J. Biomedical Science and Engineering, 2009, 2, 239-244
doi: 10.4236/jbise.2009.24037 Published Online August 2009 (http://www.SciRP.org/journal/jbise/
JBiSE
).
Published Online August 2009 in SciRes. http:// www.scirp.org/journal/jbise
Detection ischemic episodes from electrocardiogram
signal using wavelet transform
Mohammad Karimi Moridani, Majid Pouladian
Biomedical Engineering Department, Science and Research Branch, Islamic Azad University, Tehran, Iran
Received 19 November 2008; revised 16 March 2009; accepted 16 April 2009.
ABSTRACT
In this paper, we propose an algorithm for de-
tection of myocardial ischemic episodes from
electrocardiogram (ECG) signal using the
wavelet transform technique. The algorithm was
tested on data from the European ST-T change
database. Results show that this algorithm is
effective for distinguishing normal ECGs from
ischemic. We developed a method that uses
wavelets for extracting ECG patterns that are
characteristic for myocardial ischemia.
Keywords: Myocardial Ischemic; Wavelet Trans-
form; ECG
1. INTRODUCTION
The formulation and properties of an electrical impulse
through the heart muscle result in time-varying poten-
tials on the surface of the human body, which are known
as the ECG signals. The signal represents various activi-
ties of the heart [1]. Wavelet Transformation (WT) has
shown to be substantially noise-proof in ECG segmenta-
tion, and thus very appropriate for ST-T segment extrac-
tion (Figure 1). An initial downward deflection after the
P-wave is termed as, ‘Q’, the dominant upward deflec-
tion is the ‘R’ and the terminal part is denoted as ‘S’.
The T-wave represents ventricular recovery or repolari-
zation [1,2].
The ST segment usually merges smoothly and imper-
ceptibly with the T-wave.
Having a simple estimate of whether an ECG re-
cording which includes segments is characteristic for
ischemic heart or not, is one of the most interesting top-
ics for cardiologists.
Our aim in this research is to develop an algorithm
using WT for identification of myocardial ischemic epi-
sodes.
2. METHOD
For ECG parameters estimation, it is desirable that the
Figure 1. ECG waveform.
basis functions (wavelets), be symmetric/antisymmetric
Asymmetric basis will enable to detect the extreme of
wave’s peak [4].
In case of antisymmetric basis, the peak of the wave is
detected as a zero crossing.
In order to have a computationally simple analysis,
the peaks should be detected as zero crossings (which is
provided by antisymmetric basis) or local extreme
(symmetric basis) [4]. It is also desirable that basis have
a minimum number of sign changes. In practice, QRS
complex is usually considered to be symmetrical, while
T wave is less so. However, it has been shown, PR and
ST points can be estimated using biorthogonal wavelets
under the assumption of QRS complex and T wave
symmetry [2]. Moreover, having as an aim proposed
quantitative analysis, it is quite plausible to suppose ba-
sic QRS complex and T wave symmetry [4]. Biorthogo-
nal wavelets satisfy both these properties, so they are
used for DWT (discrete wavelet transform) decomposi-
tion. Hence used in this work. The filter coefficients of
both the symmetric low pass (LP) H and the antisym-
metric high pass (HP) filters G and K are given in Table
1 [2]. Decomposition and reconstruction filters satisfy
further equations:
1)()()( 2 wKwGwH (1)
where, w is frequency
piw wewH ))2/(cos()( 2
 (2)
)2/sin(4)( 2wiewGiw  (3)

240 M. K. Moridani et al. / J. Biomedical Science and Engineering 2 (2009) 239-244
SciRes Copyright © 2009 JBiSE
)(
)(1
)(
2
wG
wH
wK   (4)
1i and p=3,5,7,…
For compact support ‘p’ should be as small as possible.
Hence, it was chosen to be equal to 3. The filter coeffi-
cients corresponding to p=3 are given in Table 1.
2.1. Parameters Estimation
The parameters of the ECG signal are obtained by the
wavelet decomposition dyadic tree. This tree decom-
poses the signal initially into the smooth (low pass) and
detail (high pass) constituents [1]. The low pass compo-
nent is further decomposed into low pass and high pass.
This process is repeated over the desired number of
scales. When an ECG signal is passed through each of
the wavelet filters whose scales range from 21 to 24, as
shown in Figure 2, the detailed d1, k, d2 ,k, d3, k d4, k
and the smooth s1, k, s2, k, s3, k, s4, k filtered outputs
are obtained as shown in Figure 3. The following algo-
rithm is suggested for estimating the said ECG parame-
ters [1].
This initial filter is also based upon WT. The denois-
ing procedure is performed in three steps (MATLAB
wavelet toolbox) [2]:
1) decomposition with sym8 wavelet at level 5,
2) detailed coefficients threshold- choosing a thresh-
old for each level,
3) reconstruction.
In order to obtain ST and PR points’ values, each de-
composed ECG is segmented (Figure 1). At the first and
the second levels, QRS starting, ending and peak point are
Table 1. Filter coefficients of symmetric Low Pass (LP), anti-
symmetric High Pass (HP) and their reconstruction filters.
H G K L
0.1768
0.5303
0.5303
0.1768
0.3536
1.0607
-1.0607
-0.3536
0.1768
-0.5303
0.5303
-0.1768
-0.3536
1.0607
1.0607
-0.3536
G, decomposition high pass filter coefficients; H, decomposition low pass filter
coefficients; K, reconstruction high pass filter coefficients; L, reconstruction low
pass filter coefficients.
Figure 2. Wavelet decomposition of ECG signal. Decomposi-
tion per-formed over 4 scales.
extracted. Let us denote as -the R peak point, nl, and
n, 2 the beginning and ending points of QRS complex
respectively [2]. PR point is thus calculated as:
R
n
2
int 12 ll
R
nn
nPRpo   (5)
At the fourth level, ST-T characteristic points are ob-
tained: - the T peak point, nt1 and nt2 the beginning
and ending points of T wave, respectively [1]. ST point
is calculated as:
T
n
2
int 12 ll
T
nn
nSTpo   (6)
Wavelet decomposition introduces scale-dependent
phase delay into signals. For example, each zero cross-
ing point which corresponds to the peak of a symmetric
uniphase wave is delayed for exactly 121

J points,
where represents the scale [1].
2.2. ST Deviation Analysis
In our analysis, we have used European ST-T change
database with ischemic ECG signals sampled at 250 Hz,
two hours in duration each. The most elementary differ-
entiate/threshold algorithm has been applied to extract
each beat. Obtained signals were decomposed as de-
scribed in section 2.1 and for each beat/separate signal,
an ST deviation value was calculated as [2]:
int)(int)( STpoECGvaluePRpoECGvalue
nSTdeviatio


(7)
where ECG value(x) denotes an amplitude at the point x
of a given ECG.
Cases with specific ECG beats, that is to say with in-
distinguishable ECG features (thus impossible segmen-
tation), were excluded from further analysis. In this way,
an ST deviation value was obtained for each beat that
could be analyzed [2]. The final report might include the
comment that some beats were excluded from analysis
as well as the number of the beats omitted, and therefore
suggest a manual observation where required [2].
2.3. QRS Onset and Offset Detection
Some of the existing techniques use a series of band-
pass filters to extract QRS complexes from the ECG
signal, which under severe baseline drift and other high
frequency noises, fails to detect the characteristic points
to an acceptable accuracy. And some techniques use
neural network based adaptive identification algorithms
[3,5], which can be used for only a particular type of
pattern. The wavelet transform based technique can be
used to identify the characteristic points of the ECG sig-
nal to a fairly good accuracy, even with the presence of
severe high frequency and low frequency noises [6,7,8].
Our aim is to describe an elegant algorithm, which uses

M. K. Moridani et al. / J. Biomedical Science and Engineering 2 (2009) 239-244 241
SciRes Copyright © 2009 JBiSE
WT to identify the characteristic points of the ECG sig-
nal, and identifying the myocardial ischemic episodes.
As an alternative to the normal filtering techniques,
which use different narrow-band filters to extract the
frequency contents of the signal, wavelet transform
technique can be used [3]. In wavelet transform tech-
nique, the signal is analyzed at different frequencies with
different resolutions. It is called multi resolution analysis
(MRA) (Figure 4).
The wavelet used in this work is the quadratic spline
wavelet [9]. The reasons for choosing this particular
wavelet for the analysis purpose are as follows:
It has a very compact support and a generalized linear
phase, so there is a determinate relationship between
ECG characteristic points and the modulus maxima, or
the zero-crossing points of the WTs.
The Fourier transform (FT) of the quadratic spline
wavelet is given as:
4
)
4
4
sin
()( w
w
iww
 (8)
The FIR(Finite Impulse Response) filter coefficients
that make up the decomposition and reconstruction filter
banks and the Lipschitz coefficients for the decomposi-
tion algorithm are given in Tables 2 and 3 [3].
Figure 4. Block diagram of the entire process.
Table 2. FIR filter coefficients for quadratic spline wavelet.
N H G K L
-3
-2
-1
0
1
2
3
0.125
0.375
0.375
0.125
-2.0
2.0
0.0078125
0.054685
0171875
-0.171875
-0.054685
-0.0078125
0.0078125
0.046875
0.1171875
0.65625
0.1171875
0.046875
0.0078125
G, decomposition High Pass filter coefficient; H, decomposition Low Pass filter
coefficients; K, reconstruction High Pass filter coefficients; L, reconstruction
Low Pass filter coefficients.
Table 3. Normalization coefficients j

for the quadratic spline
wavelet.
j j

1
2
3
4
5
1.5
1.12
1.03
1..01
1.00
There is a relation between the characteristic points of
the signal and their WT at different levels [10,11,12].
For example, for the wave in Figure 5, the wavelet
transform at scale 21 is given.
The wave’s rising edge corresponds to a negative min-
ima and the dropping edge corresponds to a positive
maxima. The moduli of these maxima or minima corre-
sponding to the same edge are named as the modulus
maxima line. If the uniphase wave is symmetric to its
peak, then its peak corresponds to the zero-crossing
point of the positive maxima negative minima pair with
a delay of exactly 12 1

J points, where j represents the
scale [3]. After obtaining the wavelet transform coeffi-
cients at different scales, the next step is to find out the
ECG characteristic points from these coefficients [3].
2.4. T and P Wave Detection
T and P waves are normally low frequency, so WT at
high scale is used to locate these waves. In this work,
WT up to four scales are taken and the scale 24 is used to
locate T and P waves.
T wave creates a pair of modulus maxima with a dif-
ferent sign on W2jf (n) at scale, 24 within a time window
after the detected R-peak [3]. Since the wave is almost
symmetric to its peak, the peak of T-wave corresponds to
the zero crossing point of the modulus maximum pair
with a delay of 24-1-1 points. The peak, onset and offset
of the P-wave are detected in a way similar to those of
the T before the detected R-wave [3].
2.5. R-peak Detection
For detecting the R-peak, the modulus maxima–minima
pair is located at lowest scale, which is done by fixing a
threshold for detection [13]. So maxima–minima pairs
for other scales are located within the neighborhood of
these maxima-minima pairs. If the amplitudes of the
maxima-minima pairs, compared to those are at lower
scale, are consistent or increasing, the corresponding
Figure 5. The uniphase wave and its WT at scale 21.
Filter
Bank
Digitized
ECG WT Analysis
algorithm
ST-T
detection
Characteristic
Points
ST-T events

242 M. K. Moridani et al. / J. Biomedical Science and Engineering 2 (2009) 239-244
SciRes Copyright © 2009 JBiSE
modulus maxima-minima pair is treated as one that cor-
responds to a true R-peak [3]. This method reduces the
effect of low frequency artifacts and also the high fre-
quency non morphological noise.
2.6. Detection of MI from the ECG
Characteristic Point
Different ECG changes related to the evolution of
ischemia have been described, including T-wave am-
plitude changes, ST deviations and even alterations in
the terminal portion of the QRS complex [13]. Using
global representations for the ST–T complex instead of
a single point from the ST segment could better char-
acterizes ischemia patterns and yield better identifica-
tion of occluded artery [14,15]. The most important
ECG change associated with ischemia is the ST seg-
ment elevation or depression, with depression being
most common. Also, this can be along with T-wave
amplitude changes or even inversion [3]. So ischemia
can be detected using these two measurements. For
finding the ST depression level, a reference level is
found out at first. This is done by drawing a line be-
tween two or more P-waves where they return to the
base line (or starting of P-wave). From the character-
istic point detection algorithm, we obtain the P-wave
onset and offset of all the cardiac cycles. ST-segment
is the segment of ECG between QRS offset and T-wave
onset. The deviation of this segment from the reference
line is found out. The amplitude of T-wave is also
found by measuring the distance between T-peak and
reference line [3]. Having obtained these two values,
we can come to a conclusion, that whether the cardiac
cycle contains an ischemic episode or not.
3. INTERPRETING THE DATA
Two or more ST change episodes could be separated by
insignificant time gaps or with an unreadable ECG seg-
ment. This could lead to the conclusion that these epi-
sodes are separate, which is scarcely the case. Since the
duration of the episodes are of great importance for
making conclusions about myocardial ischemia, it was
necessary to alleviate such shortcomings [2]. Thus we
applied a low pass filter (Chebyshev Type I filter) to the
latter signal; thereby, constructed an envelope which was
informative enough [2].
Figure 6 shows typical results from the readings in a
patient with myocardial ischemia (manifested through
T-and ST-wave change) [2]. High values at vertical axis
suggest that a long ischemic episode has occurred (A
part, Figure 6). Within normal ECGs, we expect that
these values gather near zero. B part at Figure 6 is a
direct consequence of final low pass filtering and sug-
gests that a series of shorter, but still significant, ST epi-
sodes have occurred with unsubstantial time gaps among
them. B parts are not expected with normal ECGs.
Negative ordinate values are a co product of LP filtering
and bare no practical significance [2]. Beside ECG ana-
lyzed in Figure 6 which we considered to be a mid-case
in the terms of number of ST deviations, we selected two
more examples in order to demonstrate the efficiency of
the algorithm.
Figure 7 presents performance of the algorithm in the
case where there were only few ST changes, while Fig-
ure 8.
In Figure 7, we can clearly see that vertical axis
values are smaller, compared to those in Figure 6 or
Figure 8. Moreover, there are less high-valued ripples,
and they occur with significant time gaps. This sug-
gests that analyzed ECG is near normal in the terms of
ST change [2].
Figure 6. Number of ST deviations correlated with time of
consecutive appearances-function’s envelope.
Figure 7. Number of ST deviations correlated with time of
consecutive appearances-case with few ST-T deviations. (file
e115. dat in has been analyzed) [4].

M. K. Moridani et al. / J. Biomedical Science and Engineering 2 (2009) 239-244 243
SciRes Copyright © 2009 JBiSE
3.1. Estimating Algorithm’s Efficiency
Annotation files provided for ST-T change ECG signals
in contain information about ST and T-wave deviations
as well as the number of heart beat in that particular
ECG at which a change occurred [4]. ST changes are
annotated by values ‘18’ and T-wave changes by ‘19’.
In order to demonstrate similarity which exists between
the algorithm's outcome and manually annotated ST-T
changes, in Figure 6 and Figure 7 we present the manu-
ally annotated change/time dependence for the same ECGs
analyzed in Figure 7 and Figure 8, respectively [2].
Comparing Figure 7 and Figure 9, as well as Figure 8
and Figure 10, it shows the strong correlation between
automatic and manual ECG analysis. Direct relationship
exists between the time axes, too, but the algorithm's out-
come ‘compresses’ time axis due to the nature of the con-
structed array whose envelope is the actual final result [2].
Figure 8. Number of ST deviations correlated with time of
consecutive appearances-case with a considerable number
of ST-T deviations. (file E121.dat in has been analyzed) [4].
Figure 9. Manually annotated ST-T changes which corre-
spond to Figure 7. (file e115.atr in has been used) [4].
Figure 10. Manually annotated ST-T changes which cor-
respond to Figure 4. (file e115.atr in has been used) [4].
After such analysis, a physician can examine those
parts of ECG which were annotated for ST changes, if
necessary. Performance on annotating ST changes of the
proposed algorithm is the same as of, although different
wavelet base function had been used [2]. Demonstrated
performance has been verified on 10 ECGs from, and
confirms already described relationship between ECGs
automatically and manually analyzed for the stated pur-
pose [4].
4. RESULT
An array of ST deviations was transformed into an array
which correlates number of ST deviations with time of
consecutive appearances [2] to show that the algorithm
counts the number of consecutive appearances of ST
deviations greater than 0. l mV and uses that number as a
new array's element. The procedure is repeated for each
ST change episode.
ST–T change detection algorithm consists of calculat-
ing certain performance indices. They are: ST sensitivity
(STse), which is an estimation of the likelihood between
detecting an ischemic ST episode; ST positive predictiv-
ity (ST + P), an estimation of the likelihood that a detec-
tion is a true ischemic ST episode; ischemic sensitivity
(ISse), which is a fraction of true ischemia, and ischemic
positive predictivity (IS + P), which is the fraction of
detector annotated ischemia, and is true ischemia [3,16].
5. CONCLUSIONS
In this paper, we proposed an algorithm for the detection
of myocardial ischemic episodes from electrocardiogram
(ECG) signal using the wavelet transform technique.
The study is mainly aimed to use biorthogonal wave-
lets to estimate clinically significant parameters of ECG
waveforms and find out ST-T segment for detection
ischemic episodes. Comparing the measured ST and the
normal ST in test ECG provided by European ST data-

244 M. K. Moridani et al. / J. Biomedical Science and Engineering 2 (2009) 239-244
SciRes Copyright © 2009
base, the algorithm could found the exactly time, that ST
shape is changed into an abnormal shape. ST segment is
the most important diagnostic parameter to finding
myocardial ischemia therefore developed algorithm best
coincidence with the database occurs in the determina-
tion of the beginning of the ST episodes, and the worst
in the calculation of the maximum deviation. Neverthe-
less, this parameter is just used in the description of
ischemia episodes, and its importance is relatively low
with respect to the quantification of the number of epi-
sodes and their duration.
JBiSE
REFERENCES
[1] N. Sivannarayana, D. C. Reddy, (1999) Biorthogonal
wavelet transforms for ECG parameters estimation,
167-174.
[2] N. Milosavljevic and A. Petrovic, (2006) ST segment
change detection by means of wavelets, 137-140.
[3] P. Ranjith, P. C. Baby, and P. Joseph, (2003) ECG analy-
sis using wavelet transform: application to myocardial
ischemia detection, 44-47.
[4] “European ST-T change database,” www. physionet.org.
[5] N. Magaveras, T. Stamkopoulos, C. Pappas, and M. G.
Strintzis, (1998) An adaptive backpropagation neural
network for real-time ischemia episodes detection: De-
velopment and performance analysis using the European
ST-T database, IEEE Trans Biomed Eng, 45, 805-13.
[6] M. Vetterly and J. Kovacevic, (1995) Wavelet and sub-
band coding, First Edition, Upper Saddle River, NJ:
Prentice Hall.
[7] M. R. Raghuveer and S. Bopardikar, (1996) Wavelet
transforms: Introduction to theory and applications, Sec-
ond Edition, Boston, MA: Addison Wesley.
[8] (1996) Special issue on wavelets, Proc IEEE, 84(4).
[9] S. Mallat and S. Zhong, (1992) Characterization of sig-
nals from multiscale edges, IEEE Trans PAMI, 14(7).
[10] C. Li, C. Zheng, and C. Tai, (1995) Detection of ECG
characteristic points using wavelet transform, IEEE Trans
Biomed Eng, 42(1), 8-21.
[11] D. P. Golden, R. A. Wolthuis, and G. W. Hoffler, (1973) A
spectral analysis of the normal resting electrocardiogram,
IEEE Trans Biomed Eng, 20(5), 366-72.
[12] S. Mallat, (1991) Zero crossing of a wavelet transform,
IEEE Trans Inf Theory, 37(4).
[13] P. F. Cohn, (1993) Silent myocardial ischemia and infarc-
tion, 8–18, NewYork: Marcel Dekker, 87-97.
[14] F. Jager, G. B. Moody, and R. G. Mark, (1998) Detection
of transient ST segment episodes during ambulatory
ECG monitoring, Comput Biomed Res, 31(5), 305-22.
[15] J. Garcia, L Sornmo, S. Olmos, P. Laguna, (2000) Auto-
matic detection of ST–T complex changes on the ECG
using filtered RMS difference series: Application to am-
bulatory ischemia monitoring, IEEE Trans Biomed Eng;
47(9), 1195-201.
[16] F. Jager, G. B. Moody, A. Taddei, R. G. Mark, (1991) Per-
formance measures for algorithms to detect transient
ischemic ST segment changes. Computers in Cardiology,
Los Alamitos, CA: IEEE Computer Society Press, 372-96.