Content uploaded by Fahim Sufi
Author content
All content in this area was uploaded by Fahim Sufi on Jul 26, 2022
Content may be subject to copyright.
SECURITY AND COMMUNICATION NETWORKS
Security Comm. Networks 2011; 4:1351–1368
Published online 17 March 2011 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/sec.262
RESEARCH ARTICLE
Cardioids-based faster authentication and diagnosis
of remote cardiovascular patients
Fahim Sufi1∗, Ibrahim Khalil1and Ibrahim Habib2
1School of Computer Science and IT, RMIT University, Melbourne, VIC, 3000, Australia
2Department of Electrical Engineering, City University of New York, Convent Avenue at 140th Street New York, NY 10031, U.S.A.
ABSTRACT
In recent times, dealing with deaths associated with cardiovascular diseases (CVD) has been one of the most challenging
issues. The usage of mobile phones and portable Electrocardiogram (ECG) acquisition devices can mitigate the risks
associated with CVD by providing faster patient diagnosis and patient care. The existing technologies entail delay in patient
authentication and diagnosis. However, for the cardiologists minimizing the delay between a possible CVD symptom and
patient care is crucial, as this has a proven impact in the longevity of the patient. Therefore, every seconds counts in terms
of patient authentication and diagnosis. In this paper, we introduce the concept of Cardioid based patient authentication and
diagnosis. According to our experimentations, the authentication time can be reduced from 30.64 s (manual authentication
in novice mobile user) to 0.4398 s (automated authentication). Our ECG based patient authentication mechanism is up to
4878 times faster than conventional biometrics like, face recognition. The diagnosis time could be improved from several
minutes to less than 0.5 s (cardioid display on a single screen). Therefore, with our presented mission critical alerting
mechanism on wireless devices, minute’s worth of tasks can be reduced to second’s, without compromising the accuracy
of authentication and quality of diagnosis. Copyright © 2011 John Wiley & Sons, Ltd.
KEYWORDS
mission critical alerting; cardiovascular disease detection; remote monitoring; wireless monitoring; patient authentication; cardioid
*Correspondence
Fahim Sufi, School of Computer Science and IT, RMIT University, Melbourne, VIC, 3000, Australia.
E-mail: research@fahimsufi.com
1. INTRODUCTION
With the increasing rate of cardiovascular related deaths,
more and more people are inclined to join cardiovascu-
lar monitoring facilities for faster diagnosis and treatment.
Faster diagnosis for cardiovascular patient is mandatory
because of the well known proverb in cardiology: ‘Time
is Muscle’. The damage initiated by many cardiovascu-
lar diseases is an irrecoverable process. Therefore, once
the damage to cardiovascular muscle starts, the damaged
cells will remain damaged for ever. For this reason, faster
cardiovascular anomaly diagnosis and treatment resist fur-
ther complications as heinous as mortality [1--4]. Hence,
there are many researches performed on minimizing the
cardiovascular disease diagnosis and treatment time [1--8].
Mobile phone based alert mechanism for undergoing car-
diovascular rescue mechanism is a recent active research
area [5--12]. Technological breakthroughs in biosensors,
embedded computing and mobile phones are promising new
arena in mobile phone based real time patient monitoring
[5--7,13--16]. As the outcome of these rigorous researches
in mobile telecardiology, innovative business ventures have
been established [17]. In this business model, cardiovascu-
lar patients are subscribed to a remote monitoring facility
[17]. These methods provide faster detection of disease, and
faster patient care [17].
In a typical scenario as shown in Figure 1, the patient
is connected with small biosensors/medical sensors, which
collect various physiological signals like Electrocardio-
gram, Polyplethogram, Blood Oxygen Saturation level
(SpO2), motion activities (Accelerometer). These small
biosensors, then transmits the physiological signals to the
mobile phone of the patient through Bluetooth, Near Field
Communication or WiFi link. After receiving the phys-
iological signals from the biosensors, the mobile phone
compresses (for faster transmission), encrypts (for main-
taining patient privacy) and transmits to the monitoring
station via Short Messaging Service (SMS), Multimedia
Messaging Service (MMS), Hyper Text Transfer Proto-
col (HTTP) or Wireless Local Area Network (WLAN)
[5--7,13,15]. Basically, the patients are subscribed to the
cardiovascular monitoring facility. Therefore, the patients
Copyright © 2011 John Wiley & Sons, Ltd. 1351
Diagnosis of remote cardiovascular patients F. Sufi, I. Khalil and I. Habib
Figure 1. Mobile phone based cardiac patient monitoring (general scenario). Patient’s ECG acquisition device to Patient’s mobile phone
is communicated via Bluetooth Protocol. The communication among patient, doctor, hospital, and ambulance is performed over HTTP,
MMS, or SMS protocol.
need to be authenticated first before granting them access
to the monitoring facility. After the authentication mech-
anism is established cardiovascular disease recognition is
performed. Even though, manual inspection of the ECG
trace by expert cardiologist is the established methodology
for cardiovascular diagnosis, automated algorithms read-
ing ECG signal for performing abnormality detection is an
active research area for as long as past two decades [18-
-21]. While most of these algorithms being designed and
tested primarily for PC based environment, mobile phone
based cardiovascular abnormality detection algorithm is a
very recent area of research [5--9,11].
Faster patient care can be envisaged by minimizing delays
in the following component of mobile phone based remote
cardio-care:
rPhysiological data transmission (as depicted in our
earlier research [5,6,13])
rPatient authentication (as described in our earlier
research [15,22])
rCardiovascular diagnosis (as demonstrated in earlier
research [5])
According to the recent literature, the delay associated
with ECG data transmission has been reduced drasti-
cally with the usage of specialized lossless compression
techniques (designed for ECG) [5,6,13]. Therefore, our
recent research activities [5,13,6] successfully minimized
the delay associated with ECG data transmission. However,
existing mobile phone based remote monitoring systems
[23] with username and password based authentication
mechanism takes 12–35 s of time, according to our exper-
imentation. As the patient requires manual input of the
username and password, these types of solutions (in Ref-
erence [23]) are not fruitful for patients having cardiac
attacks which subsequently triggers anomaly in regular
finger movement. As a result, we came up with an auto-
mated solution with ECG based biometric with polynomial
distance measurements [15,22]. However, generation of
polynomial coefficients in mobile devices takes significant
amount of time (around 12 s).
On the other hand, reducing the delay associated with
ECG based authentication is a challenging issue for faster
patient care. Existing ECG diagnosis algorithms are mainly
based on PC and Servers [18--21] and can be classified in
three broader categories in terms of complexities.
rFiducial Technique (e.g., detection of wave onset, off-
set, amplitude, duration, slope, etc.) [24,25]
rTransformational Techniques (e.g., Wavelet transform,
Fourier Transform, Cosine Transform, etc.) [26,27]
rIntelligent Techniques (e.g., Support Vector Machines,
Fuzzy logic, Neural Network, and Other classifiers)
[21]
Before understanding these techniques, a background of
ECG signal is required. A typical ECG trace comprises
of P waves, QRS complexes, and T waves (as seen from
Figure 2). P wave reflects atrial depolarization (or mechan-
ical contraction). After P wave, QRS complex represents
ventricular depolarization (or mechanical contraction).
Lastly, T wave signifies ventricular re polarization, when
the contracted ventricles relax (or diastole). The mechani-
cal contraction and relaxation (or electrical depolarization
and re polarization) keep the oxygenated flowing around the
living body. Any abnormality in this regular process can be
picked up from the irregular ECG trace (electrical activity
of the heart). ECG is a time series data, where Y-axis shows
1352 Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd.
DOI: 10.10 02/sec
F. Sufi, I. Khalil and I. Habib Diagnosis of remote cardiovascular patients
the electrical activity in volt or millivolt range. On the other
hand, X-axis can either show time information or number
of samples.
Fiducial point based technique generally being the fastest
method of ECG diagnosis, includes multiple steps. These
steps start with the detection QRS Complex. Then, onset
and offset of QRS is detected. Next amplitude and duration
of QRS is detected. Then, other parameters such as slope
of QRS are detected. Similarly, onsets, offsets, amplitudes,
durations are calculated of P wave and T waves. Detection
of all these parameters for all the feature waves within the
ECG trace entails significant delay in mobile environment.
Transformational and Intelligent techniques are generally
more complex than fiducial point based techniques, making
them unsuitable for resource constraint mobile devices [5--
7,28].
Within this paper, we present the novel idea of cardioid
based ECG abnormality detection and patient authentica-
tion mechanism. According to the literature, this simple
yet effective method has never been reported. We have
implemented the cardioid based real-time cardiovascular
diagnosis and patient authentication mechanism on server,
hand held and mobile platform. Our experimentation results
suggest that Cardioid based abnormality detection and
patient authentication substantially minimizes the delay
associated with the treatment of cardiovascular patient.
We first present the concept mechanisms of center of
cardioid followed by its application in person identifica-
tion/authentication. Then, we demonstrated its applicability
in faster diagnosis. Both authentication and diagnosis
components are the integral part of a Mission Critical Car-
diovascular Abnormality Alerting (MCCAA) mechanism
designed to provide faster patient care and save precious
life. Our Caridioid based authentication ensures security in
MCCAA. Reliability of the MCCAA is ensured by the sup-
port of multiple communication protocols as discussed in
Section 6.
2. MOTIVATION
This section provides a general understanding of the
medical significance for faster cardiovascular care, which
is driving our research in mobile phone based alert
mechanism. To delve into the understanding of faster car-
diovascular patient care, the concept of ‘Door to Ballooning
Time (D2B)’ and ‘Symptom-Onset-to-Balloon Time’ can
be reviewed.
Door to ballooning refers to the measurement of time for
the treatment of ST segment elevation myocardial infrac-
tion (STEMI) or acute Myocardial Infraction (MI). This
is basically the time between a patient’s arrival in the
Emergency Department (ED) and balloon angioplasty (or
Balloon inflation). Delay in balloon inflation and subse-
quent insertion of mesh wire to enable free blood flow
with the heart, effectively creates the an environment where
blood gets coagulated and forms blood clot. Formation
of blood clot lead to irreversible cardiac cell damages.
ACC/AHA guidelines recommend the D2B time less than
90 min. Symptom-Onset-to-Balloon time refers to the time
interval between the patient feeling discomfort (cardiac
symptom) and catheter guide wire crosses the culprit lesion
in the cardiac cathlab. When the patient first feels a car-
diac discomfort identifying possible incident of MI, he/she
calls the ambulance. The ambulance then brings the sus-
pected MI affected patient in the ED. The ED personnel
then undergo a detailed ECG acquisition of the patient
and decide of the procedure to be taken. Based on their
decision with cardiovascular experts, the patient may be
taken to the cathlab. This lengthy process is susceptible
to deteriorate patient’s cardiovascular health. To minimize
the detrimental effect of this delay, many ambulances are
equipped with ECG equipments. Therefore, ECG can be
obtained from the patient, while the patient is being trans-
ported to the ED. The acquired ECG can then be sent to
the ED/Cardiovascular experts within the hospital, prior to
patients encounter to the hospital. Therefore, the hospital
can take early diagnosis and treatment decision. The hos-
pital can instruct the ambulance personnel, where to take
the patient. This process can bypass the patient’s admit-
tance in ED and enable urgent patients be directly undergo
surgery in cathlab. This ambulance to hospital communi-
cation can be performed by Fax, Email, MMS, HTTP, etc.
The usefulness of this type process is very recently drawn
in Reference [1]. The researchers in Reference [1] trans-
mitted real-time ECG, vital signs (blood pressure, heart
rate, and oxygen saturation) and live video directly from
the ambulance to the on duty cardiologist in the hospital.
After viewing the ECG trace and diagnosing the possi-
ble STEMI, the cardiologist in Reference [1] can activate
the catheterization laboratory. The ambulance to hospital
transmission delay was within 10 s. The crucial importance
of pre-hospital diagnosis is also reported in recent liter-
ature [4]. Researchers in Reference [2] have plotted the
relationship between ischemic time and 1 year mortality.
They have showed that each 30min of delay was associated
with a relative risk for 1-year mortality of 1.075 (95%CI
1.008–1.15; P=0.041). The conclusion drawn by them
[2] was simple; all efforts should be made to shorten the
total ischemic time. For the research presented in Refer-
ence [1], a mobile phone connected with an in-ambulance
server was deployed for enabling data transmission between
the ambulance and the server. In Reference [3] usage of
mobile phone in diagnosis of heart attack minimized delay
in treatment from 94 to 22 min. The study in Reference
[3], transmitted 12 lead ECG data to the attending cardi-
ologist’s mobile phone. This study [3], not only make it
evident that usage of mobile phone in cardiovascular mon-
itoring provides faster patient care but also it shows that
important diagnosis decisions can be made based on the
ECG plot drawn on mobile phone’s screen. In fact, taking
ECG before the patient’s admittance to the hospital signif-
icantly reduces the Symptom-Onset-to-Balloon time. Any
efforts in minimizing delays associated with patient care
impacts in saving patients life [1--5,10,12,22,29]. However,
this process demands significant improvement in terms of
minimizing end-to-end delay.
Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd. 1353
DOI: 10.10 02/sec
Diagnosis of remote cardiovascular patients F. Sufi, I. Khalil and I. Habib
Figure 2. An ECG Wave.
3. ARCHITECTURE AND SYSTEM
DESIGN
The proposed architecture of our mission critical cardio-
vascular abnormality alert system monitors, the patients
ECG in real-time. In the event of cardiovascular abnormal-
ity, the proposed alerting mechanism notifies the hospital
personnel. The hospital can come to a quick decision by
undergoing rapid diagnosis. The hospital personnel then
that initiate life saving protocols by sending emergency
rescue team, ambulance, etc. With the proposed frame-
work, faster life saving effort is established with both faster
authentication and faster diagnosis. The underlying tech-
nology of both authentication and diagnosis is focused on
a computationally inexpensive, yet effective method called
center of cardioid.
In the MCCAA architecture presented in Figure 3, when
the ECG trace of a subscribed patient is normal, the major
task established by the system is to obtain back-up of the
current data (referred as previous data). The saved data is
used to create the biometric template for authentication pur-
poses in the case of heart abnormality. Therefore, the routine
task for the scenario when the patients ECG signal remains
normal is to acquire ECG data from the acquisition device,
calculate the center of the cardioid, save the current ECG
data (provided that ECG data is normal as center of cen-
troid lies within a preset range). This range is dependent on
the monitored person, the monitoring device and the sensor
placement configuration (Lead I, II, III, V1–V6, etc.). On an
event of cardiovascular abnormality, the centroid location
will be shifted from the normal range. Since authentica-
tion must be carried out with normal ECG data containing
the patient’s biometric trait, previously saved data (with-
out abnormality) is engaged in authentication procedure.
As the abnormal data (current ECG data) hardly inherits
any trait for the patient, it cannot be used for biometric
feature extraction. Therefore, the previously saved normal
ECG trace (without the occurrence of the abnormal trait)
fetched and segregated to extract P wave, QRS complex,
and T wave. From the individual feature waves biometric
recognition data is prepared.
In the next phase, listener (of the patient’s mobile) is
turned on and a message containing the biometric features
(for patient authentication) is created and sent to the hos-
pital. The listener handles communication protocol with
the hospital. On arrival of the biometric template mes-
sage, the hospital’s biometric server performs one to many
matches against its entire subscribers list. If the hospital’s
authentication mechanism recognizes the biometric tem-
plate as a valid subscriber, then in the next step it obtains
the abnormal ECG data (current data containing the ECG
abnormalities) through the listener (already turned on) of the
patient’s mobile phone. The hospital then runs their algo-
rithms in ascertaining the seriousness of the abnormalities.
In case of urgency, patients location information (GPS coor-
dinates) is pulled through the listener and emergency team
is informed. Even if the patient is unconscious, the emer-
gency team will be able to locate the patient to undergo
life saving procedures. On the other hand, if the patient is
not a member within the hospitals enlisted patients then the
listener is turned off. In case of less serious event (deter-
mined by hospitals diagnosis algorithms), the listener is
also communicated and instructed to undergo shut down
procedures.
4. CARDIOID BASED
AUTHENTICATION MECHANISM
Template matching is the core process for any biometric
authentication. Identification template or verification tem-
plate is matched against the enrolment temple. For our
cardioid-based biometric, both identification and verifica-
tion ECG templates are commonly termed as recognition
ECG template. These recognition ECG and enrolment ECG
are matched against each other to determine the identity
of a person. These template matching processes can either
run in the patient’s mobile (for patient side biometric) or
in the hospital server (hospital side biometric) as shown in
Figure 3. Within the context of this paper, we have designed,
developed and investigated two cardioid based authentica-
tion mechanisms referred as Methods 1 and 2.
A cardioid (as shown in Figure 4) drawn from ECG sam-
ple has distinguishing features such as its area, perimeter
and center coordinates. This is obvious in Table I where
cardioids of different segments of ECGs from two differ-
ent persons are plotted. This is the basis and motivation of
our work for this novel authentication mechanism. Most of
the ECG biometric features are kept as points in Cartesian
co-ordinate system in both the methods. However, for our
Method 1 there are two parameters (Area and Circumfer-
ence), which are maintained as decimal values. Therefore,
within the context of this paper matching essentially means
obtaining two types of distances: straight line distance and
percentage distance. Therefore, enrolment ECG can be rep-
resented as follows:
E1=(ce
x,c
e
y),A
e,U
e(1)
1354 Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd.
DOI: 10.10 02/sec
F. Sufi, I. Khalil and I. Habib Diagnosis of remote cardiovascular patients
Figure 3. Architecture of Mission Critical Cardiovascular Abnormality Alerting System.
Similarly, recognition ECG can be expressed as
R1=(cr
x,c
r
y),A
r,U
r(2)
Here, (cx,cy) is the Cartesian coordinate points (super
scripts rand edenotes recognition and enrolment respec-
tively), Ais the area and Uis the circumference or perimeter.
The matching function ((E1,R
1)) for Method 1 pro-
duces a set of thresholds, ={1,
2,
3,...,
n}. In fact,
(E1,R
1)=(c2
x−cr
x)2+(c2
y−cr
y)2,
Figure 4. A typical cardioid.
Ae−Ar
Ae×100,Ue−Ur
Ue×100
=(3)
Whenever, during a matching process (as shown in
Figure 5) the threshold is less than a value, successful recog-
nition is thought to be made.
Template creations for both methods are preceded by
acquisition, loop generation and loop segregation process in
the cardioid-based system as shown in Figure 7. However,
the actual template creation for Methods 1 and 2 are quite
different. Owing to this difference in template creation for
both the methods, Method 1 provides computational inten-
sive and highly accurate ECG biometric (since this method
has more feature templates like Aand U), and Method 2
provides lower complexity ECG biometric.
rAcquisition: During the acquisition of ECG, as a
biometric entity, acquisition devices like GE ST 5500,
Alive Heart Monitor, Phillips Page Writer, etc. can be
used. After the data acquisition, data are converted to
millivolt (mV) ranges from their proprietary format.
ECG data can be de-noised using with the help of
embedded feature of the ECG acquisition device. How-
ever, for the research presented in this paper, we have
only used publicly available ECG data from MIT BIH
(http://www.physionet.org/physiobank/database/mitdb/).
MIT BIH Database contains ECG data that were
collected from numerous patients with cardiac
abnormality using ECG acquisition devices following
digitization of the data. ECG entries from this database
have been extensively used for benchmarking algo-
rithms pertaining to ECG diagnosis, compression and
other researches [5]. Mathematically, ECG can be
represented by x(n) as in Equation (1).
x(n)={x(1),x(2),x(3),...,x(N)}(4)
Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd. 1355
DOI: 10.10 02/sec
Diagnosis of remote cardiovascular patients F. Sufi, I. Khalil and I. Habib
Table I. The difference in ECG feature waves between two different individuals.
where x(1),x(2) ... are ECG samples and Nis the
length of ECG signal.
rLoop Generation: Apart from ECG, none of the bio-
metric modalities are time series signal. For biometric
detection, discarding time information is sometimes
important as this allows us to utilize techniques
adopted in other biometric modalities. This is the pur-
pose of our proposed loop generation phase. With our
loop generation, time series ECG graph is converted to
a two-dimensional loop. From this closed loop pattern,
features easily can be extracted like other popular bio-
metric mechanisms like finger print, iris, palm print,
face, etc. At the commencement of the loop generation
procedure, ECG data is differentiated first.
y(l)=x(n)−x(n−1) (5)
where, l=1,2,3,...,(M−1)
After obtaining vector x(original ECG in millivolt
range) and vector y(differentiated ECG), loop gener-
ation is plotted as a scatted xy graph. Therefore, the
X-axis of the graph holds all the mV ranges ECG
Figure 5. Two types of authentication mechanism for cardiovascular patient.
1356 Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd.
DOI: 10.10 02/sec
F. Sufi, I. Khalil and I. Habib Diagnosis of remote cardiovascular patients
Figure 6. Matching Process (method 1 as an example) in ECG
Cardioid-based Biometric.
amplitudes (vector x) and Y-axis of the graph holds
differentiated ECG (vector y). Figure 8 shows the orig-
inal ECG on the left side and the generated loop on the
right side. It is clearly seen that with the loop gener-
ation process, time information is not retained (ECG
is time varying signal). Our experimentation with 30
randomly selected ECG entries from MIT BIH reveal
all the loops are quite different. This the basis of our
research in person identification with ECG based bio-
metric, presented in this paper.
rLoop Segregation: A loop is defined as a curvature
that originates from a particular point and ends at the
same point. Therefore, in our experimentation, we used
a .Net program to detect a drawing point (in pixel)
that was previously painted. By doing so, in a window
sliding fashion, the program can efficiently detect the
previous point (or the originator of the loop) and the
current point (or the end point of the loop) as these two
points are being placed at the same pixel location (or
in a close proximity).
4.1. Method 1: ECG based person
identification with centroid, four extremas,
area, and perimeter as template
The loop resulting from QRS complex appears as the shape
of cardioid as seen in Figure 8. From the equation of the
cardioids (Equation (6)), the area (A) and the perimeter (U)
can be calculated from Equations (7) and (8).
r=2a[1 +cos(t)] (6)
A=1
22π
0
r2dϕ=..
=2a22π
0
(1 +cos ϕ)2dϕ=..
=6πa2(7)
U=2π
0dr
dϕ 2
+r2dϕ=....
=2√2aπ
0(1 +cos ϕ)dϕ=....
=16a(8)
However, the loops generated from P wave and T wave
appears to be ellipse. The equation for cardioid is given as
x2
a2+y2
b2=1 (9)
Here, ais called vertex or major axis and bis co-vertex or
minor axis. The vertex and the co-vertex for P wave loop are
apand bp. On the other hand, the vertex and the co-vertex
for T wave loop are atand bt. Our initial experimentation
shows that a, for T wave is more than three times a, for P
wave. Therefore, at>3ap
Area, Aand Perimeter (circumference), Ucan be calcu-
lated (or approximated) with the following equations.
A=π×a×b(10)
U≈π3(a+b)−(3a+b)(a+3b)(11)
Calculation of Uis based on Ramanujan’s ellipse approx-
imation. Uapproximation for ellipse with U≈2πa2+b2
2
is unsuitable for T wave, since b>3a, according to our
experimentation.
Centroid is created by the following equation, for all the
loops (QRS Complex, P wave and T wave).
centroid =N
i=1xi
N,N
i=1yi
N(12)
To calculate the extreme points as shown in Figure 8,
intersections of the coordinates (transformed) and approxi-
mated equation of the original shape (loop) are required to
be calculated. However, estimations of the exact equations
that represent the loops are computationally expensive for
mobile and embedded devices using existing curve fitting
techniques.
The procedures involved in Method 1 is summarised in
Figure 9. Advantage of Method 1, is its accuracy, with possi-
ble lower misclassification rate, as this method has two extra
template parameters (Area and perimeter). However, it has a
higher computational expense. This computational expense
is mainly caused by the custom equation based curve fit-
ting technique on the loop points. The average time taken
Figure 7. 3 Steps in biometric template creation for patient authentication.
Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd. 1357
DOI: 10.10 02/sec
Diagnosis of remote cardiovascular patients F. Sufi, I. Khalil and I. Habib
Figure 8. Cardioid based patient authentication and diagnosis system (Desktop implementation).
Figure 9. Block diagram of ECG based person identification with centroid, area and perimeter as template (Method 1).
from these custom equation based curved fitting was more
than 1.8 s on our desktop system. However, when imple-
mented on smart phones the calculation time was found
to be as high a 30 s. Therefore, for biometric authentica-
tion on remote telecardiology (involving mobile phone), a
computationally inexpensive method is required.
4.2. Method 2: ECG biometric based on
centroid and extreme points
This method, as depicted in Figure 10, is simpler com-
pared to Method 1 as it does not require computations
of Aand U. After generating the loops for the heart
Figure 10. Block diagram of ECG based person identification with centroid and four extremas.
1358 Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd.
DOI: 10.10 02/sec
F. Sufi, I. Khalil and I. Habib Diagnosis of remote cardiovascular patients
shape (originated from QRS complex), ellipse origi-
nated from T wave and ellipse originated from P wave,
15 points were initially selected for representing each
ECG sample. For the QRS loop the points are Cen-
troid (d1,c1), (fqrs(c1,y),c1), (−f qrs(c1,y),c1),
(d1,f qrs(c1,y)), (d1,−f qrs(c1,y)). For the
T wave loop the points are Centroid (d2,c2),
(fqrs(c2,y),c2), (−f qrs(c2,y),c2), (d2,f qrs(c2,y)),
(d2,−fqrs(c2,y)). For the P wave loop the points are
Centroid (d3,c3), (fqrs(c3,y),c3), (−f qrs(c3,y),c3),
(d3,f qrs(c3,y)), (d3,−f qrs(c3,y)).
The functions fqrs(x, y), f t(x, y ) and f p(x, y) are
all estimated functions, that can be approximated based on
different techniques. Curve fitting is one of the techniques,
which has been used in our earlier researches [15,22]. In
those previous researches, we have used the polynomial
coefficients as the ECG feature, which were compared to
detect human [15,22]. However, the major problem for
using coefficients as ECG feature is the size of the coef-
ficients. Multiple coefficients (as high as 32) increase the
overall template size of the feature set. Higher template
size requires higher computational expense as well as time
to perform comparison task for biometric identification.
Therefore, within this paper, we initially selected only five
points to represent shapes (Centroid and the four extreme
points for each of the shapes).
To save computational resources for resource limited
devices, approximation of the extreme points (Figure 11)
is crucial. Rather than approximating the equations of the
loops and finding the intersections with the coordinates (as
it was done for Method 1), four vertical extreme points can
be calculated for each of the loops, using the following rules.
Rule 1: Point p(xn,y
n) is the chosen to be upper extrema,
when yn>c and |(xn−d)|is minimum for all points
p(x, y) in the loop. In the same way for lower extrema,
yn<cand |(xn−d)|is minimum. Therefore: ∀p(x, y)∃yn:
|(xn−d)|
Rule 2: Point p(xn,y
n) is the chosen to be the right
extrema, when xn>dand |(yn −c)|is minimum for all the
points p(x, y) in the loop. For the left extrema xn<dand
|(yn−c)|is minimum. Therefore, ∀p(x, y)∃xn:|(yn−c)|
Based on these rules, Algorithms 1 and 2 were designed
and implemented.
From experimentations, we noted for the left and right
extreme points, yvalue is very close to c. This is because,
yis basically the change (derivative) in waveform and for
all the wave forms equivalent negative change is followed
by the positive change. From Figure 8, we can clearly see
that for all the waves (QRS Complex, P wave and T wave),
positive wave change is followed by equivalent negative
change. Left and Right extreme points are situated on the
vertically opposite side of the Centroid. And the yvalue
of the Centroid is essentially the average of change of the
waveform (which is near zero).
Hence, for all the loops, yvalues of Centroid, left and
right are least important. Yvalues for these three points
for all the loops can easily be omitted for generation of
template. The insignificance of these values (yvalues for
left and right extreme points) for identifying person also
become apparent by using Principal component analysis
(PCA).
Algorithm 1 Detection of upper and lower extreme points
of the loop
upperX, upperY, lowerX, lowerY, tempDistance
Loop for all points in the feature loop
if (xi−d)< tempDistance and yi>cthen
tempDistance =|(xi−d)|
upperX =xi
upperY =yi
endif
if (xi−d)< tempDistance and yi<cthen
tempDistance =|(xi−d)|
lowerX =xi
lowerY =yi
endif
End Loop
Algorithm 2 Detection of right and left point of the loop
rightX, rightY, leftX, leftY, tempDistance
Loop for all points in the feature loop
if (yi−c)< tempDistance and yi>dthen
tempDistance =|(yi−c)|
rightX =xi
rightY =yi
endif
if (yi−c)< tempDistance and yi<dthen
tempDistance =|(yi−c)|
leftX =xi
leftY =yi
endif
End Loop
4.3. Implementation and experimentation
results
We have implemented the cardioid based Biometric
authentication both on PC (desktop) and mobile phone envi-
ronment. Method 1 based biometric authentication took
about 24 s in PC based environment. This higher require-
ment of processing time is mainly because of the estimation
of the equation from the cardioid shape, calculation of area
and calculation of perimeter. To estimate the equation, curve
fitting tools were utilized similar to our previous research
[15,22]. After approximation of the equation of the cardioid
Area and Perimeter were calculated. Again, the calcula-
tion process for Area and Perimeter were quite different in
every individual, as the cardioid equation varied from per-
son to person. Higher computational complexity was the
Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd. 1359
DOI: 10.10 02/sec
Diagnosis of remote cardiovascular patients F. Sufi, I. Khalil and I. Habib
Table II. Uniqueness of cardioids of various MIT BIH entries.
Entry Centroid of enrolment QRS Location of enrolment Centroid of recognition QRS Location of recognition
16 265 (127, 100.8) [47:57] (128, 100.4) [1233:1243]
16 420 (146, 99.89) [66:75] (146, 100.11) [1226:1235]
16 773 (138, 100) [48:58] (138, 99.7) [982:992]
16 795 (151, 100.1) [10:20] (152, 100.2) [1198:1208]
17453 (143, 100.38) [89:102] (142:100.38) [1131: 1144]
man obstacle for implementing Method 1 based biometric
authentication on mobile environment.
In Method 1 (desktop system), area for the QRS loop
ranges from 42 to 117, while the perimeter lies between
24 and 44. T wave loop area spans between 1.5 and 7.8. T
wave loop perimeter ranges from 5.21 to 13. Similarly, P
wave loop area ranged from 2.355 to 9.42 and P wave loop
perimeter ranged from 2.13 to 8. When implemented on our
desktop system, Method 2 only consumed 300 ms and on
smart phones, it only took 2 s.
We have used the publicly available ECG entries from
MIT-BIH Normal Sinus Rhythm Database (nsrdb) to
show the difference in centroid across different individual
(Table II). Table II also shows the similarity of enrolment
centroid (for QRS only) and recognition centroid. As QRS
complex demonstrates the most uniqueness across individ-
uals, centroid of QRS are matched first. Then, the other
templates (centroid for T wave/P wave, four extremas)
are matched. However, it appears obvious that centroid of
QRS alone can uniquely identify persons. When other QRS
extremas are added as identification features, centroid and
extremas of P or T waves, and area of cardiods may not be
necessary.
4.3.1. Misclassification rate.
Both Methods 1 and 2 had only single misclassification
error (for each of them) as seen in Table III. However,
we believe that experimenting with a larger sample size
will reveal higher accuracy of Method 1 with the expense
of higher computational requirements. Table IV shows
the False Match Rate (FMR) and False Non-match Rate
(FNMR) wise comparison of Methods 1 and 2 with other
existing biometric techniques.
Table III. Misclassification rate for PRD, CC, WDM, and pro-
posed PDM.
Misclassification
Method rate (%)
PRD [3] 25
CC [3] 21
WDM [3] 11
PDM (without Alg. 1, without Alg.2) [22] 13.33
PDM (with Alg. 1, without Alg.2) [22] 6.66
PDM (with Alg. 1, with Alg.2) [22] 0
Proposed (Method 1) 1
Proposed (Method 2) 1
Table IV. FRM and FNMR across different modalities.
Modality FMR (%) FNMR (%) Reference
Face 1 10 [30]
Fingerprint 0.01 2.54 [31]
Iris 0.00129 0.583 [32]
On-line signature 2.89 2.89 [33]
Speech 6 6 [34]
ECG 6.66 6.66 PDM (without Alg. 1,
without Alg. 2) [22]
ECG 3.33 3.33 PMD (with Alg. 1,
without Alg. 2) [22]
ECG 0 0 PDM (with Alg. 1
+ with Alg. 2) [22]
ECG 0.5 0.5 Proposed (Method 1)
ECG 0.5 0.5 Proposed (Method 2)
4.3.2. Template size.
Shorter template size results in faster processing, during
one to many matches performed during person identifica-
tion. TableV shows difference in template sizes for different
ECG biometric.
If matching of 1 byte takes tbamount of time, then accord-
ing to Table V, Methods 1 and 2 consumes (69X tb) and
(63X tb) times respectively. Therefore, Method 1 is approx-
imately 4452 times faster and Method 2 is approximately
4876 times faster then Face [35] recognition template with
307 200 bytes of data. Clearly, the proposed Method 2
requires less storage and executes faster for person identifi-
cation task compared to the existing biometric mechanisms.
4.3.3. Authentication time.
To estimate how this cardiod based automated authen-
tication mechanism (Method 2) helps in mission critical
Table V. Comparison of template sizes.
Biometric data type Size in bytes
Iris [35] 512
Face [35] 153 600–307 200
Voice [35] 2048–10240
ECG [36] 600
ECG (WDM) [3] 1371
ECG (PRD / CC) [3] 2210
ECG (PDM) 340
ECG (Proposed Method 1) 69
ECG (Proposed Method 2) 63
1360 Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd.
DOI: 10.10 02/sec
F. Sufi, I. Khalil and I. Habib Diagnosis of remote cardiovascular patients
Table VI. Comparison of cardioid based automated authentica-
tion technique against user name/password based authentica-
tion (Times are in seconds. Cardioid biometric is performed on
randomly selected MIT BIH entries).
Novice users Moderate users Expert users Cardioid biometric
33.50 20.25 14.50 0.235
32.2 22.90 15.90 0.339
29.6 23.80 11.90 0.487
26.0 18.85 13.75 0.912
31.9 27.30 12.40 0.226
health application, we have compared this innovative
patient authentication mechanism with existing user-
name/password based telecardiology application. Three
different level of mobile phone users (novice, moderate
and expert) were put under the test of providing their user-
name/password pair to mimic a cardiovascular subscriber
establishing a connection with the service provider. Each
group consisted of five people. Table VI provides shows
the timing requirement of the three groups.
Apart from saving time, the obvious benefit from this
automated biometric scheme is the correctness of authenti-
cation process. If the cardiovascular patient suffers a sudden
attack, patient may enter wrong user name and password as
manual task becomes harder because of anomaly of auto-
nomic nervous system (ANS).
5. CARDIOID BASED DIAGNOSIS
We can employ this cardioid based technique for detect-
ing abnormal cardiac conditions. Any abnormality in ECG
trace can be either manually detected by the expert car-
diologists, or by automated algorithms. These algorithms
usually detect onset and offset of QRS complex, T wave
and P waves. After detection of these fiducial points (three
onsets and three offsets), width, amplitude and other param-
eters are calculated for each of the waves. Performing all
these calculations simultaneously in real time is resource
extensive. Therefore, most of the automated algorithms for
detecting heart abnormalities are designed for PC based
systems (or resource expensive ECG Acquisition devices).
Center of cardioid can play a significant role in instant detec-
tion of cardiovascular abnormality, as any sudden change
of ECG trace will cause instant change in the center of car-
dioid. As discussed earlier, calculation of center of cardioid
is a simple technique and resource efficient technique, that
can be easily implemented on mobile devices. This will help
a roaming cardiologist to instantly detect abnormalities of
cardiac conditions for ECG traces forwarded by the medical
server (medical server on the other hand, received this ECG
messages from the patient’s mobile phone). Experimenta-
tion with MIT BIH entries proves this point.
5.1. Implementation and experimentation
of cardioid based diagnosis
The diagnosis programs were implemented in Visual Stu-
dio .Net environment and tested on the pocket PC emulator
environment (as shown in Figure 12). After the successful
deployment on the emulator platform, the cardioid based
programs were deployed on HP 912 Business Messenger
Smart phone (Figure 13). Our real life experimentation
revealed that cardioid based authorization and diagno-
sis is the fastest solution for Mission Critical Alerting
mechanism, where every seconds counts towards saving
irrecoverable cardiac cell damage.
Figure 14 shows a sudden Ventricular beat (shown
by box) from record 803 of MIT BIH Supra-ventricular
Arrhythmia Database (svdb). This record corresponds to
only a subsection of the whole record (from 29:40 to
29:50 min), and a random window size of 100 samples (out
of the total 128 samples) is used for calculation of center
of cardioid. For this the normal beats the xcoordinate of
the centroid ranged from 150 to 153. On the other hand
the ycoordinate ranged from 99.77 to 100.29. One such
plot using normal beat is shown in Figure 18. However,
during the onset of the Ventricular beat the centroid was
(147, 99.73) as shown in Figure 19, when the centroid cal-
culation period was from 700 to 800. A sudden drop of
xcoordinate of the centroid identified sudden abnormality
within the ECG trace. This deviation in the values of cen-
troid clearly shows how normal heart conditions suddenly
changes. This form of presentation can help not only car-
diologists quickly diagnose cardiac abnormalities, but also
medical technicians at the hospitals. Similarly, the centroid
Figure 11. Calculation of Centroid and four extremas for QRS Complex, T Wave and P Wave.
Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd. 1361
DOI: 10.10 02/sec
Diagnosis of remote cardiovascular patients F. Sufi, I. Khalil and I. Habib
Figure 12. Mobile phone implementation of Cardioid (for ECG abnormality diagnosis within Doctors mobile phone) running in Pocket
PC emulator under MS Visual Studio 2005.
values calculated for sample range (800–900) and (810–
910) of the same record are (156, 100.35) and (156, 99.93).
These values are out of the normal ranges of centriod for
that person, and therefore, identify cardiac abnormality.
It should be noted that these values were calculated by
our software (both desktop as in Figure 13 and mobile as
in Figure 12). Figure 15 shows the entire ECG trace of
Figure 14 transformed into cardioid. Even though the plot
in Figure 15 is similar to the output of our software (both
desktop and mobile), our software has normalized the axes
(both X and Y) for obtaining integer values (as the pixels
in desktop and mobile screens are represented by integer
values). This fact (the difference in value range) is true
for the subsequent cardioid figures presented in this paper
(Figures 19–21).
Now if we look into more serious event of cardiac abnor-
mality, such as ventricularfibrillation (VF), we will also find
the center of cardioid method useful for identifying cardiac
abnormalities. We randomly selected an ECG trace contain-
ing VF onset to test the center of cardioid method. Figure 16
shows a 10 s segment (from 3:30 to 2:40 min) ECG record
(Entry no. cu01) from CU Ventricular Tachycardia Database
(cudb). The onset of VF is annotated and marked within
this trace. The normal, abnormal, and the entire section of
Figure 16 are represented by Figures 22–24.
For testing the center of cardioid method, we selected a
window size of 250 and calculated the center of cardioid
in a window sliding fashion. For all the four normal QRS
complexes of Figure 16, the centroid is (156 ±2, 100.03 ±
0.26) as shown in Figure 20. At the event of VF, the center
of cardioid changes to (160, 100.2) for segment 1100 to
1350 as visible in Figure 21.
Figure 13. Deployment of cardioid based patient authoriza-
tion and diagnosis on HP 912 Business Messenger mobile
Smartphone.
An ECG trace with 360 Hz sampling frequency and 1
minute duration spans horizontally and needs long scrolling
even on a 20 inch monitor with 1280 ×1024 resolutions.
Even if we use, a single pixel (of Mobile phone) to rep-
resent a single sample of ECG point, the mobile phone
screen needs to be 21 600 pixels wide to draw the 1 min
ECG curve (in a single screen) without the necessity of hor-
izontal scrolling (with a acquisition sampling frequency of
360 Hz). However, mobile phones used during our exper-
imentation had only 170 ˜240 pixels in width. Therefore,
the MIDlets were designed with horizontal scrolling facil-
ity to navigate forward the ECG with multiple screens
(Figure 17).
1362 Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd.
DOI: 10.10 02/sec
F. Sufi, I. Khalil and I. Habib Diagnosis of remote cardiovascular patients
Figure 14. Occurrence of Ventricular Beat from MIT-BIH Supraventricular Arrhythmia Database (svdb) entry no. 803.
Figure 15. Cardioid drawn from the entire ECG strip presented in
Figure 14 (from MIT-BIH Supraventricular Arrhythmia Database
(svdb) entry no. 803).
Further more, drawing ECG curve onto a mobile phone
screen in different than PC display, because of their variation
in coordinate system. For, for mobile phones the coordi-
nate (0, 0) starts from top left corner (Figure 22), unlike
Figure 18. Centroid Using Normal Beats of Figure 14.
bottom left corner for PC display units. Therefore, a coor-
dinate transformation is required for the ECG signal xn,
before drawing them on mobile phone screen. Equation
(13), demonstrates the transformation operation.
yn=h−xn(13)
where, his the height of supported pixels for
mobile phone Therefore, generating curves and graphs of
Figure 16. Occurrence of Ventricular Fibrillation (Ventricular beat annotated by VF) from CU Ventricular Tachyarrhythmia Database
(cudb) entry no. 01.
Figure 17. Drawing of ECG Curve in Multiple Screens of a Mobile Phone.
Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd. 1363
DOI: 10.10 02/sec
Diagnosis of remote cardiovascular patients F. Sufi, I. Khalil and I. Habib
Figure 19. Centroid Using of Abnormal Beats (i.e. during the
onset of the Ventricular beat) of Figure 14.
Figure 20. Centroid Using Normal Beats of Figure 16.
Figure 21. Centroid Using of Abnormal Beats during Occurrence
of Ventricular Fibrillation of Figure 16.
biosignals needs proper calculation and translation before
drawing them on to the mobile phone’s screen. Figure 22
shows the end result of graphing (schematic). From Fig-
ure 22, makes it obvious that a particular ECG is can be
Figure 22. Cardioid drawn from the four normal beats of
Figure 16.
spanned to Wnumber of screens, where Wis determined
by Equation (14).
W=ceil(f×t(w−1)) (14)
where, ceil denotes the ceiling operation, fis the sampling
frequency of ECG acquisition and tis the total duration of
the ECG trace.
If the mobile phone used by the doctor has moder-
ately higher resolution and supports 240 pixel in width
then 1 minute ECG (250 sampling frequency) requires Ceil
(15 000/239) or 63 screen. Therefore, if the doctor intends to
browse through the entire 1 min ECG trace, then he requires
at least 62 clicks on the mobile phone. Now if each click con-
sumes 0.5 s of delay, the mobile cardiologist needs at least
31 s (and can lead upto several minutes taking consideration
of the decision time taken by the cardiologist for each of
the screens), just to draw the complete ECG trace. Viewing
the ECG will take further time on top of the drawing time.
However, using our cardioid based diagnosis approach,
this time (i.e., clicking and viewing of the screens) could
be minimized to less than 0.5 s, as no clicking operation
is required by the cardiologist. The entire signal can be
plotted on a single screen. On that screen, the abnormal
ventricular beats will be clearly misaligned. As seen in the
cu 1 record (in Figure 20) that after first four normal beats,
VF event started (approximately from sample no. 1026). If
the cardioid is drawn from 0 to 1026 sample, we would see
the regular (normal) beat pattern for that person as seen in
Figure 16.
However, just after the initiation of the deadly VF event,
the cardioid takes a totally different pattern as shown in Fig-
ure 17. It is evident from the pictures (Figures 16 and 17)
that the regular beat and VF beat doesn’t share the same pat-
tern. If a cardiologist is given with the cardioid drawn from
the entire 10 s ECG trace (from sample 1 to sample 2500)
for the same record, then he/she will see two different pat-
terns and will instantly identify occurrences of abnormality
(as depicted in Figure 18). In fact, automated algorithms
can be developed and deployed to notify any instance of
drawing on highlighted area (termed as abnormal area) on
the mobile phone screen (canvas class [9]). In the Figure
18, we can see the possible VF region marked with boxes.
Any drawing on that box signifies occurrence of Vetricular
abnormal beat.
Algorithm 3 performs the automated abnormality detec-
tion by detecting any drawing on a boxed region. Boxed
region (x1, y1, x2, y2) is area of the Cartesian coordi-
nate (either implemented on PC or Mobile Screen) that has
been marked as abnormal region. During the regular potting
activity (looping in Algorithm 3), if the current coordi-
nate (Cur Point.X, Cur Point.Y) is bounded with the boxed
region, and then abnormality is identified. More than one
boxed region can be created for identifying multiple cardiac
abnormality.
Figure 25 represents the first 1 min ECG data from AHA
Database—01. We can see that four abnormal beats are
highlighted with boxes. If the same ECG extract is used for
1364 Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd.
DOI: 10.10 02/sec
F. Sufi, I. Khalil and I. Habib Diagnosis of remote cardiovascular patients
Algorithm 3 Cardioid based automated ECG diagnosis
Cur Point, PrePoints //the points are in Cartesian
coordinate
Cur Point =(0,0)
Pre Point =(0,0)
Cur Point =GetCurrentPoint() // Using Eqs. (4–5)
Loop
if x1≤Cur Point.X ≤x2AND y1≤Cur Point.Y ≤
y2then
Abnormality Detected = TRUE
Count =Count +1
endif
Drawline (Pre Point, Cur Point)
Pre Point =Cur Point
EndLoop
Figure 23. Cardioid drawn after the occurrence of VF in
Figure 16.
generating a cardioid, all the abnormal beats will be clearly
identified as shown in Figure 26. Within the cardioid map
all the similar beat patterns share the same region of the
screen.
Similarly, if cardioid is drawn using the Supraventricular
Arrhythmia Database (svdb) entry no. 803 that is plotted in
Figure 14, then the event of ‘V’ or wide QRS complex can
be easily visible (Figure 15).
Beat alignment operation in time series (original ECG
trace) can also reveal abnormal beats from normal beats,
as seen from Figure 24. These types of beat alignment can
be performed with direct threshold based techniques [25],
transformational techniques [26] or ever other complex
techniques involving Artificial Intelligence. In Table VII,
we have performed beat alignments based on three existing
direct methods. These three methods, namely Amplitude
based technique (ABT), First Derivative based technique
(FDT) and Second Derivative based (SDT) beat detection
(for the experimentation in Table VII) is explained in detail
by earlier researchers [25]. Using these techniques, the QRS
complex is located first and then the actual alignment is
performed. Our cardioid based technique can offer beat
alignment facility on small handheld platform in an efficient
Figure 24. Cardioid drawn from the entire ECG trace presented
in Figure 16.
Table VII. Execution time (in seconds) on HP IPAQ 912 business
messenger for cardioid, first derivative based technique, second
derivative based technique, and threshold based technique.
ECG no. ABT FDT SDT Cardioid
100 3 16 17 1
102 4 16 1 8 1
105 3 16 19 1
111 3 16 1 8 1
114 3 16 1 8 1
201 4 16 18 1
210 4 17 18 1
manner. Implementing the most simple existing beat align-
ment techniques (ABT, FDT and SDT)[25] consumes at
least three times more computational time, compared to the
presented cardioid based diagnosis as seen from Table VII.
Table VII reflects the fact that our automated cardioid based
diagnosis system (Algorithm 3) is up to 19 times faster than
convensional methods such as SDT. Even without execut-
ing Algorithm 3, drawing cardioid on mobile phone allows
the cardiologist to instantly track abnormality, saving mul-
tiple clicks (as demonstrated earlier in this section with
Figure 17).
6. DISCUSSION
In this paper, we intend to reduce the delay in cardiac care in
two main areas: patient authentication and diagnosis. The
Authentication mechanism ensures security in Mission Crit-
ical Alert (MCA) Mechanism. The five actors within this
MCA are distributed in nature and can be located anywhere.
The communication framework is tied with HTTP/ MMS/
SMS/ Bluetooth/ WiFi as seen from 1. Therefore, the pre-
sented MCA uphold elastic and distributed network. In case,
HTTP fails, patient’s mobile to hospital communication can
be executed on MMS or SMS [5], for ensuring reliabil-
ity. Within the mobile phone based wireless cardiac care
Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd. 1365
DOI: 10.10 02/sec
Diagnosis of remote cardiovascular patients F. Sufi, I. Khalil and I. Habib
Figure 25. Occurrences of Ectopic beats / premature beats (ectopic beats marked with red boxes and a suspected beat marked with
blue box) from AHA Database entry no. 01.
Figure 26. Cardioid drawn from the entire ECG strip presented
in Figure 25.
solution as seen in 1, we can observe the following five key
actors:
rPatient: This patient is monitored with portable ECG
acquisition devices. Patient is the center (or key player)
within the mobile phone based patient centric solution.
rPatient’s Mobile Phone: Patients mobile phone is
serves as the communicator between the patient and
Figure 27. ’Beat alignment in time series domain (from Refe-
rence [15]).
the hospital/cardiologist/ambulance service provider.
Also, this mobile phone performs repeated detection
facility that continuously searches for abnormality
within the patient’s ECG trace. This is done via cal-
culating the centroid of that particular ECG trace
and measuring the centroid against a set threshold
(as shown in 1). Any centroid outside the thresh-
old range signals a possible abnormality and starts
communication with the hospital. The mobile phone
also performs local authentication (before local mon-
itoring) as well as remote authentication (before
connecting and informing the hospital about an abnor-
mal event) as shown in Figure 25.
rHospital: Hospital provides the cardiovascular mon-
itoring facility to the patient, who subscribed for the
continuous monitoring facility. The hospital runs its
existing algorithms to check the validity and serious-
ness of the cardiac abnormality. In case of serious
abnormality the hospital informs the cardiologist and
the ambulance for rescuing the patient.
rAmbulance: The ambulance locates the patient by
retrieving GPS locations from the patient’s mobile
phone (a listener within the patient’s mobile phone
is responsible as in 1). Within the ambulance, a 12
lead ECG acquisition is performs and transmitted to
the hospital via HTTP or Socket routine. The Hos-
pital then views the ECG and decides on the action
plan (e.g., Surgery in catheterization lab and there-
fore, activates the catheterization lab) for the patient
along with cardiologist’s decision. The ambulance is
instructed accordingly about where to admit the patient
(e.g., catheterization lab).
rMobile cardiologist: The hospital can also receive
expert opinion on special cases from a remote car-
diologist. The cardiologist receives ECG information
in compressed and encrypted format either from the
patient or from the hospital. Cardiologist’s mobile
phone executes the cardioid based diagnosis program,
which assists in faster diagnosis of the abnormal event.
The ECG transmission from the patient to the hospi-
tal/cardiologist or from the ambulance to the cardiologist
is performed utilizing compression and encryption tech-
nology described in References [5,13]. Therefore, fast
and secured transmission is guaranteed upholding Health
1366 Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd.
DOI: 10.10 02/sec
F. Sufi, I. Khalil and I. Habib Diagnosis of remote cardiovascular patients
Insurance Portability and Accountability Act (HIPAA) act
of US (1996). The application compression and encryp-
tion technology is not the topic of this paper and our recent
research clearly reveals details of the technology [5,13]. The
value added to the MCN alert mechanism by this paper, is
by the innovative utilization of cardioid based authentica-
tion and diagnosis method and by strategically placing them
on the existing cardiac care scenario [1,3,4,8,11]. With this
piece of research in place even the fastest timing (22 min
in Reference [3]) in patient rescue can be reduced farther.
Moreover, the scenario (in 1) and the architecture (in Fig-
ure 1) outlined with this paper addresses following two main
criteria (as urged by Reference [37]) for reducing door-to-
balloon time:
rInnovative, standardized protocols
rData feedback to monitor progress and identify prob-
lems or successes
7. CONCLUSION
According to the cardiovascular experts the delay in diagno-
sis can cause significant and ever lasting damage to patient’s
heart and drastically increase the chance of reduced life
span [1,2]. Therefore, minimizing the delay in cardiovas-
cular patient care is a global urge, with cardiovascular
disease being the number one killer of modern era. Within
this paper, we endeavored in minimizing the cardiovascu-
lar patient care by harnessing the modern technological
settlements of wireless communication and portable ECG
sensors. By identifying two specific long haul bottlenecks
(authentication and diagnosis) in mobile phone based car-
diovascular patient monitoring, we have shown that several
minutes delay can be reduced to a mere 0.5 s. Moreover,
our ECG based biometric method is up to 4876 times faster
than existing biometrics systems like Face recognition (due
to lower template size). Utilizing this faster method of per-
son identification (ECG biometric) presented in this paper,
other areas demanding mission critical operations can also
benefit (e.g., military authentication). Apart from minimiz-
ing the delay in CVD patient care, we have also depicted a
structured process of alerting mechanism which might just
save your life.
REFERENCES
1. Otsuka Y, Yokoyama H, Nonogi H. A novel mobile
telemedicine system for real-time transmission of out-
of-hospital ecg data for st-elevation myocardial infarc-
tion. Catheterization and Cardiovascular Interventions,
2009.
2. Luca GD, Suryapranata H, Ottervanger JP, Antman EM.
Time delay to treatment and mortality in primary angio-
plasty for acute myocardial infarction: every minute of
delay counts. Circulation 2004; 109: 1223–1225.
3. Sillesen M, Sejersten M, Strange S, Nielsen S, Lip-
pert F, Clemmensen P. Referral of patients with
st-segment elevation acute myocardial infarction directly
to the catheterization suite based on prehospital tele-
transmission of 12-lead electrocardiogram. Journal of
Electrocardiology 2008; 41(1): 49–53.
4. Ortolani P, Marzocchi A, Marrozzini C, et al. Useful-
ness of prehospital triage in patients with cardiogenic
shock complicating st-elevation myocardial infarction
treated with primary percutaneous coronary intervention.
The American Journal of Cardiology 2007; 100(5): 787–
792.
5. Sufi F, Fang Q, Khalil I, Mahmoud SS. Novel methods
of faster cardiovascular diagnosis in wireless telecardiol-
ogy. IEEE Journal on Selected Areas in Communications
2009; 27(4): 537–552.
6. Sufi F, Fang Q, Mahmoud S, Cosic I. A mobile
phone based intelligent telemonitoring platform. 3rd
IEEE/EMBS International Summer School on Medical
Devices and Biosensors, 2006, September 2006; 101–
104.
7. Sufi F, Fang Q, Cosic I. Ecg r-r peak detection on mobile
phones. 29th Annual International Conference of the
IEEE Engineering in Medicine and Biology Society, 2007
(EMBS 2007), August 2007; 3697–3700.
8. Lee R-G, Chen K-C, Hsiao C-C, Tseng C-L. A mobile
care system with alert mechanism. Information Technol-
ogy in Biomedicine, IEEE Transactions on 2007; 11(5):
507–517.
9. Sufi F. Mobile phone programming java 2 micro edi-
tion. Proceedings of the 2007 International Workshop on
Mobile Computing Technologies for Pervasive Health-
care, Philip Island, Melbourne, December 2007; 64–
80.
10. Sufi F, Khalil I, Fang Q, Cosic I. A mobile web grid
based physiological signal monitoring system. Inter-
national Conference on Technology and Applications
in Biomedicine, 2008 (ITAB 2008), May 2008; 252–
255.
11. Khalil I, Sufi F. Mobile device assisted remote
heart monitoring and tachycardia prediction. Inter-
national Conference on Technology and Applications
in Biomedicine, 2008 (ITAB 2008), May 2008; 484–
487.
12. Sufi F, Fang Q, Cosic I. A mobile phone based intelligent
scoring approach for assessment of critical illness. Inter-
national Conference on Technology and Applications in
Biomedicine, 2008 (ITAB 2008), May 2008; 290–293.
13. Sufi F, Khalil I. Enforcing secured ecg transmission for
realtime telemonitoring: a joint encoding, compression,
encryption mechanism. security and communication net-
works. Security and Communication Networks 2008;
1(5): 389–405.
Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd. 1367
DOI: 10.10 02/sec
Diagnosis of remote cardiovascular patients F. Sufi, I. Khalil and I. Habib
14. Sufi F, Khalil I. A new feature detection mechanism and
its application in secured ecg transmission with noise
masking. Journal of Medical Systems 2009; 33(3): 121–
132.
15. Sufi F, Khalil I, Habib I. Polynomial distance measure-
ment for ecg based biometric authentication. Security and
Communication Networks, 2009 (in press).
16. Sufi F, Mahmoud S, Khalil I. A novel wavelet packet
based anti spoofing technique to secure ECG data. Inter-
national Journal of Biometrics 2008; 1(2): 191–208.
17. Cardionet: Get to the Heart of the Problem, February
2007. Available online at: http://www.cardionet.com/
18. Hamilton PS, Tompkins WJ. Quantitativeinvestigation of
qrs detection rules using the mit/bih arrhythmia database.
Biomedical Engineering, IEEE Transactions on 1986;
BME-33(12): 1157–1165.
19. Akselrod S, Gordon D, Ubel FA, Shannon DC, Barger
AC, Cohen RJ. Powerspectrum analysis of heart rate fluc-
tuation: a quantitative probe of beat to beat cardiovascular
control. Science 2007; 213(1981): 220–222.
20. Bartolo A, Clymer B, Burgess R, Turnbull J, Golish J,
Perry M. An arrhythmia detector and heart rate estima-
tor for overnight polysomnography studies. Biomedical
Engineering, IEEE Transactions on 2001; 48(5): 513–
521.
21. Kumar M, Weippert M, Vilbrandt R, Kreuzfeld S, Stoll
R. Fuzzy evaluation of heart rate signals for mental stress
assessment. Fuzzy Systems, IEEE Transactions on 2007;
15(5): 791–808.
22. Sufi F, Khalil I. An automated patient authentication sys-
tem for remote telecardiology. International Conference
on Intelligent Sensors, Sensor Networks and Informa-
tion Processing, 2008 (ISSNIP 2008), December 2008;
279–284.
23. BM, et al. Remote health-care monitoring using personal
care connect. IBM Systems Journal 2007; 46(1): 95–
113.
24. Hamilton P, Tompkins W. Compression of the ambulatory
ecg by average beat subtraction and residual differencing.
Biomedical Engineering, IEEE Transactions on 1991;
38(3): 253–259.
25. Friesen G, Jannett T, Jadallah M, Yates S, Quint S,
Nagle H. A comparison of the noise sensitivity of nine
qrs detection algorithms. Biomedical Engineering, IEEE
Transactions on 1990; 37(1): 85–98.
26. Kim B, Yoo S, Lee M. Wavelet-based low-delay ecg
compression algorithm for continuous ecg transmission.
Information Technology in Biomedicine, IEEE Transac-
tions on 2006; 10(1): 77–83.
27. Miaou S-G, Lin C-L. A quality-on-demand algorithm for
wavelet-based compression of electrocardiogram signals.
Biomedical Engineering, IEEE Transactions on 2002;
49(3): 233–239.
28. Velasco MB, Roldan FC, Ferreras FL, Santos AB, Munoz
DM. A low computational complexity algorithm for ecg
signal compression. Medical Engineering and Physics
2004; 26: 553–568.
29. Khalil I, Sufi F., Cooperative Remote Video Consultation
on Demand for e-Patients, Journal of Medical Systems,
33(6): 475–483, DOI: 10.1007/s10916-008-9208-y, 2009
30. Phillips P, Grother P, Micheals R, BoneBlackburn D,
Tabassi E, Bone M. Facial recognition vendor test 2002.
Evaluation Report, March 2003. Available online at:
http://www.frvt.org/ [Last Accessed in January 2009].
31. Maio D, Maltoni D, Cappelli R, Wayman J, Jain A.
Fv c2004: third fingerprint verification competition.
Proceesings of the First International Conference on Bio-
metric Aunthentication, Vol. 3072, 2004; 1–7.
32. Group IB. Independent testing of iris recognition tech-
nology. Final Report, NBCHC030114/0002, May 2005.
33. Yeung D-Y, Chang H, Xiong Y, et al. Sv c2004:
first international signature verification competition. In
Proceedings 2004 Biometric Authentication: First Inter-
national Conference (ICBA 2004), Hong Kong, China,
July 2004; 16–22.
34. Reynolds D, Campbell W, Gleason T, et al. The 2004 mit
lincoln laboratory speaker recognition system. In Pro-
ceedings IEEE International Conference on Acoustics,
Speech and Signal Processing (ICASSP 2005), Philadel-
phia, USA, March 2004; 177–180.
35. Yu F, Tang H, Leung V, Liu J, Lung C. Biometric-based
user authentication in mobile ad hoc networks. Security
and Communication networks 2008; 1: 5–16.
36. Wubbeler G, Stavridis M, Kreiseler D, Bousseljot R-D,
Elster C. Verification of humans using the electrocar-
diogram. Pattern Recognition Letters 2007; 28: 1172–
2275.
37. Bradley E H, et al. Achieving rapid door-to-balloon
times: how top hospitals improve complex clinical sys-
tems. Circulation 2006; 113(8): 1079–1085.
1368 Security Comm. Networks 2011; 4:1351–1368 © 2011 John Wiley & Sons, Ltd.
DOI: 10.10 02/sec