iMedPub Journals

2015

International Archives of Medicine

http://journals.imed.pub

Section: Translational Cardiology

ISSN: 1755-7682

Vol. 8 No. 62
doi: 10.3823/1661

Personalized Reduced 3-Lead System formation

methodology for Remote Health Monitoring
Applications and Reconstruction of Standard
12-Lead system
Original

Abstract
Remote Health Monitoring (RHM) applications encounter limitations
from technological front viz. bandwidth, storage and transmission
time and the medical science front i.e. usage of 2-3 lead systems
instead of the standard 12-lead (S12) system. Technological limitations
constraint the number of leads to 2-3 while cardiologists accustomed
with 12-Lead ECG may find these 2-3 lead systems insufficient for
diagnosis. Thus, the aforementioned limitations pose self-contradicting challenges for RHM. A personalized reduced 2/3 lead system is
required which can offer equivalent information as contained in S12
system, so as to accurately reconstruct S12 system from reduced lead
system for diagnosis.
In this paper, we propose a personalized reduced 3-lead (R3L)
system formation methodology which employs principal component
analysis, thereby, reducing redundancy and increasing SNR ratio, hence, making it suitable for wireless transmission. Accurate S12 system is
made available using personalized lead reconstruction methodology,
thus addressing medical constraints. Mean R2 statistics values obtained for reconstruction of S12 system from the proposed R3L system
using PhysioNets PTB and TWA databases were 95.63% and 96.37%
respectively. To substantiate the superior diagnostic quality of reconstructed leads, root mean square error (RMSE) metrics obtained upon
comparing the ECG features extracted from the original and reconstructed leads, using our recently proposed Time Domain Morphology
and Gradient (TDMG) algorithm, have been analyzed and discussed.
The proposed system does not require any extra electrode or modification in placement positions and hence, can readily find application
in computerized ECG machines.

Under License of Creative Commons Attribution 3.0 License

Sidharth Maheshwari1,
Amit Acharyya1,
Michele Schiariti2,
Paolo Emilio Puddu2
1 Department of Electrical Engineering, Indian
Institute of Technology Hyderabad.
2Department of Cardiovascular Sciences,
Sapienza University of Rome, Italy.

Contact information:
Amit Acharyya.

amit_acharyya@iith.ac.in

This article is available at: www.intarchmed.com and www.medbrary.com

International Archives of Medicine

Section: Translational Cardiology

ISSN: 1755-7682

Introduction
Advanced wireless technology, high speed internet
facility and availability of other communication systems can be used to provide the accessibility of state-of-the-art healthcare facilities to the patients in
remote and rural areas for monitoring and diagnosis
of cardiovascular diseases (CVD), one of the prime
causes of human mortality today. Tremendous and
elaborative research regarding the prevention of
CVD is being conducted by various researchers [1].
The recent advances in telemonitoring has enabled
Home Monitoring [2] and has significantly enhanced
the comfort-level of patients. These are, however,
limited to urban areas as home monitoring would
require availability of state-of-the-art health center
nearby and supervision of cardiologists. Remote
areas, generally, experience scarcity of both facilities
and experienced practitioners, hence, remote health
monitoring (RHM) can find immense application in
diagnosis and treatment of large population in remote areas.
RHM, generally, involves acquisition and wireless
transmission of ECG signals for diagnosis which posits technological constraints viz. bandwidth, storage and transmission time [3, 4, 5]. From the medical
science perspective, cardiologists have been accustomed to standard 12-lead (S12) system due to its
widespread usage since decades and S12 system
being used as routine test for CVD diagnosis. The
presence of multiple channels in S12 restricts its usage in RHM applications [5], which preferably employs
a reduced lead system with 2 or 3 leads. Numerous
signal compression techniques are available that are
employed prior wireless transmission to reduce the
data size and decompress at the receivers end to
produce the originally sent signal. However, it has
been reported that compression ratio and efficiency of signal compression algorithms decreases with
the increase in the number of channels [6]. Hence,
multichannel S12 system is discouraged for telemonitoring applications, though, advanced compression algorithms are available. The aforementioned

2015
Vol. 8 No. 62
doi: 10.3823/1661

arguments posits a requirement for an appropriate

reduced lead system for RHM scenario and availability of S12 leads for better diagnosis of CVD.
We envisage a scenario where the ECG acquired in a remote area needs to be transmitted to
nearby state-of-the-art health center. Remote areas
lack in facilities and skilled caretakers. In such a scenario, a modification in reduced lead system used
with or without standard electrode positions will
not be practical. A reduced lead system with nonstandard placement of electrodes is unlikely to provide sufficient information to cardiologists. On the
other hand, reduced lead systems such as EASI with
3-leads and standard electrode positions, similar to
Frank Vectorcardiographic (FV) system [7], would
require training of caretakers to acquire the skill of
using such systems. It has also been found to contain insufficient information for diagnosis [8]. Such
additional limitations from the RHM scenario can
be allayed by using S12 system, as it is the clinical
standard.
In this paper, we present a solution to the aforementioned limitations posed from technological and
medical science front by proposing a personalized
reduced 3-lead (R3L) system formation methodology. The proposed R3L system is derived from S12
system itself, thereby avoiding any extra electrode,
hardware or skillset requirements. At the receivers
end, we reconstruct S12 leads from proposed R3L
system accurately and reliably in a personalized
manner, which will help cardiologists in accurate
diagnosis.

Previous work and background

Existing lead systems and reconstruction
methodologies
Reconstruction of S12 system from FV system was
proposed by Dower et al [9] (Dower Transform) where the transformation matrix was obtained using the
human torso model assuming a homogeneous medium. It is also known as universal transformation
This article is available at: www.intarchmed.com and www.medbrary.com

International Archives of Medicine

Section: Translational Cardiology

ISSN: 1755-7682

matrix as it is patient independent. A population

based methodology was proposed by Dawson et al
[10] (Affine Transform), which was obtained using
linear regression technique when applied on a set of
patients. It was shown to outperform Dower transform in its accuracy to reconstruct S12 leads. Recently, a personalized methodology was proposed
which has been shown to outperform both Dower
and Affine transforms [11]. Though, the accuracy
obtained from the personalized method is superior
and sufficient for all practical purposes, the Franks
system, being a 3-lead system, cannot be used for
RHM applications due to lack of dedicated acquisition system and uncomfortable electrode placement
positions viz. behind the neck and posterior region.
There are several investigations on reconstruction
of missing precordial leads from the remaining leads
[12-15]. Three or four lead subsets e.g. I, II, V2 or
I, II, V2, V5, using patient-specific or populationspecific transformation methodologies have been
evaluated. These basis leads in the aforementioned
methodology is a subset of S12 system, hereafter
will be referred as S12S system and hence, are readily available without any extra hardware or requirement. However, Maheshwari et al [15] have shown
that such systems suffer from proximity effect and
do not produce robust results compared to that obtained from Franks leads. EASI system is a 3-lead
system having electrode positions inspired from FV
system, thus having standard positions, however,
they have been found to contain insufficient information [8] and reconstruction of S12 from S12S has
been shown to outperform reconstruction S12 from
EASI.
Trobec and Tomasic [17] proposed an algorithm
for selection of new lead system for wireless applications and it was shown to perform better than
EASI system in reconstruction of S12 system. Finlay
et al [18] proposed a new lead system using PCA
for smart textiles [19] to increase the SNR and information content. The aforementioned lead systems
involves multiple channel ECG acquisition and un Under License of Creative Commons Attribution 3.0 License

2015
Vol. 8 No. 62
doi: 10.3823/1661

conventional electrode placement positions, hence,

adoption of such systems would require further indepth investigation. Recently, Tsouri and Ostertag
[20] have used two reduced 3-lead systems viz. I, II,
V2 (S12S) and Franks XYZ leads, and reconstructed
S12 systems by employing independent component
analysis (ICA). However, as reported by the authors
that the proposed methodology is adaptive, the results still deteriorate as the recording progresses and
forthcoming QRS complexes are reconstructed. The
authors have majorly concentrated on QRS complex which might not be sufficient in diagnosis of
diseases whose biomarkers also depend upon other
ECG features. ICA also encounters the algorithmic
difficulty of order in which the independent components are obtained and no definite method to solve
this limitation has been proposed by the authors.
We think that further investigation is needed to
measure the applicability of ICA in lead reconstruction.

Previous works on lead reconstruction for

RHM application
None of the aforementioned methodologies aimed
at finding application in remote health monitoring.
In our previous works [11, 15, 21, 22], we have focused on this applications of lead reconstruction in a
RHM scenario. All possible combinations of subsets
of 3-leads of S12 system (S12S systems) have been
used to reconstruct the missing leads using personalized methodology and have been shown to produce superior results compared to other investigations
by various researchers. The S12S systems were also
compared with Franks system. It was found that
the orthogonal leads of FV system produce robust
and reliable results compared to S12S systems and
are unaffected by proximity effect.
The major thrust till now has been on accurate and reliable lead reconstruction methodology.
However, RHM applications require lead systems inclined towards signals with characteristics desired by
wireless technology today i.e. high signal-to-noise

2015

International Archives of Medicine

Section: Translational Cardiology

ISSN: 1755-7682

(SNR) ratio [18], high compression ratio [6] etc. In

this paper, we have proposed a personalized lead
system formation methodology targeting the technological requirements of remote monitoring.

Material
In this paper, two databases have been used from
PhysioNet [23]: PTBDB [24] and TWADB [25]. PTBDB
is a 290 patient 15 lead database with both S12 and
FV system simultaneously acquired and digitized at
the sampling frequency of 1 kHz. First reading of
all the 290 patients were used though some of the
patients had extreme artifacts and paced rhythm.
TWADB consists 100 records digitized at a sampling
rate of 500 Hz. TWADB contains standard 12-Lead
ECG and some 2 or 3 signal recordings. Out of these
only 72 patients had all the 12-leads of S12 recorded
and hence, these patients were used in the study.

Vol. 8 No. 62
doi: 10.3823/1661

Proposed methodology
Fig. 1 shows summary of the proposed methodology. We envisage a scenario where Standard 12-lead
ECG is being acquired in a remote or hospital-based
environment. Generally, in a remote health monitoring scenario, these leads are needed to be transmitted to nearby state-of-the-art health center for
diagnosis, storage and updating of patient's health
record. At the transmission end, the conventional
S12 acquisitions system is used to capture the ECG.
Eight independent leads of the S12 system are then
passed through the Lead Component (LC)1 system
formation module (Section 3.2.1). Using LC and S12
system together the transformation coefficients are
obtained by employing least-square (LS) fit method
(Section 3.2.2). Then, LC leads along with the transformation coefficients can be transmitted and eventually 8 independent leads of the S12 system can
be reconstructed for diagnosis at the receiver's end
in an optimum way.

Figure 1: Summary of proposed methodology. The figure has been adopted from [22] and modified.

1. The proposed reduced 3-lead system will be referred as Lead Component (LC) system throughout the paper.

This article is available at: www.intarchmed.com and www.medbrary.com

International Archives of Medicine

Section: Translational Cardiology

ISSN: 1755-7682

Reduced 3-Lead (R3L) system formation

The proposed R3L has been obtained upon employing PCA on 8 independent leads of S12 system.
PCA is a widely used technique for noise removal
and dimensionality reduction and thus, has found
applications in data compression [6], denoising [26]
and reducing information redundancy [27]. It is a
linear algebra technique used to find the most appropriate basis, called principal components, along
which when the signal is represented has a minimum covariance. With zero co-variance, the principal components (PC) are efficient representation
of the set of signals. These PCs are orthogonal to
each other and each have variances associated with
them. It is often found that most of the information
carried by the original set of signals can be found
in the first k principal components. The value of k
depends on the application.
The number of components required for our application has to be determined using Heart-Vector Projection (HVP) theory [9, 28, 29], which states that
heart can be approximated as single dipole vector,
known as heart vector H, fixed in 3-D space whose
orientation and magnitude varies during a cardiac
cycle. This dipole vector is responsible for the body
surface potential observed when electrodes are
placed on the body. The potential at any point on
the body is the projection of H on the lead vector
(L ) which is assumed to originate from the zeropotential region in the heart and terminates on the
point of location of electrode on the body, which
mathematically can be seen in equation (1).

2015
Vol. 8 No. 62
doi: 10.3823/1661

rive back the orthogonal components of the heart

vector. We have employed PCA in this work, as it
results into orthogonal components and selected
first three (k = 3) principal components viz. PC1,
PC2 and PC3 with the intention to derive the three
orthogonal heart-vector components. It should be
noted that these three leads of the LC system may
not correlate to FV leads [28], however, they can still
be assumed to be less accurate form of heart vector
components in directions different from those of
Frank leads. As mentioned before, in our previous
work [15] we have found FV system to produce
consistent and robust reconstruction results compared to other 3-lead systems, it has motivated us
to propose an orthogonal 3-lead system or the Lead
Component system.
Our objective of using PCA is to reduce the inherent redundancy among the leads of S12 system.
We start with the hypothesis that the first three
principal components carry most of the information needed for diagnosis. The plausibility of this
hypothesis is measured by reconstruction of S12
leads from the proposed Lead Components (LC)
system and its comparison with the originally measured ECG signal. Eq. 2 shows the way in which LC
system is formed for every patient. PCA is applied
on the 8 independent leads of every patient and
a personalized LC system is obtained. For further
details on PCA interested readers may refer to [30].

Where H= X + Y + Zk and L =a + b + ck . From

this we can easily infer that all leads/channels/potential of S12 system are projection of heart vector
on respective lead vectors and can be decomposed
into three orthogonal components. Hence, theoretically from the standard leads it is possible to de Under License of Creative Commons Attribution 3.0 License

Preprocessing module
The preprocessing module comprises of baseline
wandering (BW) removal based on discrete wavelet
transform (DWT) [31] and denoising based on trans-

2015

International Archives of Medicine

Section: Translational Cardiology

ISSN: 1755-7682

lation invariant wavelet transform (TIWT) [32]. The

snippet of the MATLAB code for implementation of
the preprocessing module and its methodology can
be found in [21].

Personalized transformation coefficient

generation
The generalized form of (1) can be found in (3), where l1, l2 and l3 are any three leads and a, b and c are
the corresponding coefficients (lead vector components) for the electrode placed to record potential V.
3

Using the linearity of (3), the transformation

coefficients viz. a, b and c can be calculated using
least-square fit technique, provided V, l1, l2 and l3
are known, using eq. 4. Least square fitting is an
efficient standard linear regression technique available for linear transformations, where one set of
values can be transformed to another with minimum error. Such linear systems can be represented
as (Ax=b), where the matrix A is the transformation matrix. To obtain the matrix A, we minimize
the Euclidean norm of ||Ax-b||22 using differential
equations which produces an analytical solution of
A=(XTX)-1XTb. Eq. (4) is the simple representation
of this analytical solution for our case.

For the reconstruction of S12 from proposed LC

system, V can be replaced by the S12 leads and
l1, l2 and l3 can be replaced by PC1, PC2 and PC3
in (4) to obtain the corresponding transformation
coefficients. In many previous works [12-15], S12S
system consisting of leads I, II and V2 have been
used for reconstruction of missing precordial leads
of S12 system and it is also the most often used ba-

Vol. 8 No. 62
doi: 10.3823/1661

sis lead set, hence, we compare the reconstruction

results of our proposed LC system with aforementioned S12S system. To do so, V will be replaced
by missing precordial leads and l1, l2 and l3 will be
replaced by I, II and V2 in (4). The results obtained
from both LC and S12S have been compared with
respect to the accuracy with which they reconstruct S12 system. When (4) is employed for each
patient separately, the transformation coefficients
obtained are called personalized or patient-specific,
while, when applied on a set of patients, they are
called population-based coefficients.
A training set of 5000 samples from the middle of the recording (training set) of each patient
was used in (5) for obtaining the transformation
coefficients and the whole recording was used as
the testing set. It should be noted that for both LC
system formation, transformation coefficient generation and S12 reconstruction only 8 independent
leads of S12 system has been considered, as rest
viz. III, aVR, aVL and aVF, can be readily obtained
from leads I and II and are redundant. The complete work was carried out on MATLAB (Version
7.10.0.499 R2010a).

Comparison of ECG features of

reconstructed and original signal
Computerized ECG acquisition is, generally, followed by automated analysis and interpretation
algorithms viz. feature extraction [33, 34], Selvester scoring [35, 36], fragmentation detection and
identification [37, 38]. These algorithms have been
found to be very sensitive and hence, requires accurate ECG to operate upon for reliable results.
To verify the performance of our methodology in
the context of aforementioned computerized ECG
interpretation algorithms, we have employed our
recently proposed Time Domain Morphology and
Gradient (TDMG) [33] algorithm to extract features
from PQRST complexes of both originally measured
and the reconstructed signal and computed Root
Mean Square Error (RMSE) (defined in next subsecThis article is available at: www.intarchmed.com and www.medbrary.com

International Archives of Medicine

Section: Translational Cardiology

ISSN: 1755-7682

tion IV E) to provide a detailed comparative study

and discuss efficiency of our proposed LC system
in S12 reconstruction.
We have used 49 patients from TWADB for this
module of evaluation. 23 patients were excluded
because of extreme artifacts in few or many of
the leads of these patients. TDMG [33] operates
accurately on a single PQRST complex, however,
it is not practical to manually detect and select a
PQRST complex from all the 8 leads of 49 patients
for both original and reconstructed signals using
both LC and S12S systems, which accounts to a
total of 1568 lead annotations. To automate the
process of annotation and selection of PQRST complex we used the help of two open source MATLAB
files viz. nqrsdetect.m [39] and select_train.m [40].
The former function detect the fiducial points of
QRS complexes and the latter helps in extraction
of PQRST complexes through variations in the input
arguments. To attain this complex task the results
of TDMG when operated upon PQRST complexes
of each patient were visually observed to fine tune
the input arguments so as to obtain exactly one
PQRST complex. The results obtained for the original and reconstructed signal from the TDMG algorithm were then compared using RMSE.

Performance evaluation metrics

R2 statistics, correlation coefficient (rx), regression
coefficient (bx) [10, 21, 29] and RMSE have been
used as performance evaluation metrics. All signals
were mean centered before applying these metrics.
R2 statistics has been used to evaluate the degree of
association between the measured and the reconstructed signal. Perfect retracing of the measured
wave by the reconstructed wave will be indicated
by a value 100%. Correlation coefficient (rx) [29]
is a metric to estimate the similarity between two
signals and regression (bx) [29] fairly estimates the

2015
Vol. 8 No. 62
doi: 10.3823/1661

amplitude differences between the measured and

reconstructed signal. RMSE is a good measure for
accuracy. The definitions of evaluation metrics are
as follows2:

In this paper, we have compared our results with

three previous works of various authors [10, 12, 20]
using one or more of the aforementioned evaluation metrics. As mentioned in section IV C, we have
compared our proposed reduced 3-lead system i.e.
LC system with S12S system comprising of I, II and
V2 as the basis leads. For both LC and S12S systems,
the same reconstruction methodology has been followed.

Results

Table 1 presents the mean R2, rx and bx values for

the reconstruction of S12 system from LC and S12S
systems using both PTBDB and TWADB. The average values have been calculated over 8 independent
leads of S12 system. The R2 values are higher for the
precordial leads compared to bipolar leads I and II.
The transformation of proposed LC system to S12 has

2. D Derived signal; O Originally measured signal; k denotes kth sample; n total number of samples

Under License of Creative Commons Attribution 3.0 License

2015

International Archives of Medicine

Section: Translational Cardiology

ISSN: 1755-7682

Vol. 8 No. 62
doi: 10.3823/1661

Table 1. M
 ean R2, rx and bx values of the transformation of LC to S12 (1) and S12S to S12 (2) systems for
both TWADB (72 patients) and PTBDB (first recording of 290 patients).
TWADB
I
(1)

(2)

II

V1

V2

V3

PTBDB
V4

V5

V6

AVG

II

V1

V2

V3

V4

V5

V6

AVG

R2

93.24 91.72 96.19 97.33 98.77 98.52 98.11 97.07 96.37 93.32 93.25 95.33 96.76 97.46 96.96 97.17 94.77 95.63

rx

0.974 0.956 0.984 0.989 0.994 0.993 0.992 0.987 0.984 0.970 0.968 0.978 0.985 0.989 0.987 0.987 0.975 0.979

bx

0.969 0.951 0.987 0.996 0.997 0.994 0.992 0.994 0.985 0.956 0.969 0.974 0.986 0.993 0.995 0.987 0.964 0.978

R2

100

rx

0.954

0.981 0.942 0.926 0.918 0.965

0.938

0.955 0.902 0.886 0.887 0.946

bx

0.958

0.969 0.928 0.931 0.935 0.965

0.918

0.947 0.890 0.876 0.877 0.939

100 85.25 100 96.17 88.93 82.51 79.71 91.57 100

outperformed S12S system. This confirms that the

information obtained from standard leads are better
represented in LC system than that of a subset.

100 88.04 100 90.34 77.25 73.99 75.46 88.14

Fig. 2 presents the range of R2 values with the

mean denoted by the circle with green dot and
minimum and maximum values denoted by the

Figure 2: T he mean (circle with green inside) and the range (whiskers) of R2 values for both LC to S12
(A-TWADB, B-PTBDB) and S12S to S12 (C-TWADB, D-PTBDB) transformations..

This article is available at: www.intarchmed.com and www.medbrary.com

International Archives of Medicine

Section: Translational Cardiology

ISSN: 1755-7682

whiskers. It should be noted that Fig. 2 shows

minimum R2 values being negative, which denoted
that the reconstructed signal has failed to trace
the original and is completely out-of-phase with
it. From the range of values, it can be seen that
the proposed LC system is considerably better and
reliable for the reconstruction of precordial leads.
Fig. 3 shows the comparison between the originally measured (blue) and reconstructed (red) signal. One subject each from TWADB (Fig. 3 A & B)
PTBDB (Fig. 3 C & D). The two subjects have mean

2015
Vol. 8 No. 62
doi: 10.3823/1661

R2 values of 91.69%(TWADB) and 88.11%(PTBDB),

which is very close to the overall mean values stated
in Table 1 for S12S to S12 transformation and the
corresponding values for LC to S12 transformation
were 96.71% and 92.64%. The R2 values of respective leads have been indicated in the figure. The R2
values in Fig. 3 ranges from 54.61% to 100% and
provide insight into the correspondence between
the quality of reconstruction and R2 values.
Table 2 presents the RMSE values for the features
extracted from 8 independent leads of S12 system

Figure 3: C
 omparison between the two systems (LC and S12S) for the mean case S12S patient one each
from TWADB and PTBDB. Sub-figures A and B shows comparison of 8 independent leads i.e.
I, II, V1 to V6 for the TWADB and PTBDB and sub-figures C and D show a comparison of a
mean case patient from PTBDB. The reconstructed lead (red) has been plotted over original
lead (blue) corresponding R2 values have shown on respective plots.

Under License of Creative Commons Attribution 3.0 License

2015

International Archives of Medicine

Section: Translational Cardiology

ISSN: 1755-7682

Vol. 8 No. 62
doi: 10.3823/1661

Table 2. M
 ean root mean square error (RMSE) values for various features extracted using TDMG for the
reconstruction methodology. Mean was taken over all the 49 patients (please see section IV D).
The leads I, II and V2 formed the basis leads of S12S, so the RMSE values for them were zero
and hence, has not been reported.
LC to S12
I

II

V1

V2

V3

S12S to S12

Sr

Feature (unit)

V4

V5

V6

V1

V3

P duration (ms)

P height (V)

PR interval (ms)

11.39 11.55 15.27 12.65

PR segment (ms)

10.78 6.449 15.14 8.163 5.551 9.388 5.265 5.020 18.33 11.35 16.78

Q peak (V)

QRS length (ms)

10.65 7.184 7.633 3.755

QT interval (ms)

16.08 15.8 26.98 7.551 18.61

R height (V)

47.26 51.03 47.12 99.95 60.18 88.54 48.41 31.72 78.21 182.4 480.4 158.9 90.75

S peak (V)

9.125 10.36 19.27 0.035 0.036 2.825 9.714 10.20 28.90 2.689 4.220 19.85 17.29

10

ST interval (ms)

9.184 8.857 24.25 7.143 6.612 6.122 4.367 7.796 31.59 7.225 15.06 11.76 18.16

11

ST segment (ms)

14.16 11.92 18.61 10.37 7.184 11.27 6.490 10.90 28.86 7.633 17.76 18.08 20.37

12

T duration (ms)

43.32 64.35 107.5 114.5 139.0 105.2 73.27 43.96 161.2

13

T height (V)

9.755 13.18 13.43 10.12 7.306 9.51 9.632 7.755 16.86 12.37

V4
18

V5

V6

18.82 19.63

52.89 63.59 92.53 84.82 50.52 81.96 37.58 37.03 133.3 110.1 177.7 113.4 89.27
6

7.225 9.020 9.102 19.59 11.47 12.98 12.61 17.06

12

16.04

19.58 14.33 47.73 60.85 54.27 52.51 26.20 16.27 77.22 111.6 353.4 62.59 22.27
4

6.816 5.184 6.122 9.306 9.143 11.63 10.61 10.37

10

5.469 8.531 30.61 21.59 20.33 11.10 18.65

199

202.5 135.7 98.76

22.18 32.95 55.05 58.62 71.18 53.84 37.51 22.51 82.53 101.9 103.7 69.50 50.57

between the originally measured and reconstructed

signal. Thirteen different features were extracted
for both the methodologies3. Mean improvement
of over 45% was observed in the proposed LC to
S12 compared to S12S to S12 transformation.
Fig. 4 presents the box plot [41] of overall mean
RMSE values of 13 different features for both the
transformations LC to S12 (A) and S12S to S12 (B)
The edges of the box plot are the 25th and 75th
percentiles, the whiskers extend to +/-2.7 standard
deviation () and rest are plotted individually.
Table 3 compares the reconstruction results presented by previous investigations to the one proposed in this paper. We have used the results obtained by Nelwan et al [12], Dawson et al [10] and
Tsouri et al [20]. Nelwan el al [12] have provided
median and range of correlation coefficients (rx)

for the reconstruction of missing precordial leads

from subsets of S12 system, hence, to obtain the
mean value we have employed the method proposed by Hozo et al [42]. We have used the r x value
of the basis lead set consisting leads I, II and V2
as our proposed system is a 3-lead system. They
have acquired their own private database which is
publicly not available. Dawson et al [10] provided
the R2 values obtained using PhysioNets PTBDB,
same as we have employed in our work. Tsouri et
al [20] have also used PTBDB database and provided rx values in percentages. Upon comparing our
results, we have found that the proposed LC to S12
transformation is superior to all three previously
proposed methodologies.

3. Mean was taken over 8 independent leads.

10

This article is available at: www.intarchmed.com and www.medbrary.com

2015

International Archives of Medicine

Section: Translational Cardiology

ISSN: 1755-7682

Vol. 8 No. 62
doi: 10.3823/1661

Table 3. C
 omparison of the proposed LC to S12 transformation with previously proposed methodologies.
Mean values for 8 independent leads have been presented. Different databases were used and
have been classified as follows: A PTBDB, B TWADB and C self-acquired private database.
Reconstruction Methodology

R2 values

Correlation coefficient (rx)

Proposed Lead Component (LC) system to

Standard 12-Lead system

A 95.63%
B 96.37%

A 0.984 (98.4%)
B 0.979 (97.9%)

Nelwan et al [11] Reconstruction of missing

precordial leads using I, II, V2

C 0.977 (97.7%)

Dawson et al [9] - Affine Transform

A - 81.93%

Tsouri et al [19] ICA based 12-lead ECG

reconstruction

A - I, II, V2 0.9779 (97.79%)

A - Franks XYZ 0.9755 (97.55%)

Figure 4: B
 ox plot of RMSE values. A LC to S12
and B S12S to S12. The labels 1-13
on the horizontal axis corresponds to
the respective features extracted from
TDMG as mentioned in Table 2.

Under License of Creative Commons Attribution 3.0 License

Discussion

From a wide range of R2 values in fig. 3 i.e. 54.61%

to 100%, we have found that R2 value of 80% or
above can be considered to have significant diagnostic value and a value of 90% or above is practically retracing the original waveform. For PTBDB
and TWADB using the proposed methodology, approximately, 96% of patients were found to have
mean R2 value of 80% and above. Similarly, 91%
and 93% of patients were found to have mean R2
value of 90% and above for PTBDB and TWADB
respectively. For S12S to S12 transformation, the
fraction of patients above with mean R2 values
80% were 86% (PTBDB) and 92% (TWADB). The
fraction of patients with mean R2 values 90%
were 76% (PTBDB) and 83% (TWADB). In Fig. 4, it
can be seen that RMSE has higher values for peaks
and nadirs e.g P height, Q peak etc as compared
to the features related to time axis e.g. P duration,
QRS length etc. as compared to the features related to the time axis e.g. P duration, QRS length
etc. From Tables 1 and 2 and Fig. 2-4 it can be
seen that the proposed reconstruction methodology has considerably outperformed the transformation of S12S to S12, especially, for leads V4 to
V6 in terms of accuracy and robustness in S12 lead
reconstruction which is of utmost importance for
diagnosis particularly for remote healthcare scenario. The proposed methodology has yielded a
mean improvement of 7.49% (PTBDB) and 4.8%

11

International Archives of Medicine

Section: Translational Cardiology

ISSN: 1755-7682

(TWADB) over our previously proposed personalized methodologies. This improvement is significant and shows importance of selecting appropriate lead system.
In Gregg et al. [13], the authors have discussed
the limitations on the re-usability of transformations coefficients. They showed that the reconstruction results obtained from the same transformation coefficients calculated from a particular
measurement deteriorates when electrodes are removed and replaced for another acquisition. They
attributed this deterioration mainly to the human
error involved in replacement of electrodes at the
same positions. Using our proposed methodology
the coefficients can be recalculated every time a
measurement is taken with the help of computerized ECG acquisition systems. In cases where
sudden event occurs while monitoring viz. arrhythmia, then the coefficients will have to be calculated
again. computerized ECG acquisition systems can
be readily used for such events. In the proposed
methodology, S12 leads needs to be acquired whenever signals are needed to be transmitted and
this paper mainly addresses the limitations encountered in RHM scenario. Storage of ECG data of
patients in a telemonitoring system using our proposed methodology provides the following:

Three leads require less space compared to
eight leads of S12 system and hence, can be
used to reduce the memory requirements.
Proposed methodology ensures accuracy and
robustness in the S12 lead reconstruction.
With the decrease in number of leads the compression ratio increases [6].
The proposed methodology does not require
any change in the hardware of the, today, widely used computerized ECG acquisition system.
The proposed system with the application of
PCA reduces data redundancy.
Leads enriched with condensed and most important information is retained which have high
Signal-to-Noise (SNR) ratios as compared to S12

12

2015
Vol. 8 No. 62
doi: 10.3823/1661

leads which makes them suitable for wireless or

any other mode of transmission.
The proposed system ensures the availability of
S12 system using personalized lead reconstruction methodology for diagnosis.
PCA has found many applications in signal compression algorithm [6]. However, in this paper, our
objective is to assimilate most of the information
present in S12 leads on three leads so as to reduce
the number of channels and facilitate its reconstruction. Signal compression will be the next step to be
applied before transmission. ICA has been used by
Tsouri and Ostertag [20], however, our PCA based
methodology is comparatively superior. The major
limitation of ICA is the unpredictable ordering in
which ICs are obtained. In [20], ICA was applied
on basis leads I, II and V2 and Franks XYZ leads
which already have limited information, hence, ICA
cannot not obtain any extra information. However,
we have applied PCA on all the 8 independent
leads, thereby, LC system has more information on
the S12 leads. Franks system is supposed to be an
orthogonal system with independent leads as they
are on different planes and ICA produces another
set of orthogonal independent leads, which can
be assumed to be different representation of the
same information, thereby, not gaining any extra
insight into the nature of S12 leads. Several other
algorithms exists e.g. LDA, non-linear PCA, which
can be evaluated in the context of lead reconstruction, however, this we leave upto the readers
to evaluate.

Conclusion
In this paper, we proposed a personalized and robust reduced 3-lead system viz. Lead Component
(LC) system, formation methodology targeting the
emerging field of remote health monitoring to improve diagnosis and prognosis by overcoming the
constraints from technical and medical science side.
It has been shown that the proposed LC system outThis article is available at: www.intarchmed.com and www.medbrary.com

International Archives of Medicine

Section: Translational Cardiology

ISSN: 1755-7682

performs the existing methodologies [10, 12, 15, 20,

21] by various investigators for the reconstruction of
standard 12-leads and has been verified on two widely accepted databases viz. PTBDB and TWADB of
PhysioNet. The importance of an appropriate personalized reduced lead system has been showcased.
We introduced PCA in this context to obtain and
orthogonal 3-lead personalized system which reduces dimension of the data from 8 leads to 3 leads
resulting in an increase in the SNR and Compression
ratio, which favors wireless transmission of the data
either to the medical practitioner's mobile or to the
personalized health data repository in the cloud. As
mentioned in the previous section, the personalized
coefficients for reconstruction are always recalculated and retransmitted each time the contact electrodes are positioned on the patient, thus our proposed
system would greatly reduce the attrition dequalification rate discussed by Gregg et al. [13].

Funding
This work was partly supported by the Department
of Electronics and Information Technology (DEITY),
India, under the Cyber Physical Systems Innovation
Hub [grant number 13(6)/2010-CC&BT dated 1
March 2011].

Disclosure
the acceptance of this article has been taken by the
editor-in-chief of International Archives of Medicine.

Under License of Creative Commons Attribution 3.0 License

2015
Vol. 8 No. 62
doi: 10.3823/1661

References
1. Menotti A, Lanti M, Mancini A, Zanchetti A, Laurenzi M,
Terradura Vagnarelli O, Puddu PE. Moderate leisure time physical
activity and occurrence of cardiovascular events: a 10-year
follow-up in the Gubbio Population Study. JCvD 2014;2(5):218224.
2. Bosznai I, Ender F, Santha H. Web server based remote health
monitoring system. In: 32nd International Spring Seminar on
Electronics Technology, 2009 (ISSE 2009), 13-17 May 2009. p.
1-6. (DOI: 10.1109/ISSE.2009.5206985)
3. Puddu PE, DAmbrosi A, Scarparo P, Centaro E, Torromeo C,
Schiariti M, et al. A Clinicians View of Next-Generation Remote
Healthcare System. In: Maharatna K, Bonfiglio S, editors.
Systems Design for Remote Healthcare. Springer New York,
2014. p. 1-30. (DOI: 10.1007/978-1-4614-8842-2_1)
4. Puddu PE, Morgan JM, Torromeo C, Curzen N, Schiariti M,
Bonfiglio S. A Clinical Observational Study in the CHIRON Project:
Rationale and Expected Results. In: Donnelly M, Paggetti C,
Nugent C, Mokhtari M, editors. Impact Analysis of Solutions for
Chronic Disease Prevention and Management. Springer Berlin
Heidelberg, 2012, p. 74-82. (DOI: 10.1007/978-3-642-307799_10)
5. Liszka KJ, Mackin MA, Lichter MJ, York DW, Dilip P, Rosenbaum
DS. Keeping a beat on the heart. Pervasive Comput. 2004 Oct;
3(4):42-9. (DOI: 10.1109/MPRV.2004.10)
6. Sharma LN, Dandapat S, Mahanta A. Multichannel ECG
Data Compression Based on Multiscale Principal Component
Analysis. IEEE Trans Inf Technol Biomed. 2012 July; 16(4):730-6.
(DOI: 10.1109/TITB.2012.2195322)
7. Frank E. An accurate, clinically practical system for spatial
vectorcardiography. Circulation. 1956; 13:737-49.
8. Hoekema R, Uijen GJH, van Oosterum A. On selecting a body
surface mapping procedure. J Electrocardiol. 1999; 32(2):93101. (DOI: 10.1016/S0022-0736(99)90088-2)
9. Dower GE. A lead synthesizer for the frank system to simulate
the standard 12-lead electrocardiogram. J Electrocardiol 1968;
1(1):101-16. (DOI: 10.1016/0022-0736(88)90090-8)
10. Dawson D, Yang H, Malshe M, Bukkapatnam STS, Benjamin
B, Komanduri R. Linear affine transformations between
3-lead (Frank XYZ leads) vectorcardiogram and 12-lead
electrocardiogram signals. J Electrocardiol 2009; 42(6):622-30.
(DOI: 10.1016/j.jelectrocard.2009.05.007)
11. Maheshwari S, Acharyya A, Puddu PE, Mazomenos EB, Schiariti M,
Maharatna K. Robust and accurate personalised reconstruction
of standard 12-lead system from Frank vectordcardiographic
system. Comput Methods Biomech Biomed Eng Imaging Vis.
2014. (DOI: 10.1080/21681163.2014.931029)
12. Nelwan SP, Kors JA, Meij SH, van Bemmel JH, Simoons ML.
Reconstruction of the 12-lead electrocardiogram from reduced
lead sets. J Electrocardiol 2004; 37(1):11-18. (DOI: 10.1016/j.
jelectrocard.2003.10.004)

13

International Archives of Medicine

Section: Translational Cardiology

ISSN: 1755-7682

13. Gregg RE, Zhou SH, Lindauer JM, Helfenbein ED, Field DQ.
Limitations on the Re-use of patient specific coefficients for
12-lead ECG reconstruction. In: Computers in Cardiology 2008
Sept 14-17. p. 209-212. (DOI: 10.1109/CIC.2008.4749014)
14. Field DQ, Zhou SH, Helfenbein ED, Gregg RE, Lindauer JM.
Technical challenges and future directions in lead reconstructions
for reduced-lead systems. J Electrocardiol 2008; 41(6):466-73.
(DOI: 10.1016/j.jelectrocard.2008.07.019)
15. Maheshwari S, Acharyya A, Puddu PE, Schiariti M. Reduced
lead system selection methodology for reliable standard 12-lead
reconstruction targeting personalised remote health monitoring
applications. Comput Methods Biomech Biomed Eng Imaging
Vis. 2014; 2(2):107-20. (DOI: 10.1080/21681163.2013.859097)
16. Finlay DD, Nugent CD, Kellet JG, Donnelly MP, McCullagh,
Black ND. Synthesising the 12-lead electrocardiogram: Trends
and challenges. Eur J Intern Med 2007; 18(8):566-70. (DOI:
10.1016/j.ejim.2007.04.011)
17. Trobec R, Tomasic I. Synthesis of the 12-Lead Electrocardiogram
From Differential Leads. IEEE Trans Inf Technol Biomed. 2011
July; 15(4): 615-21. (DOI: 10.1109/TITB.2011.2159236)
18. Finlay DD, Nugent CD, Donnelly MP, Lux RL. Eigenleads: ECG
Leads for Maximizing Information Capture and Improving SNR.
IEEE Trans Inf Technol Biomed. Jan 2010; 14(1):69-78. (DOI:
10.1109/TITB.2009.2022933)
19. Axisa F, Schmitt P, Gehin C, Delhomme G, McAdams E,
Dittmar A. Flexible technologies and smart clothing for citizen
medicine, home healthcare, and disease prevention. IEEE Trans
Inf Technol Biomed. Sept 2005; 9(3):325-336. (DOI: 10.1109/
TITB.2005.854505)
20.
Tsouri GR, Ostertag MH. Patient-Specific 12-Lead ECG
Reconstruction From Sparse Electrodes Using Independent
Component Analysis. IEEE J Biomed Health Inform. Mar 2014;
18(2):476-82. (DOI: 10.1109/JBHI.2013.2294561)
21. Maheshwari S, Acharyya A, Rajalakshmi P, Puddu PE, Schiariti
M. Accurate and reliable 3-lead to 12-lead ECG reconstruction
methodology for remote health monitoring applications. In:
IEEE 15th International Conference on e-Health Networking,
Applications Services (Healthcom), Oct 2013. P. 233-37. (DOI:
10.1109/HealthCom.2013.6720673)
22. Maheshwari S, Acharyya A, Rajalakshmi P, Puddu PE, Schiariti
M. Accurate and reliable 3-lead to 12-lead ECG reconstruction
methodology for remote health monitoring applications. IRBM.
Dec 2014; 35(6):341-350. (DOI: 10.1016/j.irbm.2014.07.004)
23. Goldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov
PCh, Mark RG, et al. PhysioBank, PhysioToolkit, and PhysioNet:
Components of a New Research Resource for Complex
Physiologic Signals. Circulation. 2000; 101(23):e215-e220.
24. Bousseljot R, Kreiseler D, Schnabel A. Nutzung der EKG
Signaldatenbank CARDIODAT der PTB ber das Internet.
Biomedizinische Technik. 1995; 40(1 supp):317.

14

2015
Vol. 8 No. 62
doi: 10.3823/1661

25. Moody GB. The PhysioNet / Computers in Cardiology Challenge

2008: T-Wave Alternans. In: Computers in Cardiology 2008. p.
505-8.
26. Sharma LN, Dandapat S, Mahanta A. Multiscale principal
component analysis to denoise multichannel ECG signals. CIBEC
2010: 5th Cairo International Biomedical Engineering Conference;
Dec 2010. p. 17-20. (DOI: 10.1109/CIBEC.2010.5716093)
27. Martis RJ, Chakraborty C, Ray AK. An Integrated ECG Feature
Extraction Scheme Using PCA and Wavelet Transform. INDICON
2009: Annual IEEE India Conference, Dec 2009. p. 1-4. (DOI:
10.1109/INDCON.2009.5409439)
28. Frank E. General Theory of Heart-Vector Projection. Circulation
Research. 1954; 2(3):258-70. (DOI: 10.1161/01.RES.2.3.258)
29. Levkov CL. Orthogonal electrocardiogram derived from the
limb and chest electrodes of the conventional 12-lead system.
Med Biol Eng Comput. 1987; 25(2):155-64. (DOI: 10.1007/
BF02442844)
30. Jolliffe IT. Principal Component Analysis. Springer Series in
Statistics, Springer New York 2002.
31. Zhang D. Wavelet Approach for ECG Baseline Wander Correction
and Noise Reduction. IEEE EMBS 2005: 27th Annual International
Conference of the Engineering in Medicine and Biology Society,
2005. p. 1212-5. (DOI: 10.1109/IEMBS.2005.1616642)
32. Antoniadis A, Bigot J, Sapatinas T. Wavelet Estimators in
Nonparametric Regression: A Comparative Simulation Study. J
Stat Softw. Jun 2001; 6(6):1-83.
33.
Mazomenos EB, Chen T, Acharyya A, Bhattacharya A,
Rosengarten J, Maharatna K. A Time-Domain Morphology and
Gradient based algorithm for ECG feature extraction. ICIT 2012:
IEEE International Conference on Industrial Technology, Mar
2012. p. 117-122. (DOI: 10.1109/ICIT.2012.6209924)
34. Mazomenos EB, Biswas D, Acharyya A, Chen T, Maharatna
K , Rosengarten J, et al. A Low-Complexity ECG Feature
Extraction Algorithm for Mobile Healthcare Applications. IEEE J
Biomed Health Inform. Mar 2013; 17(2):459-69. (DOI: 10.1109/
TITB.2012.2231312)
35. Horacek BM, Warren JW, Albano A, Palmeri MA, Rembert JC,
Greenfield Jr. JC, Wagner GS. Development of an automated
Selvester Scoring System for estimating the size of myocardial
infarction from the electrocardiogram. J Electrocardiol. 2006;
39(2):162-8. (DOI: 10.1016/j.jelectrocard.2005.08.013)
36. Bono V, Mazomenos EB, Chen T, Rosengarten J, Acharyya
A, Maharatna K, et al. Development of an Automated
Updated Selvester QRS Scoring System Using SWT-Based
QRS Fractionation Detection and Classification. IEEE J Biomed
Health Inform. Jan 2014; 18(1):193-204. (DOI: 10.1109/
JBHI.2013.2263311)

This article is available at: www.intarchmed.com and www.medbrary.com

International Archives of Medicine

Section: Translational Cardiology

ISSN: 1755-7682

2015
Vol. 8 No. 62
doi: 10.3823/1661

37. Maheshwari S, Acharyya A, Puddu PE, Schiariti M. Methodology

for automated detection of fragmentation in QRS complex
of Standard 12-lead ECG. IEEE EMBS 2013: 35th Annual
International Conference of the IEEE Engineering in Medicine
and Biology Society, 2013. p. 3789-3792. (DOI: 10.1109/
EMBC.2013.6610369)
38. Maheshwari S, Acharyya A, Puddu PE, Mazomenos EB, Leekha
G, Maharatna K, Schiariti M. An automated algorithm for online
detection of fragmented QRS and identification of its various
morphologies. J R Soc Interface. 2013; 10(89). (DOI: 10.1098/
rsif.2013.0761)
39. Afonso V, Tompkins WJ, Nguyen T, Luo S. ECG beat detection
using filter banks. IEEE Trans Biomed Eng. 1999; 46(2):192-202.
(DOI: 10.1109/10.740882)
4 0. Clifford GD. ECG tools. Available from: http://www.robots.
ox.ac.uk/~gari/CODE/ECGtools
41. Mathworks. Boxplot. Available from: http://www.mathworks.
in/help/stats/boxplot.html
42. Hozo SP, Djulbegovic B, Hozo I. Estimating the mean and
variance from the median, range, and the size of a sample. BMC
Med Res Methodol. Apr 2005; 5(13). (DOI: 10.1186/1471-22885-13)

Comment on this article:

http://medicalia.org/
Where Doctors exchange clinical experiences,
review their cases and share clinical knowledge.
You can also access lots of medical publications for
free. Join Now!

Publish with iMedPub

http://www.imed.pub
International Archives of Medicine is an open access journal
publishing articles encompassing all aspects of medical science and clinical practice. IAM is considered a megajournal with
independent sections on all areas of medicine. IAM is a really
international journal with authors and board members from all
around the world. The journal is widely indexed and classified
Q1 in category Medicine.

Under License of Creative Commons Attribution 3.0 License

15