition levels.
Chetan. G et al (2014) proposed an adaptable PLI canceller
utilizing Least Mean Square (LMS) algorithm to remove
50/60 Hz powerline interference from electrocardiogram
(ECG).
3. METHODOLOGY
In this section the proposed method has been described with
help of the block diagram as shown in the Fig-2.
YES
NO
Testing
Feature sets
Preprocessing
QRS Detection
(Hilbert Transform)
Training
Feature sets
Threshold setting
IS
Ischemic
Beat
ECG Signals
ST feature extraction
(Complexity Factor)
Linear Regression
Modeling
ECG Extraction
(ICA)
Fig-2: Block diagram of ICA-STCFR modeling for ST
segment Variability Analysis

_______________________________________________________________________________________
This method consists of preprocessing, ECG signal
extraction, QRS detection, STCF feature extraction, STCFR
modeling and beat classification. This method requires the
ST segment Complexity Factors (STCF) and Instantaneous
Heart Rate (IHR) of each heart cycle in order to form feature
sets for the purpose of regression modeling between STCFs
and IHRs. Finally the Ischemic beats of the ECG signals are
detected using STCFR model.
3.1 Preprocessing
The noise and artifacts due to baselinewanders, power line
interference and motion artifacts are removed in
preprocessing by performing linear filtering on corrupted
ECG [8]. The preprocessing involves three steps: 1. High
frequency noise motion and artifacts removal; 2. Baseline
wanders removal and 3. Powerline interference removal.
The bandpass digital filter with the passband of 0.05 Hz to
40 Hz is used to remove high-frequency noise and the
baselinewanders from ECG signal [9]. There are several
techniques to remove powerline interference but here in this
paper a notch filter, designed for cutoff frequency of 50 Hz,
is employed for removing powerline interference noise from
ECG signal [10].
3.2 Independent Component Analysis
In this paper Independent Component Analysis (ICA) is
used to extract the ECG signal from 12 lead ECG by
reducing multidimensionality. It is one of the Blind Source
Separation Techniques available for extracting biosignals
from the noise background [11]. ICA is used to identify and
separate mixtures of sources with little prior information.
ICA is distinguished from other methods as it looks for
components that are both statistically independent and non
Gaussian [12]. The ECG is considered as a mixture of
signals from nodes presented in the heart, various artifacts
and noise [13], [14]. Based on this consideration, ICA is
applied on preprocessed ECG signal to separate the ECG
from other noise components effectively. Once the ECG is
extracted from noise and artifacts it is easy to perform
accurate estimation of fiducial points of ECG.
The principle of basic ICA structure is described in the
Fig-3. The vectors X and S represent the observed and
source signals respectively. A is an unknown matrix of size,
n x n, and called the mixing matrix. X and S get the size,
n x m, representing „n‟ number of sources and „m‟ number
of samples in each source.
ASX  (1)
Incidentally, the justification for the description of this
signal processing technique as blind is that we have no
information on the mixing matrix, or even on the sources.
Mixing Matrix
[A]
S1(t)
S2(t)
S3(t)
Sn(t)
Source
Signals
Un-Mixing Matrix
[W]
X1(t)
X2(t)
X3(t)
Xn(t)
𝑺 𝟏(𝒕)
𝑺 𝟐(𝒕)
𝑺 𝟑(𝒕)
𝑺 𝒏(𝒕)
Separated
Signals
Fig-3: Block diagram of Independent Component Analysis
The objective is to recover the original signals, S from only
the observed vector X. Denoting the output vector by V, the
aim of ICA algorithms is to find a matrix U to undo the
mixing effect. That is, the output will be given by
UXV  (2)
Where, V is an estimate of the sources. The sources can be
exactly recovered if U is the inverse of A after a permutation
and scale change. The required ECG can be reconstructed
by using (3), where „V’ is the matrix of derived Independent
Components (ICs) with the row representing the noise or
artifacts set to zero.
VUX 1
 (3)
The ICA methods try to estimate components that would be
as independent as possible and their linear combination
should be the original data. Normally the estimation of
components is done either by iterative algorithm, which
maximizes function of independence, or by a non-iterative
algorithm, which is based on joint diagonalisation of
correlation matrices [15]. ICA has one large restriction that
the original sources must be statistically independent. This is
the only assumption that we need to take into account in
general.
After getting ICs it is necessary to determine the order of the
independent components in order to identify normal ECG,
noise and abrupt alterations. As the ICs corresponding to
noise and abrupt alterations have more distinctive properties
than that of original signal both in time and frequency
domains, we may employ the statistical properties of these
waveforms to recognize the original ECG automatically
instead of identifying visually. A random variable for which
the PDF is Gaussian distribution can be characterized
sufficiently with the mean and variance. If the PDF of
variable is non-Gaussian, then the mean and variance are not
sufficient as there will be many different constituent signals
with the same mean and variance. In ICA one more moment
known as the kurtosis is defined in addition to the first two
well known moments of random variables, the mean and the
variance. The important metrics of ICA are discussed below
The mean (first central moment) of a random variable „X‟
(samples) is defined as;




1
0
1
][
N
i
iX x
N
XEm (4)
The variance (second central moment), of a random variable
„X‟ (samples) is defined as;
2
1
0
22
)(
1
])[( X
N
i
iXX mx
N
mXE 


 (5)
The kurtosis is the fourth moment (cumulant) of a random
distribution and defined as;
4
4
])[(
)(

XmXE
XKurt

 (6)

_______________________________________________________________________________________
224
)](3[)()( XEXEXKurt  (7)
The kurtosis is used to measure the non-Gaussianity of a
PDF by specifying the relative peaking or flatness of a
distribution with respect to a Gaussian distribution.
3.3 QRS Detection
The accurate ST segmentation requires an accurate QRS
complex detection. An efficient QRS detection method
using Hilbert Transform was employed to identify the
fiducial point precisely in each heartbeat [16], [17]. The
QRS detection procedure is described with help the block
diagram as shown in the Fig-4.
Derivative
Filter
Hilbert
Transform
R-peak
Detection
QRS
Detection
ECG
Signal
QRS
Complex
Fig-4 Block diagram of QRS detector
The QRS detection method used in this paper is a naval
combination of derivative filter and Hilbert Transform.
After the Hilbert Transform, R-peak is detected by setting a
threshold to envelope of the transformed and original
signals. Later QRS complex is recognized using the slope
property of the ECG signal. The first order derivative filter
is represented by;
 )1()1(
2
1
)(  nsns
T
ts
s
d (8)
where n = 2, 3, …, M − 1, M is the total number of samples
and Ts is the sampling time.
Hilbert Transform has an interesting property that the
Transformed version of the signal at the zero crossing of the
signal exhibits peaks. Once the R-peak is determined
accurately, the other peaks of QRS complex can easily
identified with reference to the R-peak using slopes of the
ECG signal. To get a signal with zero crossings at peaks of
original signal, first the original signal must be differentiated
with respect to time. Then the zero crossings of the
differentiated ECG signal will be located at the peak
positions of the ECG signal. From the Hilbert Transform of
the differentiated ECG signal the locations of R-peaks are
identified. Hilbert Transform is a simple, efficient and
accurate method of detecting QRS complex through
detection of R-peaks of ECG waves. The Hilbert transform
of real time function, s(t) is defined as




d
t
s
tsHts 




)(1
)]([()(ˆ (9)
The Hilbert transform can also be written as a convolution
between 1/t and s(t) as;
)(
1
)]([)(ˆ ts
t
tsHts 

(10)
The spectrum of the Hilbert transform provides a very useful
environment to analyze the signals is given by;
)]([
11
)](ˆ[ tsF
t
FtsF 







(11)
)]([
11
tsFdse
s
sj2π *
f








 



(12)
)()sgn(
)]([)sgn()()](ˆ[
fSfj
tsFfjfStsF H


(13)
From (13) it is clear that the Hilbert transform of s(t) is
obtained by filtering it through a linear filter with the
transfer function, -j[sgn(f)]. Therefore the Hilbert transform
is equal to – j for positive frequency and + j for negative
frequency. Hence it is equivalent to a filter, in which the
every frequency component is allowed to pass without any
attenuation but their phases are altered by - /2 radians or +
/2 radians depending upon the sign of the frequencies.
Finally the Fourier transform of the Hilbert transform of s(t)
is given by
)()]([)( fjStsFjfS H   (14)
The analytic signal z(t) can be formed from its Hilbert
transform as;
)()()( tjststz H (15)
The envelope of this analytical signal, |z(t)| is given by;
)()()( 22
tststz H (16)
Subsequently the polar form of the analytical signal is
expressed as;
)(
)()( tj
etBtz 
 (17)
By comparing equations (16) and (17) the envelope, B(t)
and phase of z(t) can be expressed as;
and
)()()( 22
tststB H (18)






 
)(
)(
tan)( 1
ts
ts
t H
 (19)
The derivative of the (t) is
)(
)]([
)()(
)]([
)]([
)( 2
tB
dt
tsd
tsts
dt
tsd
dt
td
t
H
H



 (20)
Since the envelope B(t) coincides with s(t) when sH(t) = 0
the maximum contribution of the envelope B(t) is obtained
when s(t) = 0. Thus the Hilbert Transform converts all zero
crossing points of the signal into dominant peaks.

_______________________________________________________________________________________
3.4 STCF feature Extraction
The ST segment is the wave segment between the J-point
and K-points in a time series of ECG wave [18]. First ST
segment is extracted from the identified J-point and
conventionally located K-point either at 40 ms (in case of
tachycardia) or at 80 ms after J-point (in case of
bradycardia). In this paper a simple algorithm is used to
extract ST segments and for the identification of isoelectric
line [15]. With forward search from S-peak within a window
of 100 ms J-point was recognized as one where the ECG
signal slope is either zero or changes its sign. With
backward search within window of 20 ms from Q-peak the
onset of Q wave and offset of P wave are recognized. From
the onset of Q wave and offset of P wave the PQ-segment is
identified and the mean amplitude of PQ-segment is
considered as isoelectric line reference. Once the J-point and
K-point are known the ST-Segments are extracted by taking
the time series of the ECG signal between J-point and
K-points in each beat and formed into ST segment feature
set [19]. Later the Complexity Factors of each ST segment is
computed using the equation (21) and populated into a
feature set of ST segment Complexity Factors (STCF) [20],
[21]. The Complexity Factor [22] of ST segment is defined
as;
2
2
''
CF
x
x


 (21)
Where 2
''
x
 , is the variance of second order derivative ST
segment and 2
x is the variance of the ST segment alone.
Simultaneously with STCFs the Instantaneous Heart Rate is
also computed from the QRS complexes by using (22) for
each beat. These two feature sets, IHRs and STCFs are used
for designing and testing the ICA-STCFR model.
bpm
intervalRR
rate (IHR)ous heartInstantane
60
 (22)
3.5 Linear Regression analysis
In Linear Regression Analysis (LRA) a linear equation is
fitted between two sets of data [23]. To establish a linear
regression between two sets of dada the association between
them must be strong enough. Normally a scatter-plot is used
to assess the strength of the association between any two
sets of data. Either increasing or decreasing trends in scatter-
plot represent some significant relation between the two sets
data [24]. If X and Y are two vectors then the values of „a‟
and „b‟ are to be found such that they satisfy the linear
equation Y'i = a + bXi. The Linear Regression results in
modeling confidents „a‟ and „b‟ as;




 2
)(
))((
XX
YYXX
b
i
ii
(23)
)( XbYa  (24)
Where X and Y are the mean values of feature sets. With
the modeling coefficients a regression line is designed using
the equation Yi = a + bXi and used as reference threshold for
detecting Ischemic beats. Finally the detected beats and the
STCF values of all the beats are used for performing ST
variability analysis with help of evaluation metrics like
Ischemic Intensity Factor (IIF), Ischemic Activity Factor
(IAF) and Peak to Average Value (PAV) of Complexity
Factors (CFs).
4. RESULTS
The ECG datasets acquired from a local hospital was used
for designing regression model and testing the proposed
method. The ECG data from the first four Leads are
preprocessed separately with a bandpass filter to remove
baselinewanders, motion artifacts and powerline
interference.
For the purpose of illustration the original ECG signals from
first four Leads corresponding to the patient ID: 078 and the
extracted ECG signal after preprocessing and Independent
Component Analysis are shown in the Fig-5 and Fig-6
respectively for few cycles.
0 1 2 3 4 5 6 7 8
-3
-2
-1
0
1
2
3
4
time (sec)
Normalisedandshiftedamplitude
Lead-I
Lead-II
Lead-III
Lead-avR
Fig-5: Four Leads of original ECG of the patient ID: 078
0 1 2 3 4 5 6 7 8
-4
-2
0
2
4
time (sec)
Normalisedandshiftedamplitude
IC1
IC2
IC3
Fig-6: ICs of the ECG Leads of the patient ID: 078
0 0.5 1 1.5 2 2.5 3 3.5 4
-1
-0.5
0
0.5
1
time (sec)
Magintude
Fig-7(a): ECG after Differentiation

_______________________________________________________________________________________
0 0.5 1 1.5 2 2.5 3 3.5 4
0
0.2
0.4
0.6
0.8
1
1.2
1.4
time (sec)
Magintude
Fig-7(b): ECG after Hilbert Transformation
0 0.5 1 1.5 2 2.5 3 3.5 4
0
0.2
0.4
0.6
0.8
1
time (sec)
Magintude
Fig-7(c): Envelope of ECG after Hilbert Transform
1 2 3 4 5 6 7 8
-0.5
0
0.5
1
ECG signal with QRS complex
time (sec)
Normalisedamplitude
ECG signal
Q-peak
R-peak
S-peak
Fig-7(d): QRS of the extracted ECG of the patient ID: 078
The ECG, after extracting from Independent Component
Analysis, is processed with QRS detection algorithm using
Hilbert Transform. This algorithm involves differentiation,
Hilbert Transform (HT), Envelop detection and QRS
Complex detection. For instance the results of series of
operations in QRS detection using HT are illustrated in the
Fig-7. Next the ST segment of each beat is extracted from
the extracted ECG. Later the ST segments Complexity
Factors (STCFs) of each beat are computed using the mean
amplitude of the PQ-segment as the isoelectric reference
level. From the RR intervals of each beat cycle the
Instantaneous Heart Rate (IHR) is computed using (12) for
each beat.
= 60/RRrate (IHR)ous heartInstantane bpm (25)
Both STCF levels and Instantaneous heart rates (IHRs) are
used as feature sets for the ST analysis. For instance the
IHR and STCF levels of ECG signal of the patient ID: 078
for first 30 beats are tabulated in the Table-1. A threshold is
derived from the Linear Regression model that is designed
between IHR and STCF levels from the Table-1 to derive
design coefficients. Theses model coefficients are used to
set a threshold for beat classification. The regression model
threshold is set from the modeling coefficients as visualized
in the Fig-8. This regression line is used as a reference
threshold to compare the STCFs of the testing feature sets
Table-1: IHRs and STCFs of the ECG of patient ID: 078
Beat
No
IHR
(bpm)
STCF
Beat
No
IHR
(bpm)
STCF
1 54 0.0090 16 78 0.2575
2 45 0.0100 17 77 0.0636
3 79 0.1649 18 77 0.1163
4 78 0.2575 19 78 0.0686
5 77 0.0636 20 102 0.0352
6 77 0.1163 21 79 0.1649
7 78 0.0686 22 78 0.2575
8 102 0.0352 23 77 0.0636
9 79 0.1649 24 77 0.1163
10 78 0.2575 25 78 0.0686
11 77 0.0636 26 102 0.0352
12 77 0.1163 27 79 0.1649
13 78 0.0686 28 78 0.2575
14 102 0.0352 29 77 0.0636
15 79 0.1649 30 77 0.1163
75 80 85 90 95 100 105 110
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
IHR (bpm)
STCF
STCF
RegressionLine
Fig-8: Linear Regression between STCFs with IHRs
and the decision is made in favorer of probable Ischemic
beats if the value of STCF of the testing feature set is more
than that obtained from the regression model. Later with a
hard threshold final and true Ischemic beats are separated
from the probable Ischemic beats. The separated Ischemic
beats are populated for further ST variability analysis. Using
(13), (14) and (15) the parameters, Ischemic Intensity
Factor, Ischemic Activity Factor (per 30 beats) and Peak to
Average Value (PAV) respectively are evaluated and the ST
variability is analyzed.
T
Isc
N
N
IAFFactorIntensityIschemic )( (26)
avgIsc CFNIAFFactorActivityIschemic *)(  (27)
avg
peak
CF
CF
PAVValueAveragetoPeak )( (28)
Where NIsc is the number of detected Ischemic beats, NT is
the total number of beats, CFavg is the average STCF level
and CFpeak is the peak value of the respective ECG signal
ST segments.

_______________________________________________________________________________________
Table -2: Performance Evaluation and Comparison
Patient ID 57 77 78 1076 1077 1115
WT-STmethod
No of
Ischemic
beats
2 2 3 3 5 7
Ischemic
Intensity
Factor
(IIF)
0.067 0.067 0.100 0.100 0.167 0.233
Ischemic
Activity
Factor
(IAF)
0.003 0.001 0.004 0.003 0.055 0.016
Peak to
Average
Value
(PAV)
5.18 9.24 5.46 6.40 2.569 6.47
ICA-STCFRmethod
No of
Ischemic
beats
12 10 5 9 11 9
Ischemic
Intensity
Factor
(IIF)
0.400 0.333 0.167 0.300 0.367 0.300
Ischemic
Activity
Factor
(IAF)
0.084 0.065 0.019 0.039 0.043 0.024
Peak to
Average
Value
(PAV)
1.54 1.97 2.24 2.49 2.64 1.55
For the patient ID: 078 the parameters the IIF, IAF and PAV
are evaluated as 0.167, 0.019 and 2.24 respectively for ICA-
STCFR method. Similarly these parameters are evaluated
for all the six patients‟ ECG data using both WT-ST method
and ICA-STCFR methods and tabulated the results in the
Table-2.
057 077 078 1076 1077 1115
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Patient ID
IIF
WT-ST method
ICA-STCFR method
Fig-9: Patient wise Ischemic Intensity Factor
057 077 078 1076 1077 1115
0
0.02
0.04
0.06
0.08
0.1
Patient ID
IAF
WT-ST method
ICA-STCFR method
Fig-10: Patient wise Ischemic Activity Factor
Fig-9 and Fig-10 shows the patient wise IIF and IAF
respectively for all six patients according the Table-2 data.
From the Table-2 and the Fig-9 and Fig-10 it is obvious that
the ICA-STCFR (Independent Component Analysis – ST
segment Complexity Factor Regression modeling) method
of ST variability analysis is yielding higher Ischemic
Intensity Factor (IIF), Ischemic Activity Factor (IAF) and
Peak to Average Values (PAV) than the Wavelet Transform
(WT-ST) based method.
5. CONCLUSIONS
A new method based on ST segment Complexity Factor
(STCF) feature, Independent Component Analysis, Hilbert
Transform and Linear Regression is proposed in this paper.
The Complexity Factor is an effective and suitable feature
for detection of myocardial ischemia. The Independent
Component Analysis greatly reduces the effect of noise
present in the passband of the ECG and increases the Signal
to Noise Ratio than other methods as the ECG is extracted
by reducing the multi-dimensionality of ECG leads through
independence. The Complexity Factor improves the
accuracy of Ischemic detection by taking the variance of the
ST segment into account rather than mean of ST levels as in
WT-ST method. The regression threshold decreases the
probability of missing Ischemic beats and increases the
classification accuracy. The performance of the model was
evaluated with Ischemic Intensity Factor (IIF), Ischemic
Activity Factor (IAF) and Peak to Average Value (PAV) by
employing the designed model on the ECG signals acquired
from a local hospital. The proposed method is giving better
results compared to the WT-ST method for all the patients.
Thus the ICA-STCFR modeling for analysis of ST
variability will support the doctors significantly to recognize
the manifestation of Ischemia and diagnose the cardiac
diseases accurately.
REFERENCES
[1]. Coumel P. and Garfein O, “Electrocardiography: Past
and Future”, Annuals of the New York, Academy of
Sciences, Vol. 601, New York, 1990.
[2]. J.R. Hampton, “The ECG Made Easy”, International
Edition (English), 8th
Edition, Elsevier Health Sciences,
2013.
[3]. N. Goldschlager, “Principles of Clinical electrocar-
diography”, 13th
edition, ISBN 978-083-8579-510,
Connecticut, USA.
[4]. G.-Y. Jeong, K.-H Yu, “Design of Ambulatory ECG
Monitoring System to detect ST pattern change,” SICE-
ICASE International Joint Conference, Oct 2006.
[5]. Li, C., Zheng, C., Tai, C., “Detection of ECG
characteristic points using wavelet transforms”, IEEE
Trans. Biomed. Eng. 1995; 42:21-28.
[6]. J. Pan and W. J. Tompkins, “A real-time QRS detection
algorithm", IEEE Trans. Biomed. Eng., vol. 32, pp.
230-236, March 1985.
[7]. P. S. Hamilton, W. J. Tompkins, “Quantitative
Investigation of QRS Detection Rules Using MIT/BIH
Arrhythmia Database”, IEEE Transactions on

_______________________________________________________________________________________
Biomedical Engineering, Vol.31, No.3, pp. 1157-1165,
ISSN 0018-9294, March 2007.
[8]. Dr. Christian, “A DSP Practical Application: Working
on ECG Signal, Applications of Digital Signal
Processing”, Cuadrado-Laborde (Ed.), ISBN: 978-953-
307-406-1.
[9]. Kaur, M., Singh, B., Ubhi, JS and Seema Rani, “Digital
Filtration of ECG Signals for Removal of Baseline
Drift”, International Conference on Telecommunication
Technology and Applications Proc .of CSIT vol.5,
pp: no.105-109, 2011
[10]. Zarzoso V, Comon P, “Robust independent component
analysis for blind source separation and extraction with
application in electrocardiography”, In: Proc. EMBC-
2008, 30th Annual International Conference of the
IEEE Engineering in Medicine and Biology Society,
Vancouver, BC, Canada, 3344–3347, 2008.
[11]. P. S. Hamilton, “A comparison of adaptive and
nonadaptive filters for the reduction of powerline
interference in the ECG”, IEEE Trans. Biomed. Eng,
1996; 43:105–109.
[12]. M.P.S. Chawla, “Detection of indeterminacies in
corrected ECG signals using parameterized
multidimensional independent component analysis”,
Computational and Mathematical Methods in Medicine
Vol. 10, No. 2, 85–115, June 2009.
[13]. J. F. Cardoso, “Blind signal separation: Statistical
principles,” Proc. IEEE, vol. 86, pp. 2009–2025, Oct.
1998.
[14]. P. Comon, “Independent component analysis, a new
concept?,” Signal Processing, vol. 36, no. 3, pp. 287–
314, 1994.
[15]. K. Abed Meraim, M.G. Amin, and A.M. Zoubir, “Joint
anti-diagonalization for blind source separation”, Proc.
ICASSP-01, 2001.
[16]. D. Benitez, P. Gaydecki, A. Zaidi, and A. Fitzpatrick,
“A new QRS detection algorithm based on the Hilbert
transform,” in Computers in Cardiology 2000, 2000,
pp. 379-382.
[17]. S. L. Hahn, “Hilbert transforms,” in The Transforms
and Applications Handbook (A. Poularakis, Ed.), Boca
Raton FL: CRC Press, 1996, ch. 7.
[18]. X. Li, et al., “ST-T complex automatic analysis of the
electrocardiogram signal based on wavelet transform,”
Proc. 29th IEEE Bioengineering, pp. 144-145, 2003.
[19]. Lehtinen, R., “ST/HR Hysteresis: Exercise and
Recovery Phase ST Depression/Heart Rate Analysis of
the Exercise ECG,” J. Electrocardiol., No. 32, 1999,
pp. 198–204.
[20]. Hjorth, B., “The physical significance of time domain
descriptors in EEG analysis”, Electroenceph. Clin-
Neurophysiol.1973, 34:321–325. 4.
[21]. Hjorth, B., “EEG analysis based on time domain
properties”, Electroencephalography and Clinical
Neurophysiology, vol. 29, no.3, pp. 306-310,
September-1970.
[22]. Hjorth, B., “Time domain descriptors and their relation
to a particular model for generation of EEG
activity”, In: G. Dolce, H. Künkel (Eds.) CEAN:
Computerized EEG Analysis. Gustav Fischer, Stuttgart;
1975:3–8.
[23]. Barlow, Jesse L., “Numerical aspects of Solving Linear
Least Squares Problems”, In Rao, C.R. Computational
Statistics, Handbook of Statistics, 9, North
Holland. ISBN 0-444-88096-8:1993.
[24]. Edwards, A. L., “An Introduction to Linear Regression
and Correlation”, San Francisco, CA: W. H. Freeman,
pp. 20-32, 1976
BIOGRAPHIES
S. Thulasi Prasad received B.Tech degree
from Sri Venkateswara University,
Tirupati, Andhra Pradesh, India in
Electronics and Communication
Engineering in 1996, M.Tech degree in
Engineering from Sri Venkateswara University, Tirupati,
Andhra Pradesh, India in 2003. He has 17 years of teaching
experience and 2 years of industrial experience. Presently he
is working as Associate Professor in the Dept. of ECE in
Sree Vidyanikethan Engineering College, Tirupati, Andhra
Pradesh, India
Dr. S. Varadarajan received B.Tech degree
from Sri Venkateswara University,
Tirupati, Andhra Pradesh, India, in
Engineering in 1987, M.Tech degree from
Regional Engineering College, Warangal,
Andhra Pradesh, India in 1993 and Ph.D from Sri
Venkateswara University, Tirupati, Andhra Pradesh, India in
2005. Presently he is working as Professor in the
Department of Electronics and Communication Engineering,
Sri Venkateswara University College of Engineering,
Tirupati. His research areas are Signal and Image
Processing, Radar Signal Processing and Digital
Communications. He is a member of IETE, ISTE and IEEE.
He is the chairperson of IETE Tirupati centre. He has
teaching experience of 23 years.

St variability assessment based on complexity factor using independent component analysis, hilbert transform and regression analysis

More Related Content

What's hot (18)

Similar to St variability assessment based on complexity factor using independent component analysis, hilbert transform and regression analysis (20)

More from eSAT Journals (20)

Recently uploaded (20)

St variability assessment based on complexity factor using independent component analysis, hilbert transform and regression analysis