Accepted Manuscript

Cloud-assisted Industrial Internet of Things (IIoT)- enabled

framework for health monitoring

M.Shamim Hossain, Ghulam Muhammad

PII: S1389-1286(16)30001-9
DOI: 10.1016/j.comnet.2016.01.009
Reference: COMPNW 5803

To appear in: Computer Networks

Received date: 2 August 2015

Revised date: 6 January 2016
Accepted date: 8 January 2016

Please cite this article as: M.Shamim Hossain, Ghulam Muhammad, Cloud-assisted Industrial In-
ternet of Things (IIoT)- enabled framework for health monitoring, Computer Networks (2016), doi:
10.1016/j.comnet.2016.01.009

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service
to our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and
all legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Cloud-assisted Industrial Internet of Things (IIoT)- enabled framework for health

monitoring

M Shamim Hossaina,∗, Ghulam Muhammadb

a Software Engineering Dept, College of Computer and Information Sciences, Riyadh, Saudi Arabia
b Computer Engineering Dept, College of Computer and Information Sciences, Riyadh, Saudi Arabia

T
IP
Abstract
The promising potential of the emerging Internet of Things (IoT) technologies for interconnected medical devices

CR
and sensors has played an important role in the next-generation healthcare industry for quality patient care. Because
of the increasing number of elderly and disabled people, there is an urgent need for a real-time health monitoring
infrastructure for analyzing patients’ healthcare data to avoid preventable deaths. Healthcare Industrial IoT (HealthI-
IoT) has significant potential for the realization of such monitoring. HealthIIoT is a combination of communication

US
technologies, interconnected apps, Things (devices and sensors), and people that would function together as one
smart system to monitor, track, and store patients’ healthcare information for ongoing care. This paper presents a
HealthIIoT-enabled monitoring framework, where ECG and other healthcare data are collected by mobile devices and
sensors and securely sent to the cloud for seamless access by healthcare professionals. Signal enhancement, water-
AN
marking, and other related analytics will be used to avoid identity theft or clinical error by healthcare professionals.
The suitability of this approach has been validated through both experimental evaluation, and simulation by deploying
an IoT-driven ECG-based health monitoring service in the cloud.
Keywords: ECG monitoring, IoT-driven healthcare, cloud-assisted system, signal watermarking, Healthcare
M

Industrial Internet of Things (HealthIIoT)

ED

1. Introduction monitors, ECG monitors, and blood pressure monitors)

to avoid preventable death due to hospital or other re-
Today we are witnessing the increased use of smart lated errors. The errors may occur before, during, or
devices and communication apps in healthcare monitor- after hospitalization.
ing, and their influence on the activities of healthcare
PT

Currently, HealthIIoT is still in its preliminary stages

professionals (doctors, nurses, and hospital administra- with regards to design, development, and deployment;
tors), patients, and the healthcare industry. According however, IoT-based solutions are presently displaying a
to Gartner and Forbes, it is estimated that by 2020, the remarkable impact, and carving out a growing market
CE

Internet of Things (IoT) will contribute $1.9 trillion to in today’s healthcare industry and tomorrow’s emerg-
the global economy and $117 billion to the IoT-based ing IIoT-based healthcare monitoring solutions. IIoT
healthcare industry [1]. Based on this estimate, it is ex- has the potential to save 50,000 people each year in the
pected that the Healthcare Industrial IoT (HealthIIoT) US by avoiding preventable deaths due to hospital error
AC

will be one of the main players in the Industrial Internet [2]. It promises patient well-being and safety by coor-
of Things (IIoT)-driven healthcare industry. IIoT has dinating critical patient information and synchronizing
had a remarkable influence across many large and small related resources (e.g., healthcare staff, facilities, wear-
healthcare industries . As a result, an increasing num- able smart devices to capture real-time patient data such
ber of wearable IoT devices, tools, and apps are being as vital signs, and patient-related electronic informa-
used for different monitoring applications (e.g., glucose tion) instantly through interconnected devices and sen-
sors. Research reveals that IoT in the healthcare indus-
∗ Correspondingauthor
try can facilitate better care with reduced costs, reduced
Email addresses: [email protected] (M Shamim direct patient-healthcare staff interaction, and ubiqui-
Hossain ), [email protected] (Ghulam Muhammad) tous access to quality care [3]. Mohammed et al. de-
Preprint submitted to Computer Networks February 4, 2016
 ACCEPTED MANUSCRIPT

veloped a remote patient monitoring system using web • the introduction of a watermark into the ECG sig-
services and cloud computing [4]. Hassanalieragh et nal on the client side, to avoid a security breach in
al. discussed the opportunities and challenges of health the cloud
monitoring and management using IoT [5]. To date,
however, no comprehensive study has been published • the introduction of a user identification code to pro-
about cloud-assisted IIoT-driven health monitoring. vide customized protection of data
Safe and high-quality healthcare service is of
• the introduction of a one-class support vector ma-
paramount importance to patients. Accordingly, health-
chine classifier in the cloud
care data security and patients’ privacy are important is-
sues that will have a great impact on the future success

T
The rest of this paper is organized as follows. Sec-
of HealthIIoT [17]. One of the major issues in the IIoT-
tion 2 reports some related studies. Section 3 outlines
based healthcare system is the protection of privacy. In

IP
the HealthIIoT ecosystem, followed by a high-level data
general, a healthcare service provider receives data from
flow architecture for the HealthIIoT monitoring value
its users (such as patients) and shares them with regis-
chain, and the details of a cloud-supported HealthIIoT-

CR
tered clinics or healthcare professionals. The provider
enabled monitoring framework. Section 4 describes a
may also distribute the data to health insurance compa-
proposed health monitoring approach by considering
nies and pharmaceutical companies. Moreover, patient
ECG as healthcare data. Section 5 presents the exper-
data can be vulnerable to hackers during cloud transfer
imental results and evaluations. Section 6 concludes the
or synchronization with interconnected devices.
Therefore, we need to protect this information from
unauthorized access, which may result in the posting
of personal information in the public domain, or in in-
US paper.

2. Related Studies
AN
terference with essential medical equipment, such as a
pacemaker. A security breach of a patient’s monitoring The IIoT is an innovative technology, directly inter-
devices and data may cause the patient social embar- connecting a set of sensors and devices (such as smart-
rassment, mental disorders, or adverse physical effects phones) to collect, record, transmit, and share data for
such as a fatal heart attack. Hence, data protection in the possible analysis. The IoT has a wide range of emerg-
M

form of watermarking and authentication is very impor- ing applications [4]-[16]. Among them, the most revo-
tant in an IIoT-based healthcare system. lutionary potential application is healthcare monitoring,
To this end, this paper describes an IIoT-based health where patient healthcare data are collected from a num-
ED

monitoring framework, where health monitoring signals ber of sensors, analyzed, delivered through a network
are authenticated. As a case study, we have used electro- and shared with healthcare professionals for evaluation
cardiogram (ECG) monitoring, as ECG is an important of patient care [4], [5], [6], [10], [11], [13], [23]. A
assessment tool. By continuously monitoring an ECG more comprehensive survey of IoT for healthcare ap-
PT

signal, a healthcare professional can diagnose disease plications can be found in [7]. IoT-enabled healthcare
and prescribe medications to avoid preventable death . applications, including IoT-driven ECG monitoring, are
ECG signals are recorded via portable ECG recording discussed in the following studies [6], [10]. Li et al.
devices at home or outdoors, and sent to smartphones or [10]. presents a health monitoring service as a plat-
CE

desktops via Bluetooth technology . On the client side, a form for ECG monitoring using adaptive learning anal-
smartphone app or desktop software removes unwanted ysis model to detect abnormalities.
noise from the recorded signal, and embeds a water- Mohammed et al. developed a remote patient moni-
mark for security and authentication purposes; heartbeat toring system using web services and cloud computing
AC

is also monitored using a simple algorithm . The water- [4]. In particular, they designed an Android applica-
marked ECG signal is then transmitted to a cloud server, tion for ECG data monitoring and analysis. Data can
where temporal and spectral features are extracted and be further analyzed by third-party software if needed;
classified using a one-class support vector machine clas- however, there is no option for the cloud server to ex-
sifier. The classification decision, along with the water- tract features and classify the signal to assist the health
marked ECG, is passed to the desired healthcare profes- professional at the time the signal is received. In our
sional for analysis. The professional then sends back a proposed framework, the cloud server extracts features
decision and prescription to the cloud server. The cloud and classifies the signal, so that a preliminary analysis
then notifies the patient . The contributions of the paper decision from the cloud can be sent to the healthcare
are as follows: professionals to facilitate good patient care.
2
 ACCEPTED MANUSCRIPT

Hassanalieragh et al. discussed the opportunities and ical data are opened via malware or file sharing tools
challenges of health monitoring and management us- intentionally or unintentionally. They offered some so-
ing IoT [5]. Some challenges include slow processing, lutions to these threats, including biometric cryptogra-
handling big data, presence of too much heterogeneous phy and an advanced signal processing scheme; how-
data, and data integrity. In our proposed framework, ever, the authors did not implement these solutions in
the ECG signal is watermarked on the client side before their paper.
transmitting through the Internet to authenticate against
any attacks. Data processing is also distributed between
the client side and the cloud side to make the overall
processing faster.
Research

T
center

Jara et al. [8] present a remote monitoring framework Emergency

using IoT by proposing a protocol, called YOAPY, to

IP
response service
Patient with IoT
create a secure and scalable fusion of multi-modal sen- devices and sensors

sors to record vital signs. A cloud-based speech and

CR
face recognition framework was developed to monitor a Social media
and family
patient’s state remotely [27]. Xu et al. [9] developed a Healthcare
Storage
Analysis Replica ECG
ubiquitous data accessing method in an IoT-based sys- professionals data center verification

tem for emergency medical scenarios. They proposed a

semantic data model to store data, and a resource-based
data access method to gain control of the data ubiqui-
tously, concluding that their method could be significant
to assist decision-making in emergency medical situa-
US Smart medical device
and ‘Things’ industries
Environmental and
home control service
(e.g. temp, air, light)
Drug industries, and
pharmacies (e.g. smart
pills, and medicines)
AN
tions.
Zhang et al. [12] introduced an architecture of mobile Figure 1: Conceptual illustration and scenario for HealthIIoT ecosys-
healthcare networks, incorporating privacy-preserving tem.
data collection and secure transmission. The privacy-
preserving data collection was achieved using cryptog-
M

raphy with secret keys and private keys. Secure trans- In [19] and [21], emotion-aware or affective mobile
mission was gained using attribute-based encryption, computing frameworks have been proposed, and the
where only authorized users would have access to the authors investigated an architecture named “emotion-
ED

data. These methods are generally worthwhile; how- aware mobile cloud” (EMC) for mobile computing. Au-
ever, the main problem is computation complexity. thors in [20] proposed another framework, affective in-
Granados et al. [14] proposed web-enabled gateways teraction through wearable computing and cloud tech-
for IoT-based eHealth with an option for wired or wire- nology (AIWAC). Recently, Hu et al. [6] introduced the
PT

less services. To take advantage of wired gateways Healthcare Internet of Things (Health-IoT), attempting
in terms of power-efficiency and low cost, the authors a bridge between intelligent health monitoring and emo-
used the wired gateways in a small room or building, tional care of the patient.
where movement is restricted. Radio frequency iden- To date, we have found no comprehensive study on
CE

tification (RFID)-based eHealthcare systems were pro- cloud-assisted IIoT-driven ECG monitoring, where i)
posed in [15] and [16]. In [15], the authors proposed the ECG signal is watermarked on the client side before
a system that would capture the patient’s environmental transmission through the Internet to the cloud, and ii)
conditions, such as temperature and humidity, by RFID, the cloud server extracts features and classifies the sig-
AC

and transmit them to the cloud for a more detailed un- nal to assist healthcare professionals in providing qual-
derstanding of ambient conditions. Catarinucci et al. ity patient care.
[16] proposed an IoT-aware architecture to monitor and
assess a patient’s situation automatically by integrating 3. Proposed cloud-assisted HealthIIoT-enabled
ultra-high-frequency RFID functionality. monitoring framework
Sawand et al. [18] identified three types of threats in
an eHealthcare monitoring system. These are identity HealthIIoT can revolutionize today’s healthcare in-
threat, where the identity of the patient is lost or stolen, dustry with affordable and quality patient care by adopt-
access threat, where an intruder can access the system ing a large number of interconnected machines, wear-
illegally, and disclosure threat, where confidential med- able things (devices and sensors), and cloud-computing
3
 ACCEPTED MANUSCRIPT

T
IP
CR
US
Figure 2: Data flow architecture for HealthIIoT monitoring value chain.
AN
technologies to collect patient data in a seamless man- connected sensor devices [22]. The success of HealthI-
ner. This HealthIIoT technology will play an impor- IoT largely depends on the advancement of the cloud-
tant role in a number of health monitoring applications, computing technology and big data analytics. It cre-
to form a Healthcare Industrial IoT ecosystem. Fig- ates a platform for interconnected smart medical de-
M

ure 1 describes a comprehensive IIoT-driven health- vices to operate with large amounts of data (big data)
care ecosystem. As shown in Figure 1, one type of from anywhere at any time. The data are actually gener-
stakeholder (e.g., patient with IoT devices and sen- ated by a myriad of interconnected smart devices, com-
sors, healthcare professional, hospital or medical re- munication apps, and their usage in healthcare monitor-
ED

search center, social media and family) is connected to ing applications. Data are gathered and analyzed from
another type of stakeholder (e.g., emergency response e-health records, imaging equipment, medical sensors,
services, drug industries and pharmacies, smart medi- devices, and smartphones over the cloud. This analy-
cal devices and ’Things’ industries, environmental and sis augments the decision-making power of healthcare
PT

home control services) to form a complex HealthIIoT professionals, and helps patients have an active role in
ecosystem. It also dispatches emergency services to managing their personal health.
the patient when needed, and orders pharmacy refills. Figure 2 outlines how the flow of a patient’s health-
CE

In this ecosystem, interconnected ‘Things’ (medical de- care data (e.g., ECG signals) is captured securely; how it
vices and sensors) are coordinated. It allows fast trans- is transferred seamlessly through a connection gateway
fer of patient information among the stakeholders in a to the cloud data centers for further analysis and pro-
secured manner, such that specific patient data are avail- cessing, such as feature extraction, classification, veri-
AC

able only to a designated authorized healthcare team. fication, workload measurement, and big data manage-
Finally, cloud-based big data analytic enables analyz- ment. After being processed and securely stored in the
ing, storing, closely monitoring, and securely sharing cloud, the chain of collected data is either accessed by
the data for further review and medical recommenda- healthcare professionals, or delivered to external sys-
tions, aimed towards fulfilling the promise of Industrial tems for further industry-specific healthcare IoT solu-
IoT in regard to quality patient care, real-time patient tions.
monitoring, and avoiding hospital error.
Figure 3 shows the cloud-assisted IIoT-enabled
The Industrial IoT is the combination of big data, health monitoring framework. First, the patient’s ECG
IoT, Machine to Machine (M2M) communication, cloud signal is recorded through the connected devices and
computing, and real-time analysis of data from inter- sensors, and then after possible signal reconstruction,
4
 ACCEPTED MANUSCRIPT

enhancement, and watermarking, it is sent to the cloud- HealthIIoT manager: Manages all health-related
based system using network connections. The cloud IIoT data by assigning data to different replicated data
system validates this information to check that the pa- centers, after verifying the authenticity of the data.
tient’s information is correct, and then extracts some Monitoring and analytics: Analyzes the data by ex-
features, classifies the signal, and redirects it to the as- tracting features and applying classification techniques.
signed healthcare professionals and providers for possi- Calculates and monitors the workload of the framework,
ble patient care. such as storage and bandwidth.
The major components of the framework are de- Replica service: Because of the growing demand
scribed below. for interconnected medical devices with heterogeneous
Healthcare staff and other related stakeholders: connectivity, HealthIIoT systems handle large numbers

T
Patients upload their ECG readings through an ECG of data requests for accessing patient healthcare. There-
fore, datasets need to be replicated in multiple sites and

IP
interface, which is connected to the Internet. After
some processing, the ECG reading is stored in the cloud data centers to offer faster data access times. If one or
database, where healthcare professionals can access it more sites (data centers) are down, healthcare data can

CR
and review it for possible action based on the uploaded be accessed from other nearby sites. Generating repli-
ECG readings. cas also enables the healthcare professionals’ ECG file
ECG signal capturing and recording service: This requests to distribute the workload through the replica
service is used to record and store the ECG signal from servers, and avoid performance degradation due to net-
different devices and smartphones.
Secure transmission service: The service enables a
secure and authenticated transmission of the ECG signal
through Internet. To accomplish this, watermarking is
US work congestion. Moreover, this ensures faster access,
scalability, and a reduction in response time.
IIoT-driven healthcare service directory: Records
and stores data from the ECG capturing devices. It also
AN
embedded into the signal, and later, is extracted to verify registers and publishes different participating services.
the authenticity. It facilitates continuous care for the patients by record-
Resource allocation manager: Manages virtual ma- ing the ECG signal in portable ECG recording devices
chine (VM) resources and web services. at home or outdoors, and sends them to smartphones
or desktops. Some major elements in this directory are
M

Cloud system manager: Controls all VMs and allo-

cates suitable resources through the resource allocation the ECG capturing service, feature extraction and ECG
manager for each service, such as ECG signal and things classification service, secure transmission service, and
collection and record service manager, ECG monitoring payment service. Healthcare professionals can get ac-
ED

session manager, feature extraction and classification cess to the ECG data from this directory without visiting
manager, and finally, signal reconstruction, enhance- the patient.
ment and watermarking manager.
4. Proposed health monitoring approach
PT

1. ECG signal and things collection and record ser-

vice management: This web service is respon- Our proposed health monitoring approach consists of
sible for managing the users’ data and their re- some processing steps, which are signal enhancement,
lated health information, and storing them in the watermarking, feature extraction, ECG analysis, and
CE

database. signal reconstruction. Signal enhancement and water-

marking are done on the client side. The work flow of
2. ECG monitoring (analysis and communication
the proposed framework is shown in Figure 4.
alerts) session management: Responsible for man-
AC

aging and controlling the sessions, in addition to 4.1. Signal enhancement

locating, tracking, and evaluating the activities.
An ECG monitoring system based on the cloud was
3. Feature extraction and classification management: proposed by Pandeya et al. [24]. In their system, ECG
This web service extracts collected data upon run- data were collected by mobile devices and were sent to
ning the ECG apps on smartphones, and stores the cloud for analysis. The system was just a proto-
them in a MySQL database before sending them type, and therefore, problems remained in its fully prac-
to the HealthIIoT cloud. tical usage in term of data collection and transmission.
4. Signal reconstruction, enhancement, and water- The first issue is to ensure the effectiveness of ECG data
marking: This web service generates, records, and collection through mobile devices. Physiological arti-
tracks the performance of the monitoring function. facts can be caused by muscular activities that result
5
 ACCEPTED MANUSCRIPT

of low-pass filtering and moving average filtering of a

recorded ECG signal. From the figure, we see that the
signal looks ’clear’ after applying these two filters. This
preprocessing step is necessary to correctly detect elec-
trical waves in the ECG signal, which is critical for sub-
sequent online analysis and monitoring.

T
IP
CR
Figure 3: The proposed HealthIIoT architecture for an ECG monitor-
ing.

in small spikes, and by human motion that results in

US Figure 5: Effect of low-pass filtering (b) and moving average filtering
(c) of the recorded ECG signal (a).
AN
large swings in the recorded data. Non-physiological
artifacts can be produced by electrical interference and 4.2. Peak R detection
electrode malfunction. Electrode malfunction is initi- To process the ECG signal, determining peak R is re-
ated by loose connections, electrode misplacement, low quired. These two attributes in ECG signal are very im-
M

amount of electrode gel, wrong filter setting, fractured portant for ECG signal analysis. To detect R, we use
wires, etc. Of them, electrode misplacement is a major analytical wavelet transform (AWT) [29]. A complex
source of malfunction of ECG data acquisition [25]. Ca- AWT can be expressed by Eq. 1.
ble misplacement can even result in ECG that resembles
ED

cardiac abnormalities like ectopic rhythm [26]. φ(ω) = ωn e−ω , ω≥0

(1)
=0 , ω<0
The above equation is obtained by combining a
Hilbert transform and a wavelet transform. In Eq. 1, n is
PT

the derivative order. The advantage of using the above

equation is that it combines the time-frequency location
of a wavelet transform with the local slope information
CE

of the Hilbert transform.

After applying complex AWT, we apply thresholding
and look for persistent lines. The local maximums are
Figure 4: Overall work flow of the ECG monitoring system in the identified in the wavelet transform signal to find R (see
AC

cloud.
Figure 6).

Therefore, in the proposed framework, the recorded 4.3. ECG watermarking

ECG signals are enhanced before processing to get rid The next stage is to watermark the signal to protect it
of some of the common artifacts. In the enhancement from forgery. Watermarking is a procedure to embed
stage, the recorded ECG signal is passed through a low- some information in a signal without distorting the visi-
pass filter to suppress the high frequency components bility or credibility of the signal for ownership claim. It
that are referred to noise. A 25-point moving average has mainly two parts: watermark embedding and water-
filter is then applied to the output of the low-pass fil- mark extraction for verification. Embedding watermark
ter to smooth the signal. Figure 5 shows the effect in the ECG signal will ensure the authenticity of the
6
 ACCEPTED MANUSCRIPT

Step 2. Apply SVD on the normalized matrix. The re-

sultant Sw is a square matrix of size I × I.

Im = Uw · S w · VwT

Step 3. Multiply Sw by a watermark intensity factor, α.

S wα = α · S w

Step 4. Store U w , V w T , and α for watermark extraction.

T
Use Swα for watermark embedding.
Step 5. Divide the ECG signal into N number of beats.

IP
The beat duration is 0.6 seconds and peak R is
Figure 6: Detected peak R. located at 40% of the duration.
Step 6. Step 6: Apply two-level DWT on each beat.

CR
Take H2 (detail coefficients at level 2) for wa-
termark embedding. Store L2 (approximation
ECG signal transmitted over the cloud. We adopt a sim- coefficient) and H1 (detail coefficients at level
ple yet efficient strategy of watermarking that has rarely 1) for reconstruction of watermarked ECG sig-
been used for ECG signals. The watermarking is based
on discrete wavelet transform (DWT)-singular value de-
composition (SVD) [26] [28]. The study in [26] [28]
did not use DWT-SVD watermarking scheme in ECG
US nal.
Step 7. Form a matrix G using H2 of all the frames.
The number of rows corresponds to the number
of beats of the signal.
AN
signals, e-healthcare or cloud-based systems. DWT is
a multi-resolution technique that decomposes a signal Step 8. Apply SVD on matrix G. The resultant S s is a
into different levels of time and frequency. In the pro- square matrix of size N × N.
posed step, we use a two-level DWT that decomposes
G = U s · S s · V sT
the signal into three subbands: approximation subband
M

L2, second-level detail subband H2, and first-level de- Step 9. A new matrix, Snew , of size N × N is formed by
tail subband H1. SVD is a matrix factorization tech- using matrices Sw α and S s .
nique that decomposes a matrix into three matrices. If a (
ED

rectangular matrix A of size I × J is the input, the out- S wα (n, n) + S s (n, n), 1≤n≤I
S new =
put will be two orthogonal matrices and one diagonal S s (n, n), (I + 1) ≤ n ≤ N
matrix as follows (Eq. 2):
Step 10. Using U s , V s T , and Snew , perform inverse SVD
PT

T
AI×J = U I×I S I×J V J×J (2) to get matrix G’.
T T
where U U = II×I and V V = I J×J , which means G0 = U s · S new · V sT
that U and V are orthogonal. S is a diagonal matrix,
CE

whose diagonal entries are singular values and arranged Step 11. Using L2, G’, and H1, perform inverse DWT to
in descending order. These singular values are always get watermarked ECG signal.
real numbers. The computation of SVD is stable against
round-off errors. The fact that a slight variation in the The watermark image can be reconstructed by using
AC

values of S matrix does not affect the perception of an the following steps.
ECG signal, watermark bits can be added to the singular Step 1. Subtract S s from Snew to get Sim . Sim should be
values of S to get a robust watermarking. equal to Sw if the watermarked speech signal is
The watermark is an image consisting of the client’s not under attack.
registered ID in image format of size I × J, where I > J..
The detailed procedure is described as follows. S im = S new (n, n) − S s (n, n), 1 ≤ n ≤ I

Step 1. Normalize the watermark image matrix by 255. Step 2. Apply inverse SVD to get the normalized wa-
n o termark.
Imi, j = watermarki, j /255; 0 ≤ i ≤ I, 0 ≤ j ≤ J Im0 = Uw · S im · VwT

7
 ACCEPTED MANUSCRIPT

Step 3. Get the watermark image by multiplying the 4.5. One-class support vector machine (OCSVM) clas-
values by 255 and dividing byα. sification
n o OCSVM is a one-class classification technique, where
Im0i, j = Im0i, j × 255/α; 0 ≤ i ≤ I, 0 ≤ j ≤ J the feature space is mapped to a higher dimensional
space so that a hyperplane maximizes the distance be-
If there is no attack, Im’i, j will be the same as tween the hyperplane and the origin. In a typical SVM,
the watermark image. there are two classes, while in OCSVM, there is only
one class in the training. OCSVM is used in this frame-
The enhancement stage and the watermark stage are work, because it can detect the personal ECG as abnor-
performed on the client side. Once the ECG signal is

T
mal or not. For the experiments, MIT-BIH database was
watermarked signal is transmitted to the cloud, where it used [30]. The first group of 48 records are divided into
is processed to extract features and to analyze.

IP
two sets: first set comprises of first 3 minutes of data
in each record for training, while second set consists of
4.4. Feature extraction remaining 27 minutes of data in each record for testing.

CR
Figure 7 shows the separation of the modules in the
Several features are extracted from the ECG signal in client side and the cloud side.
the cloud server. The features include heartbeat rate
(HBR), durations of P wave, PR interval, QRS com-
plex, and QT interval, and the shape (inverted or not,
and peaked or not) of T wave. These are total of seven
features.
The HBR is determined by the inverse of the time dif-
US
AN
ference between RR intervals and expressed as beats per
minutes (bpm) as expressed by Eq. 3.

60
HBR(bpm) = (3)
M

RR intervals (sec.)
An unusual p-wave may represent ectopic atrial pace-
maker. p-wave longer than 80 millisecond (ms) can in-
ED

dicate atrial enlargement. A PR interval smaller than Figure 7: ECG signal modules on the client side and the cloud side.
120 ms may be a cause of Wolf-Parkinson-White syn-
drome, while larger than 200 ms may indicate a first
degree of atrioventricular block. QRS complex wider
PT

than 120 ms suggests a disruption of the heart’s conduc-

tion system, or sever hyperkalemia. A prolonged cor- 5. Experimental results and evaluation
rected QT interval (> 440 ms) risks for ventricular tach-
yarrhythmia, while an unusual short interval may indi- Several experiments were performed to validate the
CE

cate severe hypercalcemia. For a corrected QT interval, proposed IoT-enabled ECG signal monitoring. They are
QT interval should be normalized by the square root of described in details in the followings subsections.
the RR interval. Inverted T waves may be syndromes
of myocardial ischemia, or high intracranial pressure, 5.1. Watermarking performance
AC

while a peaked (determined by the variance of the wave) The performance of the ECG watermarking was mea-
T wave may indicate hyperkalemia or early myocardial sured in terms of imperceptibility, and robustness
infarction. against attacks [31]. Imperceptibility is a measure
Spectral features are also calculated by applying the of how much the signal is distorted perceivably. To
Fast Fourier Transform (FFT) to each beat. A 512-point measure imperceptibility, we used signal-to-noise ratio
FFT is used and first half of the magnitude output is re- (SNR), which is an objective measurement. SNR is de-
tained. The 256 bins are linearly resampled to F number fined by Eq. 4.
of bins, where F is varied to 10, 20, 30, and 40. These F
number of features are appended to previously mention Ps
S NRdB = 10 log10 (4)
seven features for classification. P s − P0s

8
 ACCEPTED MANUSCRIPT

where, P s and P s are the power of original ECG sig-

nal and the watermarked ECG signal, respectively. An-
other closely related metric is peak SNR or PSNR. In
PSNR, the numerator of logarithm in Eq. 4 is replaced
by the square of the maximum value of the pixel in the
original watermark image.

T
IP
CR
Figure 8: PSNR (dB) and SNR (dB) using the proposed DWT-SVD
based watermarking, which is compared with DWT-DCT based wa-
termarking. US Figure 10: Classification accuracy of the proposed framework.

With regard to robustness against attack, we con-

AN
sidered two common attacks, which are additive white
Gaussian noise (AWGN), and filtering of type low-pass,
high-pass, and band-pass. The measurement were ob-
tained by using a correlation factor, η, which is com-
puted by using Eq. 5.
M

N
P
wi w0 i
i=1
ED

0
η(w, w ) = s s (5)
N
P N
P
wi 2 w0 i 2
i=1 i=1
PT

where, w and w’ are the original and extracted wa-

termark, respectively, N is the number of pixels in the
watermark image. η takes the value between 0 (no rela-
tion) and 1 (perfect relationship).
CE

Figure 8 shows the obtained SNR and PSNR in

dB using the proposed DWT-SVD based watermark-
ing. The proposed scheme was compared with an-
other popular method, which is DWT-DCT (discrete co-
AC

sine transform)-based [32]. From the figure, we see

that the proposed watermarking has a very good SNR
Figure 9: Correlation factor after different types of attack. and PSNR, and it outperforms the DWT-DCT based
method.
Figure 9 shows correlation factor, η, after different
types of attacks. The attacks were applied in the cloud
server. The attacks included band-pass filtering with
passband between 8 Hz and 40 Hz, high-pass filtering
with cutoff frequencies of 40 Hz, low-pass filtering with
cutoff frequencies of 8 Hz, and 4 Hz, and AGWN of 20
9
 ACCEPTED MANUSCRIPT

in the classification section, and the other with actual

data recorded through the proposed framework. Figure
10 shows the average classification accuracy using the
two databases. Seven features correspond to features
without spectral features. With 37 features, the accu-
racy reached up to 87.7% with MIT-BIH database and
90.4% with private database. Figure 11 shows the time
spent from transmitting the data to the cloud, extracting
features, to classifying the data. One instance, three in-
stances, and fives instances of servers were used in the

T
Figure 11: Time requirement of the proposed framework. cloud. With five instance server, only around two sec-
onds were needed while using 37 features.

IP
5
x 10
3
dB, 15 dB, and 10 dB. From the figure, we see that al-

CR
Without Rep.
2.9
most in all the cases the correlation factor was 1, which With Rep.

indicates the robustness of the proposed watermarking 2.8

Data access times (sec)

algorithm. For a comparison with DWT-DCT based 2.7

method, the correlation factor of DWT-DCT based wa- 2.6

termarking is 0.98 for AWGN 20 dB attack.

Table 1: Storage Configurations.

US 2.5

2.4

2.3
AN
2.2
Config. Node Capac- Total Stor- Relative 2.1
ity (GB) age (TB) Storage 2
1 2 3 4 5
(%) Storage configuration
1 [50-1000] 20.75 80
(a) Zipf.
M

2 [50-500] 15 60
3 [50-200] 12 46.75
5
x 10
5.2

4 [20-50] 4.5 18 Without Rep.

5 [10-20] 3.5 15 5 With Rep.
ED

Data access times (sec)

4.8

5.2. Classification performance 4.6

PT

4.4
5
x 10
5.5
Without Rep. 4.2
5 With Rep.
CE

4.5 4
Data access times (sec)

1 2 3 4 5
Storage configuration
4

3.5 (b) Gaussian

AC

3
Figure 13: ECG data access time comparison for all storage configu-
2.5 ration.

2
Zipf Guassian. Random
ECG data access types
5.3. Workload of the IIoT-enabled health monitoring
Figure 12: ECG data access time using relative storage capacity 80%. service
To evaluate the proposed IIoT-based for health mon-
itoring, we used a Java-based simulator program. The
Two types of classification experiments were per- simulation environment includes cloud topology and an
formed: one with the MIT-BIH database as mentioned ECG data access pattern by a healthcare professional.
10
 ACCEPTED MANUSCRIPT

To reduce overhead and latency, ECG files are repli- 1. Shorter access times for all storage configurations are
cated so that healthcare professionals can get access to correlated with using the replication approach.
the desired data for a specific patient from neighboring
data centers. We have used a similar multi-tier cloud
80
topology structure [33] with multiple data centers using
CPU Utilization Bandwidth Utilization Memory Utilization
the following storage configurations of replica servers 70

as shown in Table 1, where the capacity of relative stor- 60

age ranges from 15% to 80%. While submitting a task,
a number of ECG files are requested for access to pa- 50

Resource (%)
tient data. The sequence of file request is handled by

T
40
three access patterns, such as Zipf distribution, random
30
distribution, and Gaussian distribution. The framework

IP
is evaluated in terms of ECG data access by healthcare 20
professionals. The access time refers to the time of com-

CR
10
pletion of all tasks for the ECG file requests. The objec-
tive is to reduce ECG data access time. In the simula- 0
0 15 30 45 60 75 90 105 120
tion, constant data access rate is considered. The total Time (s)
data access time is a summation of time needed by the
(a) ECG Transmission
disk to find a replica of an ECG file from the replica
data center disk and network communication latency for
replica transmission. The data access time, Access(R),
is calculated using the following model, Eq. 6:
US 80

70
CPU Utilization Bandwidth Utilization Memory Utilization
AN
60

X 50
Resource (%)

Access(R) = f (v, n) · Dlatency (v, r) + Gnetwork (v, r)

v∈G 40
(6)
M

30
where, G = (V, L) is undirected tree structure of the
cloud, R is a set of replica data centers, n is an ECG file 20
or sample, f (v, n) is data access frequency by a health- 10
ED

care professional (v) for an ECG file (n), r is lowest

ancestor of v in R, Dlatency is disk access latency, and 0
0 15 30 45 60 75 90 105 120
Gnetwork is network communication latency. Time (s)

We compared ECG data access time by integrat- (b) ECG Capturing

PT

ing replication and without considering replication ap-

proaches to the proposed framework. Figure 12 com- Figure 14: Workload for ECG Monitoring.
pares the ECG data access times by considering repli-
cation and non-replication strategies for the first stor- Figure 14 shows the workload of the main services
CE

age configuration, as shown in Table 1. In the major- used for health monitoring of the proposed HealthIIoT
ity of cases, a storage configuration with replication re- framework. We have concentrated on three key ser-
sulted in a shorter ECG data access time than a con- vices: ECG capturing service, transmission service, and
figuration without replication. Replicated datasets in extraction or classification service. To understand the
AC

several data center locations decrease the data access features of those workloads as they relate to ECG mon-
time because healthcare professionals have access to itoring, the run-time characteristics of those workloads
the required ECG data for a particular patient from a are collected by running the proposed health monitor-
nearby site. Moreover, the storage capacity of the data ing prototype on the Amazon Elastic Computing Cloud
centers favors using the replication approach. With a (EC2). For this purpose, we rented a VM with an
reduced storage capacity, the ECG data access times Intel R CoreT M 2 Duo, DDR3 ECC RAM at 2.53 GHZ,
for both approaches (i.e., with and without replication) 1 Gbps bandwidth, and 4.0 GB memory, running Win-
are increased, but by different magnitudes. Figure 13 dows Server 2010. The performance monitor of Win-
shows the execution times for Zipf and Gaussian distri- dows has been used to record the resource consump-
butions for all the storage configurations listed in Table tions of those workloads for memory, CPU, and network
11
 ACCEPTED MANUSCRIPT

bandwidth utilizations. [6] L. Hu, M. Qiu, J. Song, M. Shamim Hossain, “Software Defined
Healthcare Networks,” IEEE Wireless Communication magazine,
vol. 22, no. 6, Dec. 2015
6. Conclusion [7] S.M. Riazul Islam, D. Kwak, M. H. Kabir, M. Hossain, and K-S
Kwak, ”The Internet of Things for Health Care: A Comprehensive
IIoT-driven healthcare monitoring is an emerging Survey,” IEEE Access, vol.3, pp. 678-708, 2015.
[8] J. Jara, M. A. Zamora-Izquierdo, and A. F. Skarmeta, ”Intercon-
healthcare service that may potentially revolutionize nection framework for mHealth and remote monitoring based on
the healthcare industry in terms of improving access the Internet of Things,” IEEE J. Sel. Areas Commun., vol. 31, no.
to patient information, and offer quality patient care 9, pp. 47-65, Sep. 2013.
[9] B. Xu, L. D. Xu, H. Cai, C. Xie, J. Hu, and F. Bu, ”Ubiqui-
through continuous monitoring from anywhere at any
tous Data Accessing Method in IoT-Based Information System for

T
time, through a multitude of devices. With HealthIIoT, Emergency Medical Services,” IEEE Transactions on Industrial In-
healthcare professionals may be able to access patient formatics, Vol. 10, No. 2, May 2014.

IP
information, store it, and analyze it in a real-time man- [10] P. Y. Li, L. Guo, and Y. Guo, ”Enabling Health Monitoring as
a Service in the Cloud,” In Proc. IEEE Int. Conf. Conference on
ner to monitor and track the patient. However, inter- Utility and Cloud Computing, Oct. 2014, pp. 81-84.
connected wearable patient devices and healthcare data [11] M. Shamim Hossain, ”Cloud-supported Cyber-Physical Frame-

CR
(such as ECG signals) are subject to security breaches. work for Patients Monitoring,” IEEE Systems J., 2016,
To this end, this paper describes a cloud-integrated 10.1109/JSYST.2015.2470644
[12] K. Zhang, K. Yang, X. Liang, Z. Su, X. (Sherman) Shen, and H.
HealthIIoT monitoring framework, where healthcare H. Luo, ”Security And Privacy For Mobile Healthcare Networks:
data are watermarked before being sent to the cloud for From A Quality Of Protection Perspective,” IEEE Wireless Com-
secure, safe, and high-quality health monitoring. Fu-
ture work will involve testing the proposed HealthIIoT
monitoring framework for data security and notification
functions, as well as implementing a test trial with real-
US munications Magazine, pp. 104-112, August 2015.
[13] M. Shamim Hossain, and G. Muhammad ”Cloud-based Col-
laborative Media Service Framework for Health-Care,” Hindawi
International Journal of Distributed Sensor Networks, Vol. 2014
(2014), Article 858712, 11 pages
AN
world patients and health professionals. [14] J. Granados, A-M. Rahmani, P. Nikander, P. Liljeberg, and
H. Tenhunen, Web-Enabled Intelligent Gateways for eHealth
Internet-of-Things, R. Giaffreda et al. (Eds.): IoT360 2014, Part
Acknowledgment I, LNICST 150, pp. 248–254, 2015.
[15] S. Amendola, R. Lodato, S. Manzari, C. Occhiuzzi, and G. Mar-
M

rocco, ”RFID Technology for IoT-Based Personal Healthcare in

The authors extend their appreciation to the Deanship Smart Spaces,” IEEE Internet of Things Journal, Vol. 1, No. 2,
of Scientific Research at King Saud University, Riyadh, April 2014.
Saudi Arabia for funding this work through the research [16] L. Catarinucci, D. De Donno, L. Mainetti, L. Palano, L. Patrono,
group project no. RGP 228. M. L. Stefanizzi, and L. Tarricone, ”An IoT-Aware Architecture
ED

for Smart Healthcare Systems,” IEEE Internet of Things Journal,

2015. DOI 10.1109/JIOT.2015.2417684
[17] D. Meltzer, ”Securing the Industrial Internet of Things,” The
References
Information Systems Security Association (ISSA) Journal, pp. 24-
[1] TJ McCue, $117 Billion Market For Internet of 30, June 2015.
PT

Things In Healthcare By 2020, Forbes. Retrieved from [18] A. Sawand, S. Djahel, Z. Zhang, F. Naı̈t-Abdesselam, To-
http://www.forbes.com/sites/tjmccue/2015/04/22/117-billion- ward Energy-Efficient and Trustworthy eHealth Monitoring Sys-
market-for-internet-of-things-in-healthcare-by-2020 Dec. 2015. tem, China Communications, pp. 46-65, January 2015.
[2] Stan Schneider, How the Industrial Internet of Things Can Save [19] Chen, M., Zhang, Y., Li, Y., Mao, S., Leung, V.: EMC: Emotion-
CE

50,000 Lives a Year, Industrial Internet Consortium, Retrieved aware Mobile Cloud Computing in 5G. IEEE Network, 29(2):32-
from http://blog.iiconsortium.org/2015/01/how-to-industrial- 38, (2015).
internet-of-things-can-save-50000-lives-a-year.html. Dec 12, [20] Chen, M., Zhang, Y., Li, Y., Hassan, M., Alamri, A.: AIWAC:
2015 Affective Interaction through Wearable Computing and Cloud
[3] Bresnick, J. ”Healthcare Internet of Things Driving Global Technology. IEEE Wireless Communications Magazine, 22(1):20-
AC

Market Growth” Retrieved from http://healthitanalytics.com: 27, (2015).

http://healthitanalytics.com/news/healthcare-internet-of-things- [21] M. Chen, ”NDNC-BAN: Supporting Rich Media Healthcare
driving-global-market-growth, 2015. Services via Named Data Networking in Cloud-assisted Wireless
[4] J. Mohammed, A. Thakral, A.F. Ocneanu, C. Jones, C-H Lung, Body Area Networks”, Information Sciences,Vol. 284, No. 10, pp.
and A. Adler, ”Internet of Things: Remote Patient Monitoring Us- 142-156, Nov. 2014.
ing Web Services and Cloud Computing,” In Proc. IEEE Internet [22] M. Chen, J. Wan, S. Gonzalez, L. Xiaofei, V.C.M. Leung, ”A
of Things (iThings), 2014, pp. 256-263, Taipei, 1-3 Sept. 2014. Survey of Recent Developments in Home M2M Networks,” IEEE
[5] M. Hassanalieragh, A. Page, T. Soyata, G. Sharma, M. Aktas, G. Communications Surveys and Tutorials, vol.16, no.1, pp.98-114,
Mateos, B. Kantarci, and S. Andreescu, ”Health Monitoring and 2014.
Management Using Internet-of-Things (IoT) Sensing with Cloud- [23] D. Niewolny, ”How the Internet of Things Is Rev-
based Processing: Opportunities and Challenges,” In Proc. IEEE olutionizing Healthcare,” White paper, Available:
International Conference on Services Computing, pp. 285-292, cache.freescale.com/files/corporate/doc/.../IOTREVHEALCARWP.pdf,
2015. Accessed 31st July, 2015.

12
 ACCEPTED MANUSCRIPT

[24] S. Pandeya, W. Voorsluys, S. Niua, A. Khandoker, and R. Biography

Buyyaa, ”An autonomic cloud environment for hosting ECGE-
cosystem analysis in the design of open platform-based in-home M. Shamim Hossain is an Associate Professor of
healthcare terminals towards the Internet-of-Things data analy- SWE, CCIS, at King Saud University, Riyadh, KSA.
sis services,” Future Generation Computer Systems, 28 (2012) Dr. Shamim Hossain received his Ph.D. degree in Elec-
147–154.
[25] H. Tam and J. Webster, ”Minimizing electrode motion artifact
trical and Computer Engineering from the University
by skin abrasion,” IEEE Transactions on Biomedical Engineering of Ottawa, Canada. His research interests include se-
24 (1977) 134–139. rious games, cloud and multimedia for healthcare, big
[26] V. Batchvarov, M. Malik, and A. Camm, ”Incorrect electrode ca- data for multimedia, social media, and biologically in-
ble connection during electrocardiographic recording,” Europace 9
(2007) 1081–1090.
spired approach for multimedia and software system.

T
[27] M. Shamim Hossain and G. Muhammad, ”Cloud-Assisted He has authored and co-authored around 100 publica-
Speech and Face Recognition Framework for Health Monitoring,” tions including refereed IEEE/ACM/Springer/Elsevier

IP
Mobile Networks and Applications, vol. 20, no. 3, pp. 391-399, journals, conference papers, books, and book chapters.
2015.
[28] M. Ali and C. W. Ahn, ”An optimized watermarking technique He has served as a member of the organizing and tech-
based on self-adaptive DE in DWT–SVD transform domain,” Sig- nical committees of several international conferences

CR
nal Processing, vol. 94, pp. 545–556, 2014. and workshops. Recently, he received outstanding pa-
[29] H. Zhao, F. Wang, Z. Chen, and J. Liu, ”A Robust Audio Wa- per award from an IEEE Conference. He has served
termarking Algorithm Based on SVD-DWT,” Elektronika Ir Elek-
trotechnika, vol. 20, no. 1, pp. 75-80, 2014. as co-chair, general chair, workshop chair, publication
[30] J.M. Lilly and S.C. Olhede, ”On the analytic wavelet transform,” chair, publicity chair, and TPC for over 12 IEEE and
IEEE Trans. Information Theory, vol. 56, no. 8, pp. 4135-4156,
Aug. 2010.
[31] G. B. Moody and R. G. Mark, ”The impact of the MIT-BIH
arrhythmia database,” IEEE Engineering in Medicine and Biology
Magazine, vol.20, pp. 45-50, 2001.
US ACM conferences and workshops. He is on the edito-
rial board of Springer Multimedia tools and Applica-
tions (MTAP). Currently, he serves as a lead guest ed-
itor of IEEE Transactions on Cloud Computing, IEEE
AN
[32] J. Grody and L. Brutun, ”Performance Evaluation of Digital Au- Communication Magazine, Elsevier Future Generation
dio Watermarking algorithms,” Proc. the 43rd IEEE Midwest Sym-
posium on Circuits and Systems, pp. 456-459, 2000.
Computer Systems, Elsevier Computers and Electri-
[33] M. Shorfuzzaman, A. Alelaiwi, M. Masud, M. M Hassan, M. cal Engineering, Springer Multimedia tools and Appli-
Shamim Hossain,”Usability of a Cloud based Collaborative Learn- cations (MTAP), Springer Cluster Computing. Previ-
M

ing Framework to improve Learner’s Experience,” Computers in ously, he served as a lead guest editor of IEEE Transac-
Human Behavior Vol. 51, pp. 967-976, October 2015
tions on Information Technology in Biomedicine (cur-
rently JBHI), Springer Multimedia tools and Applica-
tions, and Hindawi International Journal of Distributed
ED

Sensor Networks. Dr. Shamim is a Senior Member of

IEEE and a member of ACM.
PT
CE

Ghulam Muhammad is an
AC

Associate Professor in the department of Computer En-

gineering, College of Computer and Information Sci-
ences at the King Saud University, Riyadh, KSA.
Dr. Ghulam received his Ph.D. in Electrical and
Computer Engineering from Toyohashi University and
Technology, Japan in 2006. His research inter-
ests include serious games, cloud and multimedia for
healthcare, resource provisioning for big data process-
ing on media clouds and biologically inspired ap-
proach for multimedia and software system, image
13
 ACCEPTED MANUSCRIPT

and speech processing. He has authored and co- au-

thored more than 120 publications including refereed
IEEE/ACM/Springer/Elsevier journals, conference pa-
pers, books, and book chapters.

T
IP
CR
US
AN
M
ED
PT
CE
AC

14