Internet of Things For Smart Healthcare: Technologies, Challenges, and Opportunities

Received October 2, 2017, accepted November 6, 2017, date of publication November 29, 2017,
date of current version December 22, 2017.

Digital Object Identifier 10.1109/ACCESS.2017.2775180
Internet of Things for Smart Healthcare:

Technologies, Challenges, and Opportunities
STEPHANIE B. BAKER 1, WEI XIANG2 , (Senior Member, IEEE), AND IAN ATKINSON3
1 College of Science and Engineering, James Cook University, Townsville, QLD 4811, Australia
2 College of Science and Engineering, James Cook University-Cairns Campus, Cairns, QLD 4870, Australia
3 eResearch Centre at James Cook University, Townsville, QLD 4811, Australia
Corresponding author: Stephanie B. Baker ([email protected])
ABSTRACT Internet of Things (IoT) technology has attracted much attention in recent years for its potential
to alleviate the strain on healthcare systems caused by an aging population and a rise in chronic illness.
Standardization is a key issue limiting progress in this area, and thus this paper proposes a standard model
for application in future IoT healthcare systems. This survey paper then presents the state-of-the-art research
relating to each area of the model, evaluating their strengths, weaknesses, and overall suitability for a
wearable IoT healthcare system. Challenges that healthcare IoT faces including security, privacy, wearability,
and low-power operation are presented, and recommendations are made for future research directions.
INDEX TERMS Biomedical engineering, body sensor networks, intelligent systems, Internet of
Things (IoT), communications standards, security, wearable sensors.
I. INTRODUCTION in [12], with some focus placed on communications.

Healthcare is an essential part of life. Unfortunately, the However, it is hard to draw an image of a complete system
steadily aging population and the related rise in chronic from this paper. Finally, in [9], sensing and big data manage-
illness is placing significant strain on modern healthcare ment is considered, with little regard for the network that will
systems [1], and the demand for resources from hospital support communications.
beds to doctors and nurses is extremely high [2]. Evidently, This paper therefore makes a unique contribution in that
a solution is required to reduce the pressure on healthcare it identifies all key components of an end-to-end Internet of
systems whilst continuing to provide high-quality care to Things healthcare system, and proposes a generic model that
at-risk patients. could be applied to all IoT-based healthcare systems. This is
The Internet of Things (IoT) has been widely identified vital as there are still no known end-to-end systems for remote
as a potential solution to alleviate the pressures on health- monitoring of health in the literature.
care systems, and has thus been the focus of much recent This paper further provides a comprehensive survey of
research [3]–[7]. A considerable amount of this research the state-of-the-art technologies that fall within the proposed
looks at monitoring patients with specific conditions, such model. Focus is placed on sensors for monitoring various
as diabetes [5] or Parkinson’s disease [6]. Further research health parameters, short- and long-range communications
looks to serve specific purposes, such as aiding rehabilitation standards, and cloud technologies. This paper distinguishes
through constant monitoring of a patient’s progress [7]. Emer- itself from the previous major survey contributions by consid-
gency healthcare has also been identified as a possibility by ering every essential component of an IoT-based healthcare
related works [8], [9], but has not yet been widely researched. system both separately and as a system.
Several related works have previously surveyed specific Further original contribution is made by placing focus
areas and technologies related to IoT healthcare. An extensive on LPWANs, highlighting their unique suitability for use in
survey is presented in [10], with focus placed on commer- IoT systems. The upcoming licensed-band standards, such as
cially available solutions, possible applications, and remain- NB-IoT, are compared with the competing unlicensed-band
ing problems. Each topic is considered separately, rather standards, with particular interest in suitability for healthcare
than as part of an overarching system. In [11], data mining, applications.
storage, and analysis are considered, with little mention of The remainder of this paper is structured as follows.
integration of these into a system. Sensor types are compared Section II investigates the field of Internet of Things, placing
2169-3536
2017 IEEE. Translations and content mining are permitted for academic research only.
VOLUME 5, 2017 Personal use is also permitted, but republication/redistribution requires IEEE permission. 26521
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
S. B. Baker et al.: IoT for Smart Healthcare: Technologies, Challenges, and Opportunities
focus on the provision of healthcare using IoT technologies. monitoring could be used to monitor non-critical patients
Section III examines common sensors that could be used in at home rather than in hospital, reducing strain on hospital
an IoT healthcare systems, presenting several state-of-the-art resources such as doctors and beds. It could be used to provide
sensors that have been developed in recent research. Section better access to healthcare for those living in rural areas,
IV reviews communications standards for both short- and or to enable elderly people to live independently at home
long-term communications, including a thorough analysis for longer. Essentially, it can improve access to healthcare
of the new NB-IoT standard for long-range machine-to- resources whilst reducing strain on healthcare systems, and
machine (M2M) communications. Section V discusses cloud can give people better control over their own health at all
technologies and the ways in which they can be used in times.
IoT healthcare systems. Section VI highlights areas requir- In fact, there are relatively few disadvantages of remote
ing further research and provides recommendations for this health monitoring. The most significant disadvantages
research. Section IX concludes the paper, summarizing the include the security risk that comes with having large
key findings and reiterating the areas where further research is amounts of sensitive data stored in a single database, the
required. potential need to regularly have an individual’s sensors recali-
brated to ensure that they’re monitoring accurately, and possi-
II. HEALTHCARE AND THE INTERNET OF THINGS ble disconnections from healthcare services if the patient was
The Internet of Things remains a relatively new field of out of cellular range or their devices ran out of battery. Fortu-
research, and its potential use for healthcare is an area still in nately, these issues are all largely solvable, and are already
its infancy. In this section, the Internet of Things is explored being addressed in the literature, as will be highlighted
and its suitability for healthcare is highlighted. Several pio- throughout the remainder of this paper. As progress continues
neering works towards developing healthcare IoT systems to be made to reduce the disadvantages, IoT-based systems
are discussed. Building on the recurring themes from these for remote health monitoring are becoming an increasingly
works, a generic and standardized model for future end-to- viable solution for the provision of healthcare in the near
end IoT healthcare systems is proposed, with the aim of future.
guiding the future development of such systems. As a result of the many benefits of remote health mon-
itoring, many recent researchers have identified the poten-
A. THE INTERNET OF THINGS tial of the Internet of Things as a solution for healthcare.
Many definitions of the Internet of Things exist, but at In several works, IoT healthcare systems have been developed
the most fundamental level it can be described as a net- for specific purposes, including rehabilitation, diabetes man-
work of devices interacting with each other via machine to agement, assisted ambient living (AAL) for elderly persons,
machine (M2M) communications, enabling collection and and more. While these systems have been designed for many
exchange of data [7], [10], [11]. This technology enables different purposes, they are each strongly related through
automation within a large range of industries, as well as their use of similar enabling technologies.
allowing for the collection of big data. Rehabilitation after physical injury has been a topic of
Hailed as the driver of the Fourth Industrial Revolu- particular interest for several researchers. In [7], a system has
tion [13], Internet of Things technology has already found been developed that generates a rehabilitation plan tailored to
commercial use in areas such as smart parking [14], precision an individual based on their symptoms. The patient’s condi-
agriculture [15], and water usage management [16]. Exten- tion is compared with a database of previous patients’ symp-
sive research has also been conducted into the use of IoT for toms, ailments, and treatments to achieve this. The system
developing intelligent systems in areas including traffic con- requires a doctor to manually enter symptoms, and approve
gestion minimization [17], structural health monitoring [18], the recommended treatment; in 87.9% of cases, the doctor
crash-avoiding cars [19], and smart grids [20]. agreed completely with the system, and no modifications
While the aforementioned fields appear vastly different were made to the treatment plan it proposed.
to healthcare, the research conducted within them verifies Meanwhile, in [21], mathematical models for the measure-
the plausibility of an IoT-based healthcare system. Existing ment of joint angles in physical hydrotherapy systems are
systems in other fields have proven that remote monitoring proposed, enabling the improvement of joint movement to be
of objects, with data collection and reporting, are achievable. tracked through therapy.
This can therefore be expanded and adapted for monitoring In [6], existing IoT technologies are evaluated for their
the health of people and reporting it to relevant parties such usefulness in a system for monitoring patients suffering from
as caretakers, doctors, emergency services, and healthcare Parkinson’s Disease. Their work concludes that wearable sen-
centers. sors for observing gait patterns, tremors, and general activity
levels could be used in combination with vision-based tech-
B. INTERNET OF THINGS HEALTHCARE nologies (i.e. cameras) around the home to monitor progres-
Research in related fields has shown that remote health sion of Parkinson’s Disease. Furthermore, the authors suggest
monitoring is plausible, but perhaps more important are the that machine learning could lead to enhanced treatment plans
benefits it could provide in different contexts. Remote health in the future.
26522 VOLUME 5, 2017

A practical system for the monitoring of blood-glucose Existing systems highlight that communications are also
levels in diabetic patients was proposed in [5]. This system essential for an Internet of Things healthcare system. In sev-
requires patients to manually take blood-glucose readings eral existing system models [5], [6], [22], short-range com-
at set intervals. It thereafter considers two kinds of blood- munications, such as Bluetooth, are suggested for transferring
glucose abnormalities. The first is abnormal blood-glucose sensor data to a smartphone to be processed. Long-range
levels and the second is a missed blood-glucose reading. communications such as LTE can then be used to transfer
The system then analyses the severity of the abnormality, the processed information from the patient to the healthcare
and decides who to notify; the patient themselves, caregivers provider, typically a doctor, through SMS or the Internet. The
and family members, or emergency healthcare providers such key limitation of this is that smartphones typically have lim-
as doctors. This system is practical and has been proven ited battery life, requiring frequent recharging; a patient with
realizable, though could be further improved by automating a flat battery would be a patient disconnected from health-
blood-glucose measurements. care providers. A low-powered node designed specifically for
A system aimed at detecting heart attacks was built using managing healthcare information would be preferable.
ready-made components and a custom antenna in [22]. Cloud storage capable of storing high volumes of varying
An ECG sensor is used to measure heart activity, which data was also shown to be essential to a big data healthcare
is processed by a microcontroller. This information is for- system by several previous works [9], [11], [24]. If even a
warded via Bluetooth to the user’s smartphone, where the thousand people wore a single pulse sensor that communi-
ECG data is further processed and is presented in a user cated hourly with a cloud storage database via an LPWAN,
application. The authors identify that developing heart attack there would be 168,000 new data points per week. This
prediction software would improve the system. Further number increases drastically as more people wear sensors
improvements could be made by measuring respiratory connected to the cloud storage framework, and as more kinds
rate, which is known to aid in the prediction of heart of sensors are introduced. Using the big data that will rapidly
attack [23]. form and continue to grow in cloud storage, machine learning
SPHERE [4] is a system under continuing develop- algorithms can be implemented in the high-computing envi-
ment that utilizes wearable, environmental, and vision-based ronment of the cloud. These algorithms could be designed to
(i.e. camera) sensors for general activity and health mon- mine through the large amount of data, identify previously
itoring purposes. The aim of this project it to allow older unknown disease trends, and provide diagnostics, treatment
and chronically ill patients to live in the comfort of their plans, and much more.
own homes, while their health continues to be monitored. Based on these recurring trends in the literature to date, we
This allows for intervention by caretakers and doctors if any propose and recommend a four-part model as highlighted in
issues arise. Researchers working on the project have iden- Figure 1 that will aid in the development of future Internet of
tified that machine learning would be beneficial for learning Things healthcare systems, discussed below. In the following
about conditions and for making decisions about the patient’s sections, each of the components of the proposed model are
healthcare. discussed in further detail. Existing works are presented and
evaluated in the relevant sections. Strengths and weaknesses
C. A MODEL FOR FUTURE INTERNET OF THINGS of the current technologies are presented, and recommenda-
HEALTHCARE SYSTEMS tions for future directions of research are provided.
After reviewing this wide range of existing IoT-based health-
care system, several requirements for the design of such
systems become apparent. Each of these papers emphasize 1) WEARABLE SENSOR & CENTRAL NODES
the use of sensors for monitoring patient health. All regard Wearable sensor nodes are those that measure physiological
wearable sensors, namely wireless and externally-wearable conditions. Recommended sensors are those that measure the
sensors, as essential to their respective systems. Several vital signs - pulse, respiratory rate, and body temperature -
works [4], [6] also suggest the use of environmental or vision- as these are the essential signs for determination of critical
based sensors around the home. However, this restricts the health. Further sensors that could be implemented are blood
usefulness of the system to one physical location. It would pressure and blood oxygen sensors, as these parameters are
be preferable to implement all essential sensors as small, often taken alongside the three vital signs. Special-purpose
portable, and externally wearable nodes. This would provide sensors such as blood-glucose, fall detection, and joint angle
patients with a non-intrusive and comfortable solution that sensors could also be implemented for systems targeting a
is capable of monitoring their health wherever they go. This specific condition.
would make patients more receptive to using health monitor- The central node receives data from the sensor nodes.
ing technology than they would be if implantable sensors or It processes this information, may implement some decision
cameras were required. Additionally, repairing or replacing making, and then forwards the information to an external
externally wearable nodes would be simple when compared location. A dedicated central node would be preferred to a
to implanted sensors or vision-based sensors installed in the smartphone as battery life could be improved by having only
home. functionality relevant to a healthcare IoT system.
VOLUME 5, 2017 26523

FIGURE 1. Overview of the proposed model.
2) SHORT-RANGE COMMUNICATIONS essential, as these ensure that the message sent is the same
For sensors to communicate with the central node, a short- as the message received. This is important in all healthcare
range communications method is required. There are sev- applications, but particularly in emergency situations. Lastly,
eral important requirements to consider when choosing a high availability is essential to ensure that messages will
short-range communications standard, including effects on be delivered at all times, regardless of where the patient is
the human body, security, and latency. physically located. Again, this is of particular importance to
The chosen method should have no negative effects on the time-critical applications, but is preferable for all systems.
human body, as any such effects could cause additional health
concerns for patients. It should also provide strong security 4) SECURE CLOUD STORAGE ARCHITECTURE &
mechanisms to ensure that sensitive patient data cannot be MACHINE LEARNING
accessed by an attacker. Finally, low-latency is essential for Medical information obtained from patients must be stored
time-critical systems, such as a system that monitors critical securely for continued use. Doctors benefit from knowing a
health and calls for an ambulance if the need arises. In such patient’s medical history, and machine learning is not effec-
systems, time delays could be the difference between life tive unless large databases of information are available to
and death. In applications that are not time-critical, low- it. Based on the literature, cloud storage is the most viable
latency would not need to be prioritized as highly, but is still method for storing data. However, providing accessibility
preferable. for healthcare professionals without compromising security
is a key concern [25], [26] that should be addressed by
3) LONG-RANGE COMMUNICATIONS researchers developing healthcare IoT systems.
Data obtained by the central node is not useful unless some- Additionally, machine learning has repeatedly been iden-
thing can be done with it. This data should be forwarded to a tified in the literature as a means for improving health-
database where relevant parties (such as caretakers or doctors) care systems [4], [6], [7], though it has not been widely
can securely access it. There are again several considera- explored. Machine learning offers the potential to iden-
tions when selecting a suitable long-range communications tify trends in medical data that were previously unknown,
standard for use in a healthcare system, including security, provide treatment plans and diagnostics, and give recommen-
error correcting capabilities, robustness against interference, dations to healthcare professionals that are specific to individ-
low-latency, and high availability. ual patients. As such, cloud storage architectures should be
As with short-range communications, strong security is designed to support the implementation of machine learning
important to ensure that sensitive patient data remains pri- on big data sets.
vate and cannot be altered or imitated. Low-latency is again
important in time-critical applications, such as emergency D. POTENTIAL USE CASES FOR THE PROPOSED MODEL
healthcare, where delays in communication could have detri- The generic model we have proposed for guiding develop-
mental effects on patients. High-quality error correcting capa- ment of future Internet of Things healthcare systems has
bilities and significant robustness against interference are a number of use cases. To provide context, this subsection
26524 VOLUME 5, 2017

discusses several of these use cases, which include aiding emergency services. Readings at the time of the emergency
rehabilitation, assisting management of chronic conditions, can be recorded in the patient’s health record in the cloud, and
monitoring changes in people with degenerative conditions, the doctor can append information regarding their diagnosis.
and monitoring critical health for the provision of emergency As more and more people suffer from emergency health
healthcare. conditions and have diagnoses added to their files, machine
Following our proposed model, a rehabilitation system learning could begin to be used to make connections between
for knee injuries could be developed by using wearable symptoms and possible diagnoses. This information could
accelerometer sensors on either side of the knee, to allow for then be provided to responding paramedics, ensuring that
the position and angle of the knee to be calculated. These patients receive the most appropriate care for their condition,
measurements could be recorded during several activities, and rapidly. The authors intend to work on this system in their
such as normal walking and rehabilitation exercises. They future works.
could be communicated via short-range communications to a These are only a few of the possible use cases for sys-
comfortable, wrist-wearable central node, which could then tems that could be developed based on the proposed model.
forward information to the cloud via long-range communi- Nonetheless, these use cases highlight the versatility of the
cations. In the cloud, a record of the patient’s progress will model, and the number of different situations it could be used
continue to expand with each received message. Machine to managed.
learning algorithms could be implemented to identify the
patient’s progress, predict when they will be fully rehabili- III. WEARABLE HEALTHCARE SYSTEMS
tated, and determine whether any exercises are working better WBANs have been identified as a key component of a
than others. This system could easily be adapted for other or healthcare system founded on Internet of Things technology,
additional injuries by modifying which wearable sensors are and as such the development of accurate sensors with low
used. form factor are essential for the successful development of
Our model could also be used to develop a system capable such a system. In this article, we focus on sensors that are
of assisting with the management of chronic conditions such non-obtrusive and non-invasive; we exclude sensors such as
as hypertension. Blood pressure could be monitored at several implantables. Considered are five fundamental sensors - three
locations on the body at set intervals throughout the day and for monitoring the vital signs of pulse, respiratory rate, and
communicated to the cloud via a wrist-worn central node. body temperature, and a further two for monitoring blood
Again, a comprehensive record of the patient’s blood pressure pressure and blood oxygen, both commonly recorded in a
could be built and machine learning could be used to identify hospital environment.
trends such as when the patient’s blood pressure is highest.
This information could also be used to determine optimal A. PULSE SENSORS
times for the patient to take any medication that they may Perhaps the most commonly read vital sign, pulse can be
require to manage their condition, and remind the patient of used to detect a wide range of emergency conditions, such as
that using a buzzer or alarm on the central node. cardiac arrest, pulmonary embolisms, and vasovagal syncope.
Changes in people with progressive conditions such as Pulse sensors have been widely researched, both for medical
Parkinson’s Disease could also be monitored using a sys- purposes and for fitness tracking.
tem designed in accordance with our model. Symptoms of Pulse can be read from the chest, wrist, earlobe, finger-
Parkinson’s Disease include slowed movement, tremors, gait tip, and more. Earlobe and fingertip readings provide high
problems, and balance problems [27]. Using a series of wear- accuracy, but are not highly wearable. A chest-worn system
able accelerometers, sensors could be developed to measure is wearable, but wrist sensors are generally considered most
each of these parameters. Readings could be taken at set comfortable for a long-term wearable system [28].
intervals every day and forwarded to the wrist-worn central Commercially, several fitness tracking chest straps and
node, which in turn forwards the data onto the cloud. As the wrist watches are available with pulse measurement func-
data from the patient begins to grow, machine learning can tionality. These include HRM-Tri by Garmin [29], H7 by
be used to identify the rate at which symptoms are worsening Polar [30], FitBit PurePulse [31], and TomTom Spark
for the patient. A doctor could also add records of which Cardio [32]. However, these companies all disclose that their
treatments are being used, and machine learning could be devices are not for medical use and should not be relied upon
used to identify which treatments the patient’s condition has for detecting health conditions. As such, the sensing systems
responded the best to. employed by these devices cannot be directly implemented
Finally, critical health could be monitored using a system into a critical health monitoring system.
comprised of wearable sensors that monitor vital and other Much research has been conducted into suitable methods
important signs, including pulse, respiratory rate, body tem- for sensing pulse. Sensor types developed, used, and analyzed
perature, and blood pressure. Measurements can be taken in recent works include pressure, photoplethysmographic
regularly, and if any of these parameters fall below the (PPG), ultrasonic, and radio frequency (RF) sensors.
known healthy thresholds then the central node can forward PPG sensors operate by an LED transmitting light into the
the information to the cloud, which can be used to notify artery, with a photodiode receiving the amount not absorbed
VOLUME 5, 2017 26525

Pressure sensors and PPG sensors are combined

in [35] and [36], where pulse sensor modules are developed
with arrays of nine PPG sensors and one pressure sensor.
Pulse is taken from multiple points on the wrist, providing
clear pulse readings and the potential to use these readings
for diagnostics of certain diseases such as diabetes.
Diagnostics through pulse sensing is also investigated
in [37], where pressure, PPG, and ultrasonic sensors are
compared. Reasonable accuracy was achieved with all three,
but the authors concluded that specific diseases required
FIGURE 2. Photoplethysmographic pulse sensor. diagnosis using different sensor types; pressure was found to
be best for arteriosclerosis, while ultrasonic was superior for
diabetes.
by the blood, as shown in Figure 2. Changes in the amount of An et al., designed a non-conventional pulse sensor using
light can be recorded and a pulse rate can thus be determined. an RF array module in [38], with the aim of measuring several
In [28], PPG sensors are used to measure pulse, pulse locations on the wrist in case the received pulse signal at
rate variability, and blood oxygen in one small wrist-wearable one point becomes noisy due to movement. Reasonable pulse
sensor. As motion affects the accuracy of pulse readings from readings were achieved when compared to a reference signal,
PPG sensors, an accelerometer is used to check for move- but still do not appear as clear as those obtained with the tradi-
ment. When motion is high, the device goes into a low power tional sensor types. This type of pulse sensor shows promise,
state and does not record pulse. This is not entirely suitable but further work is clearly required to make it reliable in a
as pulse may be relevant when motion is high, such as when a critical healthcare scenario.
person is seizing or suffering cardiac issues during exercise. Based on these works, it is strongly recommended that PPG
Improving the accuracy of pulse sensors during motion would sensors are used for pulse sensing. These have repeatedly
be preferred to disregarding readings when movement levels been proven to be effective for measuring pulse rate, and
are high. techniques have already been developed to algorithmically
In [33], the effects of motion on PPG sensors are reduced reduce the impacts of noise on the signal quality.
by using two different LED light intensities and comparing
the amount of light received at the photodiode. Significant B. RESPIRATORY RATE SENSORS
improvement in signal quality is seen as motion artefacts are Another of the vital signs is respiratory rate, or the number
greatly reduced through this technique. of breaths a patient takes per minute. Monitoring respiration
Pressure sensors aim to mimic a healthcare professional could aid in the identification of conditions such as asthma
manually reading the radial pulse by pressing down with their attacks, hyperventilation due to panic attacks, apnea episodes,
fingers. As shown in Figure 3, the sensor is placed firmly lung cancer, obstructions of the airway, tuberculosis, and
against the wrist, and pressure is continuously measured to more.
acquire a pulse waveform. Due to the importance of respiration, many previous
works have developed sensors for measuring respiratory rate.
In inspecting the previous works, several types of respiratory
rate sensor emerge. The first is a nasal sensor based on a
thermistor, as is used in [39]. The principle that these sensors
are based on is that air exhaled is warmer than the ambient
temperature. As such, the sensor uses the rise and fall of tem-
perature to count the number of breaths taken. This is shown
to work reasonably well, but accuracy may be compromised
by other sources of temperature fluctuations - for example
if worn by a chef working in a kitchen. It is also not highly
wearable, as it is obstructive and easily noticeable.
FIGURE 3. Pressure-based pulse sensor. Echocardiogram (ECG) signals can also be used to
obtain respiration rate. This is called ECG Derived Respi-
In [34], a flexible and highly-sensitive pressure sensor for ration (EDR), and is used in [40] to determine respiration
pulse detection is developed and tested, showing promising patterns and detect apnea events. This method reads respira-
results. However, increasing the sensitivity to better detect tory rate reasonably well, but is again limited by the wear-
pulse also increases the amount of noise that is detected ability. ECG contacts are uncomfortable and would likely
due to movement of the wearer. This sensor was tested in cause irritation to the skin if used continuously. Additionally,
at-rest conditions, and further research would be required to ECG contacts are not reusable and would need to be regularly
determine that it performed well during motion. replaced.
26526 VOLUME 5, 2017

Respiratory rate can also be calculated using a microphone using these sensors, rather than on developing new sensors
to detect respiration, as was done in [41]. In this study, entirely.
focus was placed on detecting wheezing - a symptom com-
mon in asthmatics. The limitation of using a microphone C. BODY TEMPERATURE SENSORS
is that it would be extremely susceptible to any external The third vital sign is body temperature, which can be used
noise, and would therefore not be suitable as a long-term to detect hypothermia, heat stroke, fevers, and more. As such,
wearable. body temperature is a useful diagnostics tool that should be
One study [42] developed a fiber optic sensor in an elastic included in a wearable healthcare system.
substrate, that was sensitive enough to measure vibrations Recent works surrounding the measurement of body tem-
caused by respiration. This was shown to work in a single perature all use thermistor-type sensors. In [46] and [47],
test, but it is not known whether it would work well under the common negative-temperature-coefficient (NTC) type
all conditions. It is likely that this sensitive material would be temperature sensors were used, while positive-temperature-
susceptible to noise from other sources of vibration, including coefficient (PTC) sensors were considered in [48] and [49].
walking. Further testing should be conducted. In all studies, the thermistors were shown to measure a
A pressure-type sensor was developed in [43]. Two capac- suitable range of temperatures for monitoring the human
itive plates are placed in parallel, with one resting on the body, with acceptable levels of error. Therefore, it is strongly
abdomen. During breathing, the plates move further apart recommended that these sensor types continue to be used by
and then closer together during inhalation and exhalation future system designers.
respectively, allowing for calculation of respiratory rate. This The accuracy of temperature sensing is limited by how
study showed a 95% confidence in respiratory rate calcula- closely the sensor can be placed to the human body. As such,
tions when compared to a nasal sensor. This is fairly accurate, several works [48], [49] focused on developing sensors
and far more wearable than the nasal sensor it was compared printed onto thin, flexible polymers with adhesive backing
to. However, the nature of a pressure sensor may mean it is that could be attached directly to human skin. Whilst this
susceptible to noise if it is affected by external pressures, such is an interesting advancement, the work in [46] shows that
as while walking into wind. temperature can also be measured with relative accuracy
A common method of measuring respiratory rate is to use using a temperature sensor embedded in textiles. Thus, it is
a stretch sensor, as was done in [44]–[46]. Stretch sensors are recommended that system designers should use textiles to
those where properties change in response to the application hold temperature sensors until electronics printed on flexible
of tensile force, such as being stretched during inhalation. polymer can be more easily manufactured.
In [44], the designed sensor was made from a ferroelectric
polymer transducer, which generated a charge when a tensile D. BLOOD PRESSURE
force was applied. Measuring the changes in this charge allow Whilst not a vital sign itself, blood pressure (BP) is fre-
for calculation of respiratory rate. This sensor appeared to quently measured alongside the three vital signs. Hyperten-
obtain a clear signal, but accuracy was not verified through sion (high BP) is a known risk factor for cardiovascular
comparison with respiratory rate calculated by other means. disease, including heart attack. It is also one of the most com-
In [45] and [46], the respiratory rate sensors were based on mon chronic illnesses, affecting 32% of adult Australians.
changes in resistance. When a tensile force is applied to the Of those affected, 68% had uncontrolled or unmanaged
sensor, resistance increases. The changes in voltage caused hypertension [50]. As such, incorporating BP into a WBAN
by varying resistances can be used to calculate the breathing for healthcare would provide vital information for many
rate. patients.
Each of the stretch sensor types was shown to be effective Nonetheless, designing a wearable sensor for continuously
in calculating respiratory rate, but Atalay et al. [45] admit and non-invasively monitoring blood pressure remains a chal-
that motion artifacts were present during walking and other lenge in the field of healthcare IoT. A significant number of
movements. Additionally, in [46] it was found that breathing works [51]–[54] have attempted to obtain an accurate esti-
was accurate within 3.3 breaths per minute when sitting at mate of BP through calculation of pulse transit time (PTT) -
a desk; the margin of error increased when movement was the time taken between pulse at the heart and pulse at another
introduced. Therefore, a limitation of these sensors is that location, such as the earlobe or radial artery. Another work
other movements can cause tensile force to be applied to the endeavored to measure this property between the ear and
sensor in such a way that the sensor mistakes the movement wrist [55], while another looked to calculate it between the
for breathing. palm and the fingertip of a hand [56]. PTT is known to be
Evidently, many different sensor types exist for measuring inversely proportional to systolic blood pressure (SBP), and is
respiratory rate. The main factor in choosing a sensor type typically determined using an electrocardiogram on the chest
for a WBAN thus becomes the wearability. Therefore, stretch and a PPG sensor on the ear, wrist, or alternate location.
sensors are strongly recommended for implementation into The outcomes of each of these works indicate that the use
future systems. Future work should focus on developing algo- of PTT to calculate BP is not yet suitable. PTT is depen-
rithms and techniques to improve robustness against motion dent on several other factors, including arterial stiffness and
VOLUME 5, 2017 26527

blood density [54]. In ideal conditions, where devices had

been calibrated to the individual and the individual remained
relatively still during testing, reasonable results were acquired
by the aforementioned studies that utilized one measurement
at the chest and another at the wrist. Measurements taken
between the ear and wrist were shown to be inaccurate in [55].
FIGURE 4. Absorbance-mode vs. reflective-mode PPG sensors for pulse
Additionally, PTT was measured with reasonable accuracy oximetry.
between the palm and fingertip in [56], but the study did
not manage to convert this to blood pressure. This should
be investigated further, given some promise was shown in angle so that some light reflects to a photodiode on the same
measuring PTT and given that the design is the most wearable side of the appendage. These are called absorbance-mode and
option for monitoring blood pressure presented in this survey. reflectance-mode PPG sensors respectively.
For systems measuring between the heart and wrist, one study Classically, pulse oximeters are worn as a finger clip wired
identified that regular recalibration of devices would likely be to a medical monitor. Several recent works have attempted
required as the human body changes over time [51]. to make more portable devices. In [57], a low-power pulse
Another issue with these types of systems is that, while oximeter is designed with the aim of improving wearabil-
non-invasive, they are still obstructive. Usually a chest- ity. Two techniques are used to reduce power consumption.
wearable ECG is required in addition to some other device, The first - named ‘‘minimum SNR tracking’’ - continuously
and the connection between them may be wired. One calculates the current signal-to-noise-ratio (SNR) and adjusts
study [55] identified this issue and opted to use two eas- the length of time that the LED is in the ‘‘on’’ state for
ily wearable PPG sensors - one on the earlobe, and one accordingly - the higher the SNR is, the longer that the LED
on the wrist - to estimate pulse arrival time (or time taken needs to be on to gain accurate readings. The second, named
to travel) between these locations and thus estimate blood ‘‘PLL tracking’’, estimates when the peaks and troughs of
pressure. The results were promising, showing reasonable the PPG signal are likely to occur, and samples only at these
measurements for healthy subjects in different positions (such times to acquire this important information. Up to 6x less
as sitting and standing). However, the measurements taken power was consumed through implementing both techniques,
were not compared to measurements from a traditional cuff- and the worst error recorded was a 2% difference between
based sphygmomanometer. Such comparison would aid in actual and measured blood oxygen levels. This is a significant
analyzing the accuracy of the PPG-based system. contribution towards making pulse oximeters more wearable,
While no system has yet been developed for accurately but reductions in the level of error are desirable.
measuring blood pressure continuously using a comfortably An in-ear reflective pulse oximeter was designed in [58].
wearable device, this is a field worthy of further research. It is This was designed to detect blood oxygen levels even when
suggested that this could be achieved by developing a device the patient is suffering from shock, hypothermia, or other con-
that utilizes two or more PPG sensors placed along the arm to ditions that may cause blood centralization and lead to pulse
calculate PTT. Blood pressure is certainly a valuable param- being undetectable at the fingertips. The oximeter sits inside
eter in healthcare, and the ability to monitor it continuously the ear canal without sealing it, ensuring there is no disruption
would greatly improve the quality of healthcare that could be to hearing. Reasonable accuracy in measuring blood oxygen
provided through a WBAN-based system. levels was achieved in clinical testing on surgical patients,
but the authors concluded that their sensor should be used
E. PULSE OXIMETRY SENSORS in addition to finger pulse oximeters, not as an alternative.
Pulse oximetry measures the level of oxygen in the blood. This is a sensible idea for wearable healthcare systems that
Like blood pressure, blood oxygen level is not a vital sign, are providing remote care, as it would be preferable to detect
but does serve as an indicator of respiratory function and when centralization is occurring.
can aid in diagnostics of conditions such as hypoxia (low The most wearable option would be a wrist-worn sensor, as
oxygen reaching the body’s tissues). As such, pulse oxime- many people are accustomed to wearing bracelets or watches
try is a valuable addition to a general health monitoring and would not find this uncomfortable. In [59], a reflective
system. pulse oximeter was designed to be worn on the wrist. The
Pulse oximeters measure blood oxygen by obtaining PPG design is concave in shape, blocking out much external light
signals. Usually, two LEDs - one red, one infrared - are and improving robustness against noise. The trade-off is that
directed through the skin. Much of this light is absorbed by it makes the device larger, but size reductions are then made
the hemoglobin in the blood, but not all. The amount of light by performing data processing off-node. Overall, it is more
not absorbed is measured by receiving photodiodes, and the wearable than other designs, but would still benefit from
difference between the received lights is used to calculate being miniaturized and made more wearable. Additionally,
blood oxygen. As highlighted in Figure 4, LED lights can it could be used to detect pulse and skin temperature as well -
either be passed through an appendage (normally a finger) combining three essential sensors into a single, comfortably
to a photodiode on the opposite side, or can be directed at an wearable node.
26528 VOLUME 5, 2017

Overall, the works in improving pulse oximetry do not future use in controlling lower limb prosthetics. Detection of
focus on finding new means to measure blood oxygen sat- gait events, rather than general gait patterns, has also been
uration, but instead focus on making wearable devices that considered in several works. In [73], three accelerometers are
utilize the well-known existing techniques. Research should placed on the hip, knee, and ankle of advanced Parkinson’s
continue in this direction, focusing on wrist-wearable pulse Disease sufferers. Features are extracted from the data, and
oximetry. an anomaly detection scheme is used successfully detect
freezing of gait, a common Parkinson’s symptom that causes
F. OTHER WEARABLE SENSORS FOR HEALTHCARE a temporary loss of motor function and regularly leads to
Aside from the sensors that measure critical health parame- falls. Detection of any gait anomaly is investigated in [74],
ters, there are several special-purpose wearable sensors that where a waist-worn device comprised of a microcontroller
may be useful in systems focused on monitoring a specific and tri-axial accelerometer is used to monitor gait. Through
condition. Echocardiograms (ECGs) can be used to evaluate the implementation of feature extraction and an advanced
heart health, and several wearable sensors have been devel- anomaly detection algorithm, a system is created that can
oped to acquire these signals. In [60], an armband-based detect approximately 84% of gait anomaly periods that last
ECG sensor was developed and measures with reasonable for 5 seconds. Accuracy was higher when detecting severe
accuracy. ECG sensors have also successfully been devel- anomalies.
oped for integration in helmets [61] and more traditional People living with diabetes are considered in several works
chest-straps [62]. aiming to develop a non-invasive blood glucose monitor, none
The helmet in [61] also features an electroencephalogram of which currently exist on the market. In [75], an optical
(EEG) sensor. EEGs measure brain activity, and could gener- sensor measures blood glucose levels through the fingertip,
ally be used to monitor seizures, sleep disorders, and progress while in [76] an ultrawide band microwave technique is
after a head injury. Other EEG systems have been developed used for blood glucose level detection in an earlobe-attached
for specific purposes, such as for detecting driver drowsi- sensor.
ness [63] or stress management [64]. Both systems measure These sensors are not the only special-purpose sensors that
EEG through a relatively wearable headband. have been researched or could be researched, but they are
Fall detection can be useful for monitoring elderly people, some of the key types. They are applicable to many common
as they are particularly prone to falls and resulting injuries. diseases or conditions, and should be considered for inclusion
In [65], a tri-axial accelerometer inside a smartphone is used by designers of systems that will focus on monitoring of
by machine learning algorithms to classify the user’s posture, specific ailments.
which the best algorithms showing classification accuracy
of 99.01%. A related study found that the classification algo-
IV. COMMUNICATIONS STANDARDS
rithms used for posture detection were much less accurate
Communications related to Internet of Things for healthcare
when performing fall detection [66], suggesting that further
can be classified into two main categories: short-range com-
training or alternate algorithms may be required for this
munications, and long-range communications. The former is
purpose. In a more recent work on fall detection, a wearable
used to communicate between devices within the WBAN,
camera was used in [67], with rapid changes in scenery used
whilst the latter provides connection between the central
to detect falls. This showed an accuracy of 93.78% and 89.8%
node of the WBAN and a base station (such as a healthcare
in indoor and outdoor environments respectively. In their
provider). In this paper, both types of communications are
earlier work [68], accelerometer data was combined with an
considered with equal importance.
earlier version of their wearable camera system, showing 91%
accuracy in detecting falls. An accelerometer, a gyroscope,
and a magnetometer were used to accurately detect falls A. SHORT-RANGE COMMUNICATIONS
in [69], with the authors then adding a barometer to even more In the context of wearable healthcare systems, short-range
accurately detect changes in height in [70]. The latter work communications are often used between nodes, particularly
showed that fall detection was performed with no lower than between sensor nodes and the central node where data pro-
99.38% accuracy and up to 100% accuracy across a series cessing occurs. Although short-range communications stan-
of tests. This is an exceptional result, and suggests that this dards can be used for other purposes (i.e. developing mesh
fall detection system could be implemented into healthcare networks for smart lighting), this survey focuses on the pur-
applications immediately. pose of developing a small WBAN that is comprised of only
Gait detection can also be useful in monitoring the elderly, a few sensors and a single central node.
as well as those with specific conditions such as Parkinson’s Many short-range communications standards exist, but
Disease (PD). Gait detection for those recovering from stroke perhaps the most commonly used ones in IoT are Bluetooth
or suffering from PD was considered in [71], with foot- Low Energy (BLE) and ZigBee. The key features of these two
worn sensors designed to measure many parameters including standards are highlighted in Table 1, and this section further
step size and walking speed. A sensor for gait detection in analyzes these standards and considers their suitability for
lower limb amputees was developed in [72], with the aim of implementation into an IoT healthcare system.
VOLUME 5, 2017 26529

TABLE 1. Comparison of short-range communication standards.
1) BLUETOOTH LOW ENERGY Curve Hellman-Diffie (ECHD) algorithm, which uses a pub-
BLE was developed by the Bluetooth Special Interest lic key and private keys unique to each device, to secure
Group (SIG) to provide an energy-efficient standard that key exchange. Two keys are exchanged between master and
could be used by coin-cell battery operated devices, including slave - a Connection Signature Resolving Key (CSRK) and
wearables. It also aimed to enable IoT, connecting small an Identifying Resolving Key (IRK). The former is used to
peripheral devices to processing devices such as smart- provide authentication for unencrypted data, whilst the latter
phones [77]. provides privacy and the device’s identity [80]
BLE is used in a star topology, which is suitable for health- Encryption is also available in BLE, using Advanced
care applications. The central node would act as the center of Encryption Standard (AES). Specifically, a 128-bit AES
the star topology, with sensors linked to it. The sensors will cypher is used to protect the data from potential
have no need to communicate with each other directly. attackers [80].
The range for BLE is 150m in an open field [77]; it would Man-in-the-middle, eavesdropping, and identity attacks
be much less in non-ideal conditions. It also has a low latency are effectively protected against by the security features of
of 3ms, and a high data rate of 1Mbps [78]. The range is BLE. This is crucial in a healthcare environment where sen-
clearly sufficient for use in a healthcare WBAN where nodes sitive patient data is being exchanged.
are physically proximal, and the extremely low latency is Classic Bluetooth has previously been used in IoT for
ideal for applications such as emergency health. healthcare works including a blood pressure monitoring sys-
BLE operates in the 2.4GHz band, a band also used by tem [81] and a system for early detection of Alzheimer’s
classic WiFi and ZigBee. This may subject it to some noise, disease [82], as it has been optimized for IoT; unlike Classic
but robustness to interference is implemented through fre- Bluetooth.
quency hopping across carefully selected channels and a Overall, BLE is extremely well suited to healthcare appli-
24-bit cyclic redundancy check (CRC) [79]. This method cations. It is secure and features good range, low latency,
makes BLE robust enough to noise for use in a healthcare low power consumption, and robustness to interference. This
system. standard is highly recommended to designers, as it is cur-
Power consumption in BLE is extremely low. In [77], it is rently the most suitable standard for implementation into
shown that a 180 mAH coin cell battery could run a BLE chip wearable healthcare systems.
for 18 continuous hours, making 21.6 million transactions.
However, if the chip was powered off when not needed, 2) ZIGBEE
battery would last much longer. If a health sensor transmitted The ZigBee standard was designed by the ZigBee Alliance,
its data every 30 seconds (or 2,880 times per day), then specifically for providing low-cost, low-power networks for
the battery could theoretically run the BLE chip for around M2M communications. It builds on the IEEE 802.15.4 phys-
20.5 years if not for the fact that it would die from self- ical standard [83]. It is commonly known as the standard for
depletion well before then. With careful hardware design and mesh networks, but it can also be used in the star topology
low-energy programming, BLE would clearly be suitable for required of a WBAN with one central node and many sensing
healthcare applications. nodes.
Security has been implemented in a variety of ways for Different ZigBee modules provide different characteristics
BLE. Firstly, there are four possible pairing models. The in terms of range, data rate, and power consumption. The
newest and most secure of these, LE Secure Connections, simplest XBee has a range of up to 30m in an urban environ-
implements a numeric comparison method and an Elliptical ment, and outputs only 1mW of power for transmission [84].
26530 VOLUME 5, 2017

The XBee Pro has a higher range of 90m in the same condi- Overall, ZigBee is reasonably well-suited to healthcare
tions, but the transmit power output is significantly higher at applications. It provides robustness to interference and sev-
63mW [84], the XBee Pro 900 XSC can reach up to 610m in eral security mechanisms. Several implementations are pos-
an urban environment, but with 250mW of power being used sible to optimize range, data rate and power consumption
to transmit [85]. There are ZigBee-based solutions for a wide for specific applications. XBee modules were examined
variety of applications, but for the use case of a healthcare as a case study, and it was found that low-power, high
WBAN the XBee 1mW would be suitable. Only a small data rate modules existed with suitable range for healthcare
range is needed for on-body communications, so choosing the applications.
lowest-power solution is preferable. The main drawback of using ZigBee is that key exchange
Data rates are also variable. XBee and XBee Pro have a can be compromised unless implemented extremely well by
data rate of up to 250kbps [84], while the XBee Pro 900 XSC the manufacturer. This could pose a risk to healthcare sys-
has a maximum data rate of only 10kbps [85]. Clearly, the tems where sensitive patient data is being exchanged regu-
latter module has sacrificed data rate in addition to power effi- larly. Additionally, ZigBee is not commonly implemented in
ciency to achieve its long range. In a healthcare environment, devices such as smartphones, while BLE typically is. This
it would be preferable to opt for a higher data rate, as this will makes it less compatible with existing devices, and there-
reduce the latency in the system and ensure critical health data fore it is suggested that it would be better suited to fixed-
is delivered timely. location, standalone purposes such as home automation than
ZigBee can operate at a range of frequencies, includ- it is to wearable healthcare systems. It is therefore recom-
ing 868MHz, 900MHz, and 2.4GHz bands, depending on mended that system designers prefer BLE for wearable sen-
the module chosen. Each of these bands faces interference. sors over ZigBee, particularly in applications where privacy is
The 2.4GHz band is shared by Bluetooth and WiFi, while critical.
many long-range communications systems utilize the unli-
censed 868MHz and 900MHz bands in Europe and America B. LONG-RANGE COMMUNICATIONS
respectively. ZigBee uses CSMA-CA to reduce collisions, Low-Power Wide-Area Networks (LPWANs) are a subset of
and implements re-transmission if messages sent are not long-range communications standards with high suitability
acknowledged [86]. for IoT applications. The range of a LPWAN is generally
Several security features are provided by ZigBee, though several kilometers, even in an urban environment. This is
most are optional and must be enabled by the network devel- significantly longer than the range of traditional IoT com-
oper. ZigBee’s security model is largely based on 128-AES munications types such as WiFi or Bluetooth, whose ranges
encryption, and offers types of security keys - a link key, are in the order of meters and thus would require extensive
a network key, and a master key. The network key is manda- and costly mesh networking or similar to be plausible for
tory. It is shared by all devices on the network, and is a healthcare.
network-layer security mechanism that cyphers all transmis- LPWANs also have significant advantage over cellular
sions within the network. The link key is optional, and is used networks such as 3G in that they are designed to support
to secure communications at the application layer. Master short bursts of data infrequently. This is suitable for a large
keys are also optional, and are used to secure the creation and number of healthcare applications, including monitoring gen-
sharing of link keys [87]. eral health and receiving hourly updates, monitoring criti-
Despite these security measures, a recent study [88] found cal health and receiving emergency calls, and rehabilitation
that it was relatively easy to exploit a ZigBee network during where updates may only be necessary once daily. This design
a device join, by sniffing the link key being exchanged. This principle also allows for low-power device design, which in
compromises the network key and thus the entire network. turn ensures that the designed healthcare devices will operate
The researchers in this study did identify that the security for longer before human interaction is required to recharge
flaws were not to do with ZigBee itself, but rather with or change batteries. This reduces the risk of patients being
the way that manufacturers implemented key exchange and offline, and provides more convenience to the wearer. Based
initialization. If ZigBee is to be implemented in healthcare on these advantages, it is suggested that LPWANs are the
systems, only ZigBee modules with all optional security best solution for transmitting data from the central node to
mechanisms and robust key management should be used. the cloud for storage or further processing.
ZigBee has already been used in several works relat- The most prominent standards for LPWANs are Sigfox and
ing to healthcare. In [89], a system that detects wandering LoRaWAN. While these standards are well-established, they
Alzheimer’s patients and alerts their caregivers was devel- face competition from emerging standards including NB-IoT.
oped using a ZigBee mesh network. In [62], a wearable In this section, existing LPWAN standards are considered in
ECG sensor was developed using ZigBee to communicate terms of suitability for an IoT healthcare system, and recom-
with a central monitoring device. Improving ZigBee for use mendations are made. A table summarizing the three main
in biomedical application was considered in [90], where standards discussed is also included in Table 2 to provide a
a low-power transceiver was developed with robustness to snapshot of their features, enabling easy comparison between
interference. these standards.
VOLUME 5, 2017 26531

TABLE 2. Comparison of long-range communication standards.
1) SIGFOX technologies operating in the sub-GHz spectrum, no globally

Perhaps the simplest of the LPWAN standards, Sigfox pro- available band exists for Sigfox’s use. Operating in an unli-
vides limited functionality but is widely deployed compared censed band allows Sigfox to occupy a wider bandwidth, but
to other standards listed. It is a protocol developed in the comes with the disadvantage of increased interference, which
first four layers of the OSI model, and its base stations bear may be an issue in healthcare.
similarity to those in cellular - antennas mounted on towers. To increase resistance to interference, Sigfox sends pay-
Sigfox uses a star topology, and nodes are designed to be loads in three consecutive frames, each in different pseudo-
uplink only to improve battery efficiency. It is possible for random sub-carriers and over different propagation paths.
a node to receive downlink, but it must explicitly request it. This improves the likelihood that the message will be received
As acknowledgement of receipt is important for health data, intact, and thus reduces the disadvantages of increased inter-
downlink would have to be requested. Unfortunately, a limi- ference in the unlicensed bands.
tation of Sigfox is that downlink can only be requested 4 times Sigfox has a high network capacity and can support
per day [91]. approximately 50,000 nodes with a single gateway [92]. This
There is a trade-off in design between latency and range. is comparable to NB-IoT’s 52,547 nodes, which has been
If a receiver has higher sensitivity, it can detect weaker sig- shown to be able to support 40 devices per household assum-
nals; thus, the distance a signal can travel is increased. Sigfox ing a household density equivalent to London’s and an inter
opted to maximize range to around 9.5km in urban areas [92] site distance of 1732m [95]. In rural areas such as Australia’s
by using slow D-BPSK modulation and a low bit rate of Burdekin Shire, which had a population of 17,784 people in
100 bits per second (bps) [93]. In rural areas, Sigfox can 2011 [96] and is 4880km2 in area [97], one well-positioned
reach a range of up to 50km [94]. The high latency of Sigfox base station would enable every resident of the region to be
is a drawback for its use in healthcare applications, as it is connected to healthcare providers via Sigfox. This is signifi-
important for messages to be delivered quickly in this critical cant for healthcare applications, especially in regional areas.
context. Security is implemented in Sigfox by signing each message
Sigfox operates in the unlicensed bands of 868MHz in with the device’s private key [91]. This reduces the risk of
Europe and 915MHz in the US. As with other LPWAN spoofing attacks or interception, but does not eliminate it.
26532 VOLUME 5, 2017

A sophisticated attack targeting the node hardware or service 7.2km in an urban area at a rate of 0.25-5.5 kbps. This is a
provider could still reveal the unique keys, compromising a significantly faster data rate than Sigfox, with about 2.3km
patient’s healthcare system. less range. It is suggested that for healthcare, this trade-off
The limit of 140 messages per day also minimizes the is worthwhile as low latency is essential. More range can be
impact that a spoofing attack could have. However, in a obtained by simply installing more base stations.
healthcare environment, even a small number of fraudulent To allow consumers to choose a solution that suits them
messages could have significant impacts on the wellbeing of best, LoRaWAN specifies three classes of device that can
patients. be used. Class A is the most battery efficient, providing
Encryption and scrambling methods are supported by downlink for only a small window after uplink. This would be
Sigfox [91], but developers must implement these themselves suitable for healthcare, as downlink would only be required to
within the 12-byte payload. If implemented well, these meth- receive acknowledgement that the health data was delivered.
ods could reduce the risk of sensitive patient data being Like Sigfox, LoRaWAN operates in the unlicensed bands
intercepted. of 868MHz in Europe and 915MHz in the US. This carries
Overall, Sigfox is suitable for non-critical applications the advantage of wider spectrum availability, but also the
where speed of message delivery and acknowledgements disadvantage of increased exposure to potential interference.
of receipt are non-essential, such as in smart parking and Security is provided by the LoRaWAN standard. A unique
automated street lighting. In such an application, a breach key is assigned to each node on the network; this key is known
of security would cause annoyance rather than danger. only to the node and to the network provider. Theoretically,
However, in healthcare, successfully transmitting messages this would eliminate man-in-the-middle attacks as intercepted
with relative speed is essential. Any compromise of security data would be encoded and not decipherable [99].
could be detrimental for an individual’s health, or could affect Unfortunately, a node’s unique key could become a single
the integrity of medical databases. For this reason, we recom- point of failure for the whole system should the key be discov-
mend that Sigfox not be used for critical healthcare appli- ered through sophisticated hardware hacking of the node, or
cations. It is therefore strongly recommended that system through an attack on the network server. If a key was illicitly
designers instead consider alternative LPWANs for critical obtained, then the attacker could use it to decipher all future
healthcare applications. messages from the node, or could send false messages to the
base station whilst pretending to be the node.
2) LoRa & LoRaWAN For the most part, LoRaWAN is reasonably well-suited to
Technical information about the LoRa & LoRaWAN stan- healthcare applications due to its range, latency, and network
dards is presented in [98], written by the LoRa Alliance. This capacity. Interference could potentially cause issues while
subsection overviews the key components of the standard operating in unlicensed bands, but a more significant issue
based on this source, and thus the interested reader is referred is that of key management. To be truly secure, proper key
to it for further information about LoRa & LoRaWAN. management must be implemented by the developers and ser-
LoRa is a physical layer protocol that utilizes chirp spread vice providers so that sensitive healthcare data and important
spectrum techniques over a wide bandwidth of at least medical databases are protected from malicious attacks, as
125kHz. This provides low-power, long-range communica- LoRaWAN does not provide a complete security solution.
tions with high resilience to intentional or environmental
interference. 3) NB-IoT
LoRaWAN is built on top of the LoRa standard, in the Standardized in the recent 3GPP Release 13, NB-IoT operates
network layer. It utilizes a star topology, and nodes are in the licensed bands of GSM or LTE and provides long-
asynchronous; they only communicate when they need to, range, low-power communications. As NB-IoT has been
such as after an event or scheduled measurement. Scheduled developed based on LTE, much of the existing LTE hardware
messages from nodes would suit long-term monitoring appli- can be used to deploy it rapidly and effectively [95], [100].
cations, while event-driven messages from nodes would suit There are three different ways in which NB-IoT can be
emergency monitoring. deployed, allowing easier coexistence with existing networks.
LoRaWAN also has a high network capacity, ensur- These deployment modes are in-band, guard-band, and stand-
ing many messages can be passed over the network at alone. Using in-band mode involves reserving LTE Physical
the same time. Each gateway can support approximately Resource Blocks (PRBs) from the existing LTE network, for
40,000 nodes [92]. While this is lower than the capacity of use by NB-IoT. In guard-band mode, NB-IoT utilizes the
Sigfox, it would still be suitable for use in urban and regional bandwidth of an existing LTE carrier’s guard-band. Finally,
areas if optimal base station positioning was considered in stand-alone mode, GSM carriers can be re-farmed and
thoroughly. used for NB-IoT, or NB-IoT can exist in entirely new band-
Already, LoRaWAN has been successfully deployed in width [101], [102].
several areas including parts of Europe, America, and Operating within licensed bands has the significant advan-
the Asia-Pacific region. With a maximum link budget tage of reduced risk of interference. One potential disadvan-
of 155 dBm, messages can travel over a range of around tage is that there will likely be a higher cost to use NB-IoT
VOLUME 5, 2017 26533

than there is for unlicensed standards. Just as mobile phone of existing LTE infrastructure. This lack of deployment limits
users pay a fee to be able to use LTE, NB-IoT device users the immediate usefulness of NB-IoT, but the standard will
will likely have to pay a connection fee for the use of NB-IoT. likely be deployed rapidly on a large scale due to the ability
The exact magnitude of these costs is not yet known as to reuse existing 3G hardware. When this happens, NB-IoT
NB-IoT is yet to be widely deployed. Nonetheless, a decrease is highly recommended for use, as it offers many favourable
in interference is likely to be worth the additional cost when characteristics that make it the most suitable standard for use
considering healthcare systems, as QoS in these applications in healthcare systems.
is critical.
Due to a high receiver sensitivity of 164dB, NB-IoT 4) OTHER STANDARDS
achieves a range of up to 15km. Despite the long range, speed Several other LPWAN standards have been developed for
remains relatively high, with a maximum uplink data rate of operation within unlicensed, sub-GHz bands, but these
250kbps [103]. The significant data rate and wide range are have been minimally deployed compared to SigFox and
ideal for healthcare applications, as messages can travel a LoRaWAN. They also feature unique hardware, making
reasonable distance within an acceptable time frame for even them harder to deploy on the same wide-scale that NB-IoT
the most critical health events. could be deployed on. These standards include Symphony
Battery life was a fundamental consideration in the design Link, nWave, Weightless, and NB-Fi. Outside of the sub-
of NB-IoT. The power efficiency has some dependence on GHz bands but still within ISM bands lies Ingenu Ran-
which mode NB-IoT operates in. In [95] it was found that dom Phase Multiple Access (RMPA). Meanwhile, in the
the life of 5Wh batteries were 2.6 and 2.4 years in stand- licensed bands, two additional standards besides NB-IoT
alone and in-band modes respectively when a 50 byte mes- have been developed by 3GPP. These standards are Extended
sage was being sent every 2 hours. When a 50-byte message Coverage GSM for IoT (EC-GSM) and Long-Term Evolu-
was only sent once per day, the battery lifetime increased to tion Machine Type Communications Category M1 (LTE-M
18.0 and 16.8 years respectively. Applications such as long- or eMTC). Each of these standards is briefly discussed in this
term monitoring would likely require several transmissions section.
per day, while emergency health monitoring system would Symphony Link [106], [107] is a synchronous protocol that
transmit short ‘‘heartbeat’’ messages occasionally. Longer uses LoRa as its physical layer, and serves as an alternative to
messages would only be sent when an emergency condition is LoRaWAN. It is used in a star topology, but allows for the use
detected. For each of these applications, the energy consump- of repeaters to provide more hops. All messages sent using
tion of NB-IoT is sufficiently low, and allows for minimal Symphony Link are acknowledged, and message length can
interaction by the system wearer. be longer. It is robust and secure standard, but is less energy-
NB-IoT can also support a minimum of 52,547 nodes per efficient due to the need for nodes to frequently synchronize
base station. As mentioned in Section IV.B.1., this would with the network. The increased power consumption and
be sufficient to support 40 devices per home, assuming a lack of deployment are the main limiting factors in using
household density equivalent to that of London [95]. With an Symphony Link for healthcare purposes.
average household population of 2.47 people [104] and ideal nWave [108] is an ultra-narrow band technology using
positioning of cell sites, every resident of London could wear a star topology. It boasts high capacity and a range of
over 15 healthcare sensors, each successfully communicating up to 10km. nWave has previously been used for smart
directly with the closest base station. The capacity of NB-IoT farming and smart parking. Little further information is made
is clearly sufficient for providing wide-spread healthcare. available by this company with regards to their technology.
State-of-the-art 3GPP S3 security is used by NB-IoT, Weightless offers three standards to give end
with mechanisms on both the transport and application lay- users more choice - Weightless-N, Weightless-P, and
ers [105]. There are several mandatory mechanisms includ- Weightless-W [109]. Weightless-N was developed through
ing entity authentication, device identification, user identity a partnership with nWave [110], and offers uplink-only sim-
confidentiality, and data integrity. Optional mechanisms are plex communications with low battery consumption and a
also available for ensuring application authentication, data range around 4.1km. Weightless-P offers two-way commu-
confidentiality, and more. As eavesdropping is a real threat nications, but with lower range and shortened battery life.
to radio communications, it is recommended that the optional Weightless-W has the lowest battery life, but has similar
encryption mechanisms are also used to protect sensitive range to Weightless-N whilst enabling two-way communica-
health data. With all mandatory and optional mechanisms in tions. Of the Weightless standards, Weightless-P would likely
place, NB-IoT will likely be suitably secure for healthcare be best suited to most healthcare applications due to its two-
applications. way communications and longer battery life. However, the
Overall, NB-IoT is suitable for healthcare applications. short range of 2km is a limiting factor even in a dense urban
It is secure, supports communications over a long range, has environment.
high energy efficiency, and can support many devices. The NB-Fi by WAVIoT [111] is another unlicensed UNB tech-
most significant drawback is the current lack of deployment, nology that operates in a star topology. It is full duplex, with a
though this is expected to occur rapidly due to the reusability range of 16.6km in an urban environment. The trade-off made
26534 VOLUME 5, 2017

is a high latency of 30 seconds on uplink and 60 seconds V. CLOUD-BASED IoT HEALTHCARE SYSTEMS
on downlink [112]. This may be suitable for non-critical Cloud technologies have been widely researched due to
systems, but would not be suitable when an emergency health their usefulness in big data management, processing and
alarm needs to be transmitted. analytics. Several related works have surveyed the litera-
Ingenu RPMA [113] is the only well-known LPWAN stan- ture on using cloud technologies for IoT purposes such as
dard operating in the unlicensed 2.4GHz band. This design smart grid [115] and mobile cloud computing for smart-
choice allows for higher transmission power and antenna gain phones [116], [117], where complex computations are
to be used, but this uses extra power. Additionally, the pop- offloaded from low-resource mobile devices to the high-
ularity of the 2.4GHz band means that messages transmitted power environment of the cloud, before the result is returned
using Ingenu RPMA are likely to be subjected to interference. to the mobile device. These works consider data storage and
Ingenu’s range is short compared to its competitors, only data processing as key advantages of cloud technologies.
reaching 4.6km in urban areas [92]. High capacity and QoS Further related works have reviewed the state of cloud-
are provided by Ingenu, which would be advantageous in centric healthcare. The use of cloud technology for health
healthcare. However, the low range for high power consump- record storage is considered in [118], which also overviews
tion is not ideal for a battery-powered wearable system. This cloud technologies as a complete field. Storage is consid-
standard would likely be better suited to IoT applications ered further [11] and [119], with particular focus on how
where permanent power supplies are available for sensor a large database could be used for data analysis and trend
nodes. determination.
EC-GSM is a licensed band standard that was introduced to While each of these related works provides valuable insight
improve GSM for IoT usage, as GSM has already been widely into the field of cloud technologies, there is no known article
used for M2M communications due to its high deployment that considers all advantages, disadvantages, challenges, and
and low cost devices. EC-GSM can be enabled by updating opportunities that cloud presents to healthcare systems based
the software on existing GSM gateways. This allows for on WBANs and the IoT. In this section, we bridge the gap
extremely fast roll-out, and allows operators to extend the in the literature by presenting recent works regarding cloud-
useful lifetime of legacy 2G gateways. EC-GSM provides centric healthcare, analyzing key challenges, and providing
coverage improvements of up to 20dB, and each gateway recommendations for future research directions.
can support up to 50,000 devices [114]. The data rate offered
by EC-GSM is less than 140kbps for both uplink and down- A. CLOUD FOR HEALTHCARE
link [103], slower than NB-IoT. Currently, EC-GSM has sim- Much research has been conducted in recent years regarding
ilar capabilities to that of NB-IoT and thus would be suitable the benefits of cloud for healthcare applications. These bene-
for use in healthcare. However, EC-GSM operates on legacy fits stem from the three primary services that can be provided
gateways while NB-IoT operates on modern LTE gateways. by cloud technologies in healthcare environments:
It is therefore suggested that NB-IoT will outlast EC-GSM • Software as a Service (SaaS) - provides applications to
as a widely utilized standard, and thus it would be preferable healthcare providers that will enable them to work with
to develop healthcare systems based on NB-IoT rather than health data or perform other relevant tasks.
EC-GSM. • Platform as a Service (PaaS) - provides tools for virtual-
LTE-M is another licensed band standard that has been ization, networking, database management, and more.
developed to utilize the full capacity of an LTE carrier using • Infrastructure as a Service (IaaS) - provides the physical
multiplexing techniques, whilst improving battery life and infrastructure for storage, servers, and more.
coverage. The major difference between LTE-M and classic These services can be used to achieve a variety of tasks,
LTE is that the former introduces new power-saving meth- but two key uses are easily identified in the literature; big
ods not implemented in the latter [114]. With a data rate data management and data processing. These two different
of up to 1Mbps [103], LTE-M allows for more advanced concepts are presented separately in this section. However, it
IoT applications, but has limited range and can only support is also highlighted that both are essential for a state-of-the-art
around 20,000 nodes per gateway [92]. LTE-M is certainly an IoT healthcare system, and thus should be included together
ideal solution for a system where high speed, large amounts in future cloud system designs.
of data, and advanced features are required. However, this
is not the case for healthcare. A healthcare system that is 1) BIG DATA MANAGEMENT
intermittently transmitting small amounts of important data Big data is regularly characterized by the so-called 5 V’s -
would benefit from long-range, high-capacity gateways, like volume, velocity, variety, veracity and value. Volume refers
those seen in NB-IoT. In LTE-M, these features crucial to to the amount of data generated, while velocity refers to the
healthcare are traded off to provide the higher data rate and speed at which it is generated. Variety is the general variance
enhanced functionality. LTE-M is an excellent standard for in data types, while veracity is the uncertainty surrounding
some applications, but is not suitable for healthcare when what data types may later be added. Finally, value refers to
compared to other cellular IoT standards. what information can be gained from the big data set.
VOLUME 5, 2017 26535

The challenge in big data management lies in designing a health record, but rather to amass a big data set that machine
system that can handle the characteristics of a specific big learning could be applied to. The authors have placed signif-
data set. In healthcare, each of the characteristic 5 V’s are icant focus on managing all the characteristic Vs of big data,
important to consider, as a wide variety of data from patient but there are still some improvements that could be made.
name, age and gender to vital sign values as taken at regular While health data is stored, it appears that patients and doctors
intervals would need to be stored for many systems. Regularly cannot easily access a patient’s complete medical history. It is
measured data would create significant velocity, and lead to suggested that implementing techniques from the systems
an increased volume of total data rapidly. Additionally, new focusing on storage of health records could further its overall
kinds of data may be added regularly as new sensors are usefulness.
developed to measure previously unmonitored health signs. Enabling machine learning through appropriate big data
Finally, machine learning to perform diagnostics or provide management in the cloud is considered in [122]. The authors
treatment plans would be extremely valuable in a healthcare identify that cloud storage is useful for storing high volumes
context, so a cloud storage framework for healthcare would of data in such a way that value can be extracted from it. In this
need to enable value. As all of the characteristics of big system, physiological parameters and frequency of medical
data are important to healthcare applications, recent research visits are both stored in the patient’s health record. This
in this area has focused on storing a wide variety of data information can then be used by machine learning algorithms
generated by voluminous IoT systems in an organized manner to determine the patient’s condition. To prove this, machine
that may be useful for later data analysis. learning was successfully applied for the detection of flu.
In one recent work [24], volume and variety of healthcare Accuracy in classifying the flu steadily increased as more
data is considered, and a data storage model suitable for sensors were included in the WBAN, reaching 98% accuracy
emergency healthcare is developed. It aims to organize het- when 14 parameters were measured. This suggests that the
erogeneous physiological data and make it readily accessible authors have implemented data management to a reasonably
to relevant healthcare providers during an emergency. IaaS high standard. However, it is unclear whether the system is
and SaaS are used to reach these goals. One area of improve- capable of rapidly expanding as the velocity and veracity
ment in this work is that access control is only considered of data increases over time. Further testing would likely be
on an elementary level. Additionally, high levels of human required to ensure that the designed cloud architecture is still
interaction are required to enter and maintain patient health suitable when the database is non-static and continuously
records. For a WBAN-based system, automation techniques expanding.
would need to be implemented to store data from sensors in An extremely thorough work on cloud for healthcare is
the appropriate areas of the health record. presented in [123]. In this work, all services of cloud tech-
In [120], cloud technologies for a WBAN system is consid- nologies are utilized to create a robust system. Patients are
ered. All three services are used to create the cloud module monitored by their WBANs, with their data stored in the
in this system. SaaS is used to provide applications that allow cloud securely. A signature-based access control mechanism
authorized parties to work with health data, PaaS provides prevents unauthorized users from accessing data in the cloud.
tools for virtualization and database management, and IaaS PaaS is used to provide virtualization techniques that support
provides the hardware and required infrastructure. This sys- resource management and scheduling. Finally, SaaS in the
tem is primarily used to create and store health records, and to form of a user-friendly application that allows healthcare
allow health practitioners to check on their patient’s state as professionals to access patient data that they are authorized
required. Patients set up a profile, configure who has access to view. An alert system is also present in the software,
to their data, and decide whether they want monitoring to be triggering alarms when abnormal physiological parameters
continuous, on-request, or periodic. This system appears very are detected. To prove that value can be obtained from their
sophisticated, but further work needs to be done to ensure that big data management scheme, machine learning was applied
their proposed cloud model would be suitable as the volume to ECG signals and was shown to successfully classify con-
and variety of data increased over time. Additionally, further gestive heart failure in up to 98.9% of cases. Overall, this is an
security mechanisms would likely need to be incorporated to excellent example of how cloud technologies can be utilized
ensure that the patient’s data is truly secured and private. in healthcare. To get more value out of the datasets, further
Another sensor-based system is designed in [121], with the machine learning could be implemented. Additionally, it may
aim of monitoring patients’ emotional states. Big data man- be worth considering methods for reducing latency as the
agement is crucial in this system, as they aim to draw links velocity and volume of data inflow increases, as currently
between emotional responses and physiological changes. delay is shown to rise quite rapidly as the dataset expands.
Large amounts of physiological data are stored in the cloud Big data management for healthcare is not just a theoretical
module, organized sufficiently to enable data mining tech- concept. It is already being implemented in certain parts of
niques for the extraction of important information. To maxi- the world. In Australia, the Government has recently intro-
mize storage space, algorithms have been applied to remove duced the My Health Record scheme [124] , which utilizes
redundant or non-useful data from the database. The primary cloud storage methods. A patient’s My Health Record can
focus of cloud storage in this system is not to maintain a contain information about any allergies, conditions, current
26536 VOLUME 5, 2017

medications and treatments, pathology test results, and diag- and diseases, determining possible diagnoses based on those
nostic images. The patient can decide who has access to given to previous patients, developing suitable treatment
these records under normal circumstances. In an emergency plans for individual patients based on what has worked for
where the patient is unable to provide the information them- similar patients in the past, and much more. Each of these
selves (i.e. they are unconscious following an accident), then applications reduces human uncertainty and thus would help
limited-time emergency access can be granted to the respond- patients receive the most suitable care, as soon as possible.
ing healthcare providers so that the patient can receive the Cloud storage enables machine learning to perform rapidly
best possible treatment rapidly [125]. and effectively by providing large databases and high com-
The benefits of cloud technologies for big data manage- putational power. Standalone mobile devices would not have
ment are clear. It allows for virtually unlimited storage space, the storage capacity or computational resources to analyze
the provision of many useful services, and enables acces- data through machine learning, and thus it is essential that
sibility for patients and doctors. This gives patients more data is sent to the cloud. Information that could be obtained
control over their own healthcare, and simultaneously enables through machine learning includes disease trends, connec-
doctors to provide more suitable treatments without having to tions between symptoms, and the development of suitable
even meet with their patient in person. Additionally, big data treatment plans for individual patients.
management schemes that are designed to meet all 5 charac- Several researchers have identified the usefulness of com-
teristics of big data will enable data mining, machine learn- putational offloading in healthcare environments. In [120],
ing, and other forms of detailed analysis. This could lead to readings from WBAN sensors are transmitted to a mobile
new medical discoveries by identifying previously unknown phone, where some basic processing occurs. The information
trends in patient progression through an illness, finding new is then transmitted forward from smartphone to cloud, where
links between symptoms and conditions, determining new advanced processing occurs using feature selection and clas-
treatments that may be suitable for various conditions, and sification techniques. The meaningful information generated
much more. Big data management is essential for the IoT can then be stored or forwarded to healthcare practitioners.
healthcare system of the future. The primary weakness of this is that it is reliant on a smart-
phone, which would run out of battery within days at best.
2) DATA PROCESSING AND ANALYTICS It would be preferable for raw data to be transmitted straight
There are several types of data processing that can be per- to the cloud for complete processing, using a low-power
formed using cloud technologies, but the most relevant are communications standard such as the previously discussed
computational offloading and machine learning. Computa- NB-IoT. Additionally, the information obtained through pro-
tional offloading involves using the cloud to perform complex cessing is not analyzed further; instead, it is passed directly
data processing beyond the capabilities of low-resource wear- to a doctor who manually observes the result. In many cases,
able devices. By sending raw or partially processed sensor it would be possible to implement classification algorithms
data to the cloud, the computing resources of many machines that alert the doctor when an abnormal reading is detected in
can be utilized for processing. Using this high-powered com- a patient’s physiological signs.
puting environment over processing on the standalone mobile In [121], cloud computing is used to process the complex
device offers many advantages; more complex algorithms can raw data and send the meaningful results back to the patient
be executed, results can be obtained significantly faster, and through their sensing system. This is a strong concept, as it
battery life will be extended in mobile devices due to less allows for the processing power of cloud to be utilized by
processing occurring internally. complex sensors, whilst also enabling the patient to rapidly
Complicated sensor nodes such as those measuring ECG access their results and share them with a doctor as needed.
data, blood pressure, or accelerometers for fall detection Computational offloading for data processing is used
would benefit greatly from computational offloading. For in [123] to determine shapes of ECGs and evaluate whether
example, ECGs have a standard shape, and different devi- the shape is consistent with congestive heart failure. This
ations from this shape can indicate several different heart evaluation would be far too complicated to perform on a wear-
problems including arrhythmia, heart inflammation, and even able device, and thus is a perfect example of the usefulness
cardiac arrest. A small, low-powered sensor node could not of cloud technologies in data processing. The obtained ECG
analyze ECG readings rapidly using machine learning algo- information is stored after processing in the patient’s health
rithms to determine the patient’s state of health. However, if record, enabling them to view their results and share them
the raw data was offloaded to cloud, high-power processing with their doctor. Additionally, an alert can be triggered to
could be performed to determine the shape of the ECG before emergency services workers if the ECG shows heart rhythm
machine learning algorithms compare it to the characteristic consistent with congestive heart failure.
shape, identify any serious differences between the shapes, Some research comparing machine learning algorithms has
and determine what condition is causing them. also been conducted. In [126], deep neural networks (DNNs)
Machine learning can also be applied to large datasets are compared with gradient boosting decision tree (GBDT),
so as to obtain meaningful information from them, includ- logistic regression (LR), and support vector machine (SVM)
ing identifying previously unknown links between symptoms algorithms for predicting stroke. In this context, DNNs
VOLUME 5, 2017 26537

performed the best with 87.3% accuracy, while the GBDT for them to obtain insurance. Worse still, if the malicious
and LR algorithms were similar in performance at 86.8% attacker altered a patient’s health record, it could have detri-
and 86.6% accuracy respectively. SVMs performed the worst, mental effects on the patient’s health.
with only 83.9% accuracy. This suggests that DNNs are Access control policies and data encryption are two means
well suited to prediction tasks, but due to the complexity of securing cloud-centric healthcare systems. An access con-
of the algorithm they would need to run on cloud storage trol policy specifies who is authorized access to the patient’s
frameworks. health data, and how much access they are allowed. It would
Meanwhile, in [127], multi-layer perceptron (MLP), SVMs also implement an authentication mechanism (e.g. password,
for regression (SVR), generalized regression neural net- facial recognition, etc.) that verifies the identity of the party
works (GRNN), and k-nearest neighbour (kNN) regression attempting to access the data. Meanwhile, data encryption
approaches were compared for determining a person’s psy- provides security for the data whilst in data storage. Strong
chological wellness index based on five key parameters of data encryption would prevent an attacker from reading sen-
psychological health. In contrast to the previous work, in this sitive health information, even if they did gain access to the
case the SVM-based algorithm performed the best, while the database.
neural network approach performed second-worst. These two Some research has been conducted into developing secu-
comparison articles highlight the fact that the best machine rity mechanisms robust enough for healthcare applications.
learning algorithm for healthcare does not exist; rather, there In [128], a sophisticated access control scheme named
are algorithms that may be suitable for one context but ‘‘SafeProtect’’ is proposed, focusing on giving patients con-
completely unsuitable for another. As such, future system trol over their information. The patient creates a policy that
designers should compare machine learning algorithms that allows specific healthcare providers to access their health
may work for their purpose and determine which algorithm record, and can enforce limitations. The patients data is
has the most desirable characteristics for the system they are encrypted and stored in cloud storage. If a healthcare provider
designing. wants to access the patient’s health record, they must enter
Evidently, there are limited works that investigate compu- their credentials. Credentials are checked for validity before
tational offloading for IoT healthcare specifically. However, data is decrypted for the authorized healthcare provider. If the
it has been studied extensively for mobile devices such as healthcare provider has been assigned limited access by the
smartphones. The aforementioned surveys in [116] and [117] patient, then the policy control application will monitor their
present comprehensive overviews of the works in this field, behavior. If illegal actions are performed, such as the user
and demonstrate that energy efficiency and higher processing pressing the Ctrl + C shortcut for a party that is not allowed
capabilities can be readily achieved when utilized computa- to copy, then the action will be blocked and the patient will
tional offloading to the cloud. The interested reader is referred be notified that the healthcare provider tried to perform an
to these surveys for an overview of the related field of mobile illegal action. This is an intelligent mechanism to and could
cloud computing. easily be expanded to look for attempts to print or other
Overall, using computational offloading for data process- copy-based actions such as taking a screenshot using the
ing is invaluable in healthcare systems, particularly when ana- ‘‘Prt Scr’’ button. A significant advantage of the SafeProtect
lyzing complicated physiological parameters and conditions. scheme is that if the policy changes, keys do not need to be
Data processing techniques may also aid in the organized regenerated; healthcare providers’ credentials can simply be
storage of health data, as it can generate meaningful infor- added to or removed from the policy. The authors identify
mation such as a standard ECG from many different types of that they have not protected against all possible means of
ECG sensors. This is extremely useful in standardizing health copying and distributing healthcare information, making this
records, and would enable machine learning techniques to the main area in which future improvement can be made.
be applied to a big data set with more ease. Due to these Additional monitoring of keyboard shortcuts and ports for
benefits, computational offloading is vital for IoT healthcare illegal actions would help increase security. Another potential
systems to ensure that even the most complicated physio- improvement is immediately revoking access to the patient’s
logical parameters can be monitored, enabling the highest data if the healthcare provider performs an illegal action.
possible standard of healthcare for the patient. However, these improvements are minor and relatively easy
to make. Overall, SafeProtect is a sophisticated scheme that
B. SECURITY AND PRIVACY IN THE CLOUD adequately protects patients data from being accessed or used
Security remains a key issue in cloud-based systems. In a in an unauthorized manner, and the mechanisms implemented
healthcare environment, it is essential that a patient’s health could easily be expanded to protect against more actions.
information is readily accessible to authorized parties includ- In [129], a secure cloud scheme is proposed for ‘‘enhanced
ing doctors, nurses, specialists, and emergency services. It is living environments’’, which are generally comprised of both
also essential that the patient’s sensitive health data is kept wearable devices worn by the patient and smart home devices,
private. If malicious attacks revealed the patient’s health data, with the aim of supporting independent living for elderly
it could have many negative ramifications for the patient, or disabled persons. Access control is policy-based, where
including exposing them to identity theft or making it difficult access is given to authorized people. To ensure that only
26538 VOLUME 5, 2017

authorized people can access health records, biometrics are local server means that the health data cannot be easily shared
used to ensure that the person is who they claim to be. Users with another health authority. This could be problematic if a
are asked to provide a fingerprint, and facial recognition patient was travelling overseas and was hospitalized, as the
software calculates certain distances and angles on the face. healthcare providers would not be able to readily access the
The user is also asked to blink a random number of times, required health data. It would be valuable to make this system
so that a photograph cannot be used to ‘‘trick’’ the system. more accessible to authorized users, without compromising
Security of the data stored in the cloud is implemented by the high level of security that it clearly provides.
attempting to conceal possible weaknesses of the accessing In [132], fully homomorphic encryption (FHE) is consid-
applications. This is achieved by creating several applications ered as a scheme for protecting the security of the data. FHE
with the same functionality but varying implementations, allows for public-key encryption to be implemented before
and dynamically swapping them in and out of execution. being sent to the cloud. It also allows mathematic functions
While this system appears to be a strong solution, it would to be performed over on the data whilst it is encrypted,
likely benefit from encryption of patient data in the cloud. enabling basic machine learning without decrypting data in
Additionally, the identity management server adds another the cloud. If data is requested by an authorized user, the
vulnerability to the system, as information about many users’ key authority system can use a secure channel to provide
identities could be revealed if an attacker gained access to them with the secret key required to decrypt the data. This
this server. Ensuring that this server is secure is almost as is a promising approach to combining security with machine
important as securing patient’s healthcare data, as it is still learning, but it is not yet ready to be implemented into
sensitive information pertaining to the users of the system. IoT-based healthcare systems. In this work, it is shown that
Signal scrambling is considered as a means for encryption significant computational resources are required for FHE to
in [130]. In this application-layer scheme, a small portion of be successful. Additionally, only limited arithmetic can be
the data - termed ‘‘tiny data’’ - is used as the scrambling key, performed on the encrypted data. Whilst this work shows
and is shared between authorized parties. To reduce the risk improvement on previous FHE schemes, improving the speed
that a brute force attack could be used to determine what the and computational capabilities of FHE schemes remains an
tiny data used is, algorithms are in place to prevent the tiny active field of research.
data from containing statistically significant characteristics, FHE was also compared to traditional Advanced Encryp-
thus lowering the probability that such an attack would be tion Standard (AES) and emerging Attribute-Based Encryp-
successful. This scheme would be suitable in addition to other tion (ABE) in [133]. AES is a 128-bit block cypher, and the
security mechanisms, especially considering its position in scheme has already been optimized to run on small, lower-
the application layer allows for it to be built on top of exist- power devices. It is an accepted industry standard. ABE
ing communications standards. Further work could be done focuses on allowing access to multiple authorized parties,
regarding securing the exchange of tiny data, as this process much like the other schemes discussed in this section. The
could currently be susceptible to man-in-the-middle attacks if three types of encryption were compared in depth, and it was
the communications standard used is not secure. Additionally, found that there is no perfect solution for encryption. AES
it may be useful to occasionally generate new tiny data to use is the only scheme that is simple and low-latency enough for
for scrambling and unscrambling. This would likely decrease wearable devices, but ABE is the best scheme for enabling
the likelihood of a brute force attack succeeding. multiple parties to access private data. FHE is the only one
A steganography-based approach to access control is pre- that can provide computation and thus enable machine learn-
sented in [131]. In addition to protecting electronic health ing, but it is the slowest and requires the highest amount of
records with encryption techniques, this scheme also conceals storage.
the data within an ECG signal. This means that if the data Kocabas et al. [133] suggest that a scheme that uses
was intercepted by an attacker, they would likely assume that AES during acquisition, and then converts to FHE using an
it was simply a standard ECG signal, and would not even AES-to-FHE conversion scheme would be a suitable solution.
discover the hidden, encrypted message that is the health One advanced AES-to-FHE conversion scheme was proposed
record. After encrypting and concealing the health record, the in [134], whereby the conversion can be performed iteratively
health authority stores several parameters that are required without decrypting the data at any stage. This conversion
to reconstruct the data at their local servers. If the data is process requires significant computational power and time,
requested, authorization is performed using facial recognition but would be achievable within a cloud storage framework.
and the location of person making the request. If the user is We agree that this type of AES-to-FHE scheme is a rea-
authorized, then the health authorization recovers the data and sonable means for securing patient data. However, with the
provides it to them. While this appears to be a very secure clear advantages of ABE for cloud security in healthcare, we
system, it is limiting. Firstly, the use of location as a means believe that it would be beneficial to use it for acquisition
for authentication limits flexibility. For example, a paramedic and provision of health data. It is therefore suggested that
would likely not be able to access a patient’s health data an optimized ABE-to-FHE conversion scheme that at no
whilst travelling to their location. Additionally, the fact that point decrypts and re-encrypts the data would be extremely
the health authority stores the decryption parameters on their valuable for securing patient data.
VOLUME 5, 2017 26539

Overall, there is still significant research opportunity in specifically on the need for privacy and security in health-
developing a completely suitable encryption scheme for care applications. All of these works have made significant
healthcare systems that rely on both low-power wearables and improvements on previous methods, but there is still no per-
big data cloud storage. Increasing speed, decreasing compu- fect solution for security within the cloud.
tational requirements, and enabling high-level machine learn- Finally, several works have identified that machine
ing are all areas for improvement that should be investigated learning is extremely important in healthcare applications.
further. Machine learning as a whole is a topic that has been widely
considered by previous researchers. However, minimal works
VI. FINDINGS AND RECOMMENDATIONS exist on implementing machine learning specifically for diag-
Upon completion of this thorough survey of existing tech- nostics or other health-related purposes.
nologies, several lessons have been learned. In this section,
we will present a summary of these lessons, followed by B. RECOMMENDATIONS FOR FUTURE WORKS
providing recommendations for future work with the aim The lessons learned through conducting this survey highlight
of directing researchers to areas that would fill the most several areas for further research. In terms of sensors, much
significant gaps in the literature. progress has been made but there are still no available devices
that match the accuracy of hospital-grade devices without
A. LESSONS LEARNED compromising energy efficiency or wearability. This is espe-
In terms of the key sensor types, it was found that there are cially true of complex devices such as blood pressure and
several options for suitable pulse sensors, while researchers respiratory rate sensors, both of which would be invaluable to
agree that thermistor-type temperature sensors are already the field of medicine. As such, further research efforts should
suitable for use in measuring human body temperature. The be made towards improving the quality of these sensors until
photoplethysmographic method for implementing blood oxy- they are highly accurate, reliable, and comfortably wearable.
gen level monitoring is also widely agreed upon. The issues In our own future works, we will be placing particular focus
that remain with these devices are primarily making them on developing a blood pressure monitor that is more wearable
robust against motion and ensuring energy efficiency without than the works presented in this paper, without compromis-
compromising accuracy. Meanwhile, there is little consen- ing accuracy. We will also look at reducing the impact of
sus regarding the most suitable respiratory rate sensor for motion on sensors, particularly for respiratory rate and pulse
general-purpose use, and there remains much work to be done sensors.
on blood pressure sensors in order to achieve an accurate and In terms of communications standards, it would be worth-
truly wearable sensor that could be deployed on a wide scale. while to develop wearable healthcare systems that are reliant
The progress made towards suitable solutions largely on the emerging NB-IoT standard. As this is an extremely
reflects the frequency with which each topic has been con- new standard, no known work has implemented it into a
sidered in the literature. There are many papers that exist healthcare environment despite its obvious advantages for
on pulse, body temperature, and blood oxygen monitoring, this field. In our own future works, we will be implementing
both current and classical. On the other hand, respiratory rate NB-IoT into healthcare devices to confirm its suitability,
monitoring with wearable sensors is a newer concept in the before using it as the foundation communications standard
literature, and is further divided into subsections for varying for a healthcare system that is being developed in accordance
sensor types (nasal, stretch, pressure, and so on). Meanwhile, with the model proposed by this paper.
research on blood pressure sensors within the literature is Data storage using cloud technologies has been extensively
minimal, and is very much still in its infancy when compared considered, but data processing is an area in which further
to research on other sensor types. research should be conducted. The development of cloud-
In terms of short-range communications standards, it was based algorithms that are capable of processing raw data from
learned that Bluetooth Low Energy has the highest suitability complex sensors and extract meaningful information about a
for healthcare, and can be immediately implemented into person’s health should be continued.
healthcare systems being designed now and in the future. Machine learning is another branch of data processing
As discussed in Section IV.A., BLE has already been imple- that would be extremely valuable in healthcare scenarios.
mented by systems discussed in a reasonable number of If applied in the high-power computing environment of
papers. Meanwhile, it was found that NB-IoT has the high- the cloud, machine learning could provide diagnostics for
est suitability for low-power, long-range communications in patients, make new discoveries about disease trends, and
healthcare, and will likely be deployed rapidly once the stan- aid in the development of treatment plans. Despite these
dard is finalized, due to the ability to re-use existing cellular clear advantages, machine learning has not yet been widely
station hardware for this purpose. explored for healthcare applications, providing a significant
From our research, it is found that the problem of cloud research opportunity. This opportunity should be seized by
storage has largely been solved. Additionally, extensive researchers aiming to make notable improvements to the field
research has been conducted into improving security mech- of IoT-based healthcare. In our own future work, we will be
anisms within the cloud. Several works have focused more investigating clustering and logistic regression algorithms as
26540 VOLUME 5, 2017

a means for providing diagnostics based on vital and other future research. Machine learning and the development of a
sign information. secure yet lightweight encryption scheme for cloud storage
There is still much room for improvement in security and were the two areas that provide the most opportunity for
privacy for cloud-based healthcare. No known encryption researchers seeking to make significant improvements in the
scheme is ideal for protecting data whilst providing acces- field of IoT-based healthcare.
sibility for authorized parties and enabling machine learning.
ABE and FHE are schemes that provide appealing character- REFERENCES
istics, but are not lightweight enough for implementation into [1] Australian Institute of Health and Welfare. (2014). Australia’s
wearable devices. Improving these schemes is the first active Health. [Online]. Available: http://www.aihw.gov.au/WorkArea/
DownloadAsset.aspx-?id=60129548150
area of research. Upon improving the schemes individually, [2] E. Perrier, Positive Disruption: Healthcare, Ageing and Participation in
a lightweight ABE-FHE hybrid scheme should be considered, the Age of Technology. Sydney, NSW, Australia: The McKell Institute,
as it could potentially provide all the desirable characteristics 2015.
[3] P. Gope and T. Hwang, ‘‘BSN-care: A secure IoT-based modern health-
for cloud-based healthcare security. care system using body sensor network,’’ IEEE Sensors J., vol. 16, no. 5,
Overall, there is no known end-to-end system for general pp. 1368–1376, Mar. 2016.
or specific purposes that contains all components in our [4] N. Zhu et al., ‘‘Bridging e-health and the Internet of Things: The SPHERE
proposed model; wearable sensors, short- and long-range project,’’ IEEE Intell. Syst., vol. 30, no. 4, pp. 39–46, Jul./Aug. 2015.
[5] S.-H. Chang, R.-D. Chiang, S.-J. Wu, and W.-T. Chang, ‘‘A context-
communications, cloud-based storage, and machine learning. aware, interactive M-health system for diabetics,’’ IT Prof., vol. 18, no. 3,
Developing such a system would be a significant achievement pp. 14–22, May/Jun. 2016.
in the field of IoT-based healthcare, and should be considered [6] C. F. Pasluosta, H. Gassner, J. Winkler, J. Klucken, and B. M. Eskofier,
‘‘An emerging era in the management of Parkinson’s disease: Wearable
as the ultimate goal for researchers in this area. In our own technologies and the Internet of Things,’’ IEEE J. Biomed. Health Inform.,
future works, we will be striving to reach this goal through vol. 19, no. 6, pp. 1873–1881, Nov. 2015.
the development of a wearable, IoT-based system for the [7] Y. J. Fan, Y. H. Yin, L. D. Xu, Y. Zeng, and F. Wu, ‘‘IoT-based
smart rehabilitation system,’’ IEEE Trans. Ind. Informat., vol. 10, no. 2,
provision of emergency healthcare that incorporates health pp. 1568–1577, May 2014.
sign monitoring, machine learning for diagnostics, and long- [8] S. Sarkar and S. Misra, ‘‘From micro to nano: The evolution of wire-
range communications via LPWANs to notify emergency less sensor-based health care,’’ IEEE Pulse, vol. 7, no. 1, pp. 21–25,
Jan./Feb. 2016.
service providers when a patient needs urgent help. [9] Y. Yin, Y. Zeng, X. Chen, and Y. Fan, ‘‘The Internet of Things
in healthcare: An overview,’’ J. Ind. Inf. Integr., vol. 1, pp. 3–13,
VII. CONCLUSION Mar. 2016. [Online]. Available: http://www.sciencedirect.com/science/
article/pii/S2452414X16000066
In this work, we have proposed a unique model for future [10] S. M. R. Islam, D. Kwak, H. Kabir, M. Hossain, and K.-S. Kwak,
IoT-based healthcare systems, which can be applied to both ‘‘The Internet of Things for health care: A comprehensive survey,’’ IEEE
general systems and systems that monitor specific condi- Access, vol. 3, pp. 678–708, 2015.
[11] D. V. Dimitrov, ‘‘Medical Internet of Things and big data in healthcare,’’
tions. We then presented a thorough and systematic overview
Healthcare Inform. Res., vol. 22, no. 3, pp. 156–163, Jul. 2016. [Online].
of the state-of-the-art works relating to each component of Available: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4981575/
the proposed model. Several wearable, non-intrusive sensors [12] C. C. Y. Poon, B. P. L. Lo, M. R. Yuce, A. Alomainy, and Y. Hao, ‘‘Body
were presented and analyzed, with particular focus on those sensor networks: In the era of big data and beyond,’’ IEEE Rev. Biomed.
Eng., vol. 8, pp. 4–16, 2015.
monitoring vital signs, blood pressure, and blood oxygen lev- [13] P. K. Schwab, The Fourth Industrial Revolution: What it Means, and How
els. Short-range and long-range communications standards to Respond. Cologny, Switzerland: World Economic Forum, 2016.
were then compared in terms of suitability for healthcare [14] Smart Parking. (2017). SmartEye, SmartRep, and RFID Technology—
Westminster City Council—London. [Online]. Available: http://
applications. BLE and NB-IoT emerged as the most suitable www.smartparking.com/keep-up-to-date/case-studies/3-500-vehicle-
standards for short-range and long-range communications in detection-sensors-and-epermit-technology-in-the-city-of-westminster-
healthcare respectively. london
[15] University of New England. (2017). SMART Farm. [Online]. Available:
Recent works utilizing cloud technologies for data storage http://www.une.edu.au/research/research-centres-institutes/smart-farm
were presented, and showed that cloud is the best means for [16] Sensus. (2017). Smart Water—Smarter at Every Point. [Online]. Avail-
storing and organizing big data in healthcare. It is also shown able: www.sensus.com/internet-of-things/smart-water
[17] H. El-Sayed and G. Thandavarayan, ‘‘Congestion detection and propaga-
by several works that significantly better data processing can tion in urban areas using histogram models,’’ IEEE Internet Things J., to
be performed in the cloud than can be performed by wearable be published.
devices with their limited resources. The most significant [18] C. A. Tokognon, B. Gao, G. Tian, and Y. Yan, ‘‘Structural health moni-
toring framework based on Internet of Things: A survey,’’ IEEE Internet
drawback of using cloud is that it introduces security risks, Things J., vol. 4, no. 3, pp. 619–635, Jun. 2017.
and as such we presented several works focused on improving [19] K. M. Alam, M. Saini, and A. E. Saddik, ‘‘Toward social Internet of
security in the cloud. It was found that access control policies vehicles: Concept, architecture, and applications,’’ IEEE Access, vol. 3,
and encryption can significantly enhance security, but that no pp. 343–357, Mar. 2015.
[20] S. Tan, D. De, W.-Z. Song, J. Yang, and S. K. Das, ‘‘Survey of security
known standard is suitable for immediate application into a advances in smart grid: A data driven approach,’’ IEEE Commun. Surveys
wearable, IoT-based healthcare system. Tuts., vol. 19, no. 1, pp. 397–422, 1st Quart., 2017.
Based on our analysis of state-of-the-art technologies in the [21] R. C. A. Alves, L. B. Gabriel, B. T. D. Oliveira, C. B. Margi, and
F. C. L. D. Santos, ‘‘Assisting physical (hydro)therapy with wireless
fields of wearable sensors, communications standards, and sensors networks,’’ IEEE Internet Things J., vol. 2, no. 2, pp. 113–120,
cloud technology, we identified several significant areas for Apr. 2015.
VOLUME 5, 2017 26541

[22] G. Wolgast, C. Ehrenborg, A. Israelsson, J. Helander, E. Johansson, [45] O. Atalay, W. R. Kennon, and E. Demirok, ‘‘Weft-knitted strain sensor for
and H. Manefjord, ‘‘Wireless body area network for heart attack detec- monitoring respiratory rate and its electro-mechanical modeling,’’ IEEE
tion [education corner],’’ IEEE Antennas Propag. Mag., vol. 58, no. 5, Sensors J., vol. 15, no. 1, pp. 110–122, Jan. 2015.
pp. 84–92, Oct. 2016. [46] P. Aqueveque, C. Gutiérrez, F. S. Rodríguez, E. J. Pino, A. Morales, and
[23] M. A. Cretikos, R. Bellomo, K. Hillman, J. Chen, S. Finfer, and E. P. Wiechmann, ‘‘Monitoring physiological variables of mining workers
A. Flabouris, ‘‘Respiratory rate: The neglected vital sign,’’ Med. J. at high altitude,’’ IEEE Trans. Ind. Appl., vol. 53, no. 3, pp. 2628–2634,
Austral., vol. 188, no. 11, pp. 657–659, 2008. May/Jun. 2017.
[24] B. Xu, L. D. Xu, H. Cai, C. Xie, J. Hu, and F. Bu, ‘‘Ubiquitous data [47] P. Narczyk, K. Siwiec, and W. A. Pleskacz, ‘‘Precision human body
accessing method in IoT-based information system for emergency medi- temperature measurement based on thermistor sensor,’’ in Proc. IEEE
cal services,’’ IEEE Trans Ind. Informat., vol. 10, no. 2, pp. 1578–1586, 19th Int. Symp. Design Diagnostics Electron. Circuits Syst. (DDECS),
May 2014. Apr. 2016, pp. 1–5.
[25] I. Olaronke and O. Oluwaseun, ‘‘Big data in healthcare: Prospects, chal- [48] T. Nakamura et al., ‘‘Development of flexible and wide-range polymer-
lenges and resolutions,’’ in Proc. Future Technol. Conf. (FTC), Dec. 2016, based temperature sensor for human bodies,’’ in Proc. IEEE-EMBS Int.
pp. 1152–1157. Conf. Biomed. Health Inform. (BHI), Feb. 2016, pp. 485–488.
[26] J. Zhou, Z. Cao, X. Dong, and A. V. Vasilakos, ‘‘Security and privacy [49] A. Eshkeiti et al., ‘‘A novel self-supported printed flexible strain sensor
for cloud-based IoT: Challenges,’’ IEEE Commun. Mag., vol. 55, no. 1, for monitoring body movement and temperature,’’ in Proc. IEEE SEN-
pp. 26–33, Jan. 2017. SORS, Nov. 2014, pp. 1615–1618.
[27] Shake It Up Australia Foundation. (2017). Symptoms of Parkinson’s. [50] Heart Foundation. (2017). High Blood Pressure Statistics. [Online].
[Online]. Available: https://shakeitup.org.au/understanding- Available: www.heartfoundation.org.au/about-us/what-we-do/heart-
parkinsons/symptoms-of-parkinsons/ disease-in-australia/high-blood-pressure-statistics
[28] J. Ženko, M. Kos, and I. Kramberger, ‘‘Pulse rate variability and blood [51] S. S. Thomas, V. Nathan, C. Zong, K. Soundarapandian, X. Shi, and
oxidation content identification using miniature wearable wrist device,’’ R. Jafari, ‘‘BioWatch: A noninvasive wrist-based blood pressure monitor
in Proc. Int. Conf. Syst., Signals Image Process. (IWSSIP), May 2016, that incorporates training techniques for posture and subject variabil-
pp. 1–4. ity,’’ IEEE J. Biomed. Health Inform., vol. 20, no. 5, pp. 1291–1300,
[29] Garmin. (2017). HSM-Tri. [Online]. Available: https://buy.garmin.com/ Sep. 2016.
en-AU/AU/p/136403 [52] H. Lin, W. Xu, N. Guan, D. Ji, Y. Wei, and W. Yi, ‘‘Noninvasive and
[30] Polar. (2017). H7 Heart Rate Sensor. [Online]. Available: https://www. continuous blood pressure monitoring using wearable body sensor net-
polar.com/au-en/products/accessories/H7_heart_rate_sensor works,’’ IEEE Intell. Syst., vol. 30, no. 6, pp. 38–48, Nov./Dec. 2015.
[31] FitBit. (2017). FitBit PurePulse. [Online]. Available: https://www.fitbit. [53] Y.-L. Zheng, B. P. Yan, Y.-T. Zhang, and C. C. Y. Poon, ‘‘An arm-
com/au/purepulse band wearable device for overnight and cuff-less blood pressure mea-
[32] TomTom. (2017). TomTom Spark Cardio. [Online]. Available: https:// surement,’’ IEEE Trans. Biomed. Eng., vol. 61, no. 7, pp. 2179–2186,
www.tomtom.com/en_au/sports/fitness-watches/gps-watch-cardio- Jul. 2014.
spark/black-large/ [54] D. Griggs et al., ‘‘Design and development of continuous cuff-less blood
[33] H. Lee, H. Ko, C. Jeong, and J. Lee, ‘‘Wearable photoplethysmographic pressure monitoring devices,’’ in Proc. IEEE SENSORS, Oct./Nov. 2016,
sensor based on different LED light intensities,’’ IEEE Sensors J., vol. 17, pp. 1–3.
no. 3, pp. 587–588, Feb. 2017. [55] Y. Zhang, M. Berthelot, and B. P. Lo, ‘‘Wireless wearable photoplethys-
[34] Y. Shu, C. Li, Z. Wang, W. Mi, Y. Li, and T.-L. Ren, ‘‘A pressure sensing mography sensors for continuous blood pressure monitoring,’’ in Proc.
system for heart rate monitoring with polymer-based pressure sensors IEEE Wireless Health (WH), Oct. 2016, pp. 1–8.
and an anti-interference post processing circuit,’’ Sensors, vol. 15, no. 2, [56] J. Wannenburg and R. Malekian, ‘‘Body sensor network for mobile
pp. 3224–3235, 2015. health monitoring, a diagnosis and anticipating system,’’ IEEE Sensors
[35] D. Wang, D. Zhang, and G. Lu, ‘‘A novel multichannel wrist pulse system J., vol. 15, no. 12, pp. 6839–6852, Dec. 2015.
with different sensor arrays,’’ IEEE Trans. Instrum. Meas., vol. 64, no. 7, [57] S. V. Gubbi and B. Amrutur, ‘‘Adaptive pulse width control and sampling
pp. 2020–2034, Jul. 2015. for low power pulse oximetry,’’ IEEE Trans. Biomed. Circuits Syst., vol. 9,
[36] D. Wang, D. Zhang, and G. Lu, ‘‘An optimal pulse system design by no. 2, pp. 272–283, Apr. 2015.
multichannel sensors fusion,’’ IEEE J. Biomed. Health Informat., vol. 20, [58] B. Venema, H. Gehring, I. Michelsen, N. Blanik, V. Blazek, and
no. 2, pp. 450–459, Mar. 2016. S. Leonhardt, ‘‘Robustness, specificity, and reliability of an in-ear pulse
[37] W. Zuo, P. Wang, and D. Zhang, ‘‘Comparison of three different types oximetric sensor in surgical patients,’’ IEEE J. Biomed. Health Informat.,
of wrist pulse signals by their physical meanings and diagnosis perfor- vol. 18, no. 4, pp. 1178–1185, Jul. 2014.
mance,’’ IEEE J. Biomed. Health Informat., vol. 20, no. 1, pp. 119–127, [59] G. Pang and C. Ma, ‘‘A neo-reflective wrist pulse oximeter,’’ IEEE
Jan. 2016. Access, vol. 2, pp. 1562–1567, 2014.
[38] Y.-J. An, B.-H. Kim, G.-H. Yun, S.-W. Kim, S.-B. Hong, and J.-G. Yook, [60] V. P. Rachim and W.-Y. Chung, ‘‘Wearable noncontact armband for
‘‘Flexible non-constrained RF wrist pulse detection sensor based on mobile ECG monitoring system,’’ IEEE Trans. Biomed. Circuits Syst.,
array resonators,’’ IEEE Trans. Biomed. Circuits Syst., vol. 10, no. 2, vol. 10, no. 6, pp. 1112–1118, Dec. 2016.
pp. 300–308, Apr. 2016. [61] W. Von Rosenberg, T. Chanwimalueang, V. Goverdovsky, D. Looney,
[39] S. Milici, J. Lorenzo, A. Lázaro, R. Villarino, and D. Girbau, ‘‘Wireless D. Sharp, and D. P. Mandic, ‘‘Smart helmet: Wearable multichannel
breathing sensor based on wearable modulated frequency selective sur- ECG and EEG,’’ IEEE J. Transl. Eng. Health Med., vol. 4, 2016,
face,’’ IEEE Sensors J., vol. 17, no. 5, pp. 1285–1292, Mar. 2017. Art. no. 2700111.
[40] C. Varon, A. Caicedo, D. Testelmans, B. Buyse, and S. van Huffel, [62] E. Spanò, S. D. Pascoli, and G. Iannaccone, ‘‘Low-power wearable ECG
‘‘A novel algorithm for the automatic detection of sleep apnea monitoring system for multiple-patient remote monitoring,’’ IEEE Sen-
from single-lead ECG,’’ IEEE Trans. Biomed. Eng., vol. 62, no. 9, sors J., vol. 16, no. 13, pp. 5452–5462, May 2016.
pp. 2269–2278, Sep. 2015. [63] G. Li, B.-L. Lee, and W.-Y. Chung, ‘‘Smartwatch-based wearable EEG
[41] D. Oletic and V. Bilas, ‘‘Energy-efficient respiratory sounds sensing for system for driver drowsiness detection,’’ IEEE Sensors J., vol. 15, no. 12,
personal mobile asthma monitoring,’’ IEEE Sensors J., vol. 16, no. 23, pp. 7169–7180, Dec. 2015.
pp. 8295–8303, Dec. 2016. [64] U. Ha et al., ‘‘A wearable EEG-HEG-HRV multimodal system with
[42] X. Yang et al., ‘‘Textile fiber optic microbend sensor used for heartbeat simultaneous monitoring of tES for mental health management,’’ IEEE
and respiration monitoring,’’ IEEE Sensors J., vol. 15, no. 2, pp. 757–761, Trans. Biomed. Circuits Syst., vol. 9, no. 6, pp. 758–766, Dec. 2015.
Feb. 2015. [65] J. Wannenburg and R. Malekian, ‘‘Physical activity recognition from
[43] S. D. Min, Y. Yun, and H. Shin, ‘‘Simplified structural textile respiration smartphone accelerometer data for user context awareness sensing,’’
sensor based on capacitive pressure sensing method,’’ IEEE Sensors J., IEEE Trans. Syst., Man, Cybern., Syst., vol. 47, no. 12, pp. 3142–3149,
vol. 14, no. 9, pp. 3245–3251, Sep. 2014. Dec. 2017.
[44] I. Mahbub et al., ‘‘A low-power wireless piezoelectric sensor-based res- [66] P. Vallabh, R. Malekian, N. Ye, and D. C. Bogatinoska, ‘‘Fall detection
piration monitoring system realized in CMOS process,’’ IEEE Sensors J., using machine learning algorithms,’’ in Proc. 24th Int. Conf. Softw.,
vol. 17, no. 6, pp. 1858–1864, Mar. 2017. Telecommun. Comput. Netw. (SoftCOM), Sep. 2016, pp. 1–9.
26542 VOLUME 5, 2017

[67] K. Ozcan, S. Velipasalar, and P. K. Varshney, ‘‘Autonomous fall detection [90] J. Gil et al., ‘‘A fully integrated low-power high-coexistence 2.4-GHz
with wearable cameras by using relative entropy distance measure,’’ IEEE ZigBee transceiver for biomedical and healthcare applications,’’ IEEE
Trans. Human–Mach. Syst., vol. 47, no. 1, pp. 31–39, Feb. 2017. Trans. Microw. Theory Techn., vol. 62, no. 9, pp. 1879–1889, Sep. 2014.
[68] K. Ozcan and S. Velipasalar, ‘‘Wearable camera- and accelerometer-based [91] Sigfox. (2017). Getting Started. [Online]. Available: http://makers.
fall detection on portable devices,’’ IEEE Embedded Syst. Lett., vol. 8, sigfox.com/getting-started/
no. 1, pp. 6–9, Mar. 2016. [92] WAVIoT. (2016). Comparison of LPWAN Technologies. [Online].
[69] P. Pierleoni, A. Belli, L. Palma, M. Pellegrini, L. Pernini, and S. Valenti, Available: http://dgmatics.com/technology/waviot-lpwan-technology-
‘‘A high reliability wearable device for elderly fall detection,’’ IEEE comparison
Sensors J., vol. 15, no. 8, pp. 4544–4553, Aug. 2015. [93] VT Networks. (2016). Sigfox Network Overview for Technical
[70] P. Pierleoni et al., ‘‘A wearable fall detector for elderly people based Professionals. [Online]. Available: http://vtnetworks.ie/wp-content/
on AHRS and barometric sensor,’’ IEEE Sensors J., vol. 16, no. 17, uploads/2016/12/Sigfox-Network-Overview.pdf
pp. 6733–6744, Sep. 2016. [94] Thinxtra. (2017). Empowering Internet of Things in Australia,
[71] H.-C. Chang, Y.-L. Hsu, S.-C. Yang, J.-C. Lin, and Z.-H. Wu, ‘‘A wearable New Zealand and Hong Kong. [Online]. Available: https://www.thinxtra.
inertial measurement system with complementary filter for gait analysis com/
of patients with stroke or Parkinson’s disease,’’ IEEE Access, vol. 4, [95] R. Ratasuk, B. Vejlgaard, N. Mangalvedhe, and A. Ghosh, ‘‘NB-IoT
pp. 8442–8453, 2016. system for M2M communication,’’ in Proc. IEEE Wireless Commun.
[72] H. F. Maqbool, M. A. B. Husman, M. I. Awad, A. Abouhossein, Netw. Conf., Apr. 2016, pp. 1–5.
N. Iqbal, and A. A. Dehghani-Sanij, ‘‘A real-time gait event detection for [96] Burdekin Shire Council. (2017). Population of the Burdekin Shire.
lower limb prosthesis control and evaluation,’’ IEEE Trans. Neural Syst. [Online]. Available: http://www.burdekin.qld.gov.au/council/business/-
Rehabil. Eng., vol. 25, no. 9, pp. 1500–1509, Sep. 2017. population-of-the-burdekin-shire/
[73] T. T. Pham et al., ‘‘Freezing of gait detection in Parkinson’s disease: [97] National Regional Profile: Burdekin (Statistical Area Level 2)—
A subject-independent detector using anomaly scores,’’ IEEE Trans. Environment/Energy, Austral. Bureau Statist., Canberra, Australia, 2013.
Biomed. Eng., vol. 64, no. 11, pp. 2719–2728, Nov. 2017. [Online]. [98] LoRaWAN: What is it? A Technical Overview of LoRa and LoRaWAN,
Available: http://ieeexplore.ieee.org/document/7845616/ 1st ed., LoRa Alliance, Beaverton, OR, USA, 2015. [Online]. Available:
[74] G. Cola, M. Avvenuti, A. Vecchio, G.-Z. Yang, and B. Lo, ‘‘An on-node https://www.lora-alliance.org/lorawan-white-papers and https://docs.
processing approach for anomaly detection in gait,’’ IEEE Sensors J., wixstatic.com/ugd/eccc1a_ed71ea1cd969417493c74e4a13c55685.pdf
vol. 15, no. 11, pp. 6640–6649, Nov. 2015. [99] R. Miller, LoRa security: Building a Secure LoRa Solution, 1st ed.,
[75] S. Haxha and J. Jhoja, ‘‘Optical based noninvasive glucose monitor- Basingstoke, UK: MWR InfoSecurity, 2016. [Online]. Available:
ing sensor prototype,’’ IEEE Photon. J., vol. 8, no. 6, Dec. 2016, https://labs.mwrinfosecurity.com/assets/BlogFiles/mwri-LoRa-security-
Art. no. 6805911. guide-1.2-2016-03-22.pdf
[76] X. Xiao and Q. Li, ‘‘A noninvasive measurement of blood glucose concen- [100] N. Mangalvedhe, R. Ratasuk, and A. Ghosh, ‘‘NB-IoT deployment
tration by UWB microwave spectrum,’’ IEEE Antennas Wireless Propag. study for low power wide area cellular IoT,’’ in Proc. IEEE 27th Annu.
Lett., vol. 16, pp. 1040–1043, 2016. Int. Symp. Pers., Indoor, Mobile Radio Commun. (PIMRC), Sep. 2016,
pp. 1–6.
[77] K.-H. Chang, ‘‘Bluetooth: A viable solution for IoT? [Industry Perspec-
[101] NB-IoT, 1st ed., Huawei Technologies Co. Ltd., Shenzhen,
tives],’’ IEEE Wireless Commun., vol. 21, no. 6, pp. 6–7, Dec. 2014.
China, 2017. [Online]. Available: http://www.huawei.com/minisite/
[78] Bluetooth. (2009). SIG Introduces Bluetooth Low Energy Wireless Tech-
iot/img/nb_iot_whitepaper_en.pdf
nology, the Next Generation Bluetooth Wireless Technology. [Online].
[102] R. Ratasuk, N. Mangalvedhe, Y. Zhang, M. Robert, and
Available: https://www.bluetooth.com/news/pressreleases/2009/-12/17/
J.-P. Koskinen, ‘‘Overview of narrowband IoT in LTE Rel-13,’’ in
sig-introduces-bluetooth-low-energy-wireless-technologythe-next-
Proc. IEEE Conf. Standards Commun. Netw. (CSCN), Oct. 2016,
generation-of-bluetooth-wireless-technology
pp. 1–7.
[79] M. R. Palattella et al., ‘‘Internet of Things in the 5G era: Enablers,
[103] LTE Evolution for IoT Connectivity, 1st ed., Nokia, Espoo, Finland, 2017.
architecture, and business models,’’ IEEE J. Sel. Areas Commun., vol. 34,
[Online]. Available: https://resources.ext.nokia.com/asset/200178
no. 3, pp. 510–527, Mar. 2016.
[104] 2011 Census Snapshot: Households Families, Census Inf. Scheme,
[80] Security, Bluetooth Low Energy, 1st ed., Bluetooth, Kirkland, WA, London, U.K., 2012.
USA, 2017, pp. 1–5. [Online]. Available: https://www.bluetooth.com/~/ [105] 3GPP Low Power Wide Area Technologies, 1st ed., GSMA, London,
media/files/specification/bluetooth-low-energy-security.ashx U.K., 2016. Available: https://www.gsma.com/iot/3gpp-low-power-
[81] M. Singh and N. Jain, ‘‘Performance and evaluation of smartphone based wide-area-technologies-white-paper/
wireless blood pressure monitoring system using Bluetooth,’’ IEEE Sen- [106] Symphony Link vs. LoRaWAN, 1st ed., Link Labs, Annapolis, MD,
sors J., vol. 16, no. 23, pp. 8322–8328, Dec. 2016. USA, 2017. [Online]. Available: https://www.link-labs.com/whitepaper-
[82] H. T. Cheng and W. Zhuang, ‘‘Bluetooth-enabled in-home patient mon- symphony-link-vs-lorawan
itoring system: Early detection of Alzheimer’s disease,’’ IEEE Wireless [107] A Comprehensive Look at Low Power, Wide Area Networks for ‘Internet
Commun., vol. 17, no. 1, pp. 74–79, Feb. 2010. of Things’ Engineers and Decision Makers, Link Labs, Annapolis, MD,
[83] Digi. (2017). ZigBee Wireless Standard. [Online]. Available: USA, 2016.
https://www.digi.com/resources/standards-and-technologies/rfmodems/ [108] nWave Network, nWave, Brussels, Belgium, 2017.
zigbee-wireless-standard [109] Which Weightless Standard?, Weightless, Cambridge, U.K., 2017.
[84] Digi. (2009). XBee/XBee-Pro RF Modules (Datasheet). [Online]. [110] Weightless-N Standard Accelerates, Weightless, Cambridge, U.K., 2014.
Available: https://www.sparkfun.com/datasheets/Wireless/Zigbee/- [111] What is NB-Fi Protocol, WAVIoT, Houston, TX, USA, 2017.
XBee-Datasheet.pdf [112] Frequently Asked Questions, WAVIoT, Houston, TX, USA, 2017.
[85] XBee-PRO 900HP/XBee-PRO XSC RF Modules, Digi, Shah Alam, [113] How RPMA Works: An Educational Guide, 1st ed., Ingenu, San
Malaysia, 2014. Diego, CA, USA, 2017. [Online]. Available: https://www.ingenu.com/
[86] ZigBee Alliance. (2017). ZigBee IP and 910IP. [Online]. Available: portfolio/how-rpma-works-white-paper/
http://www.zigbee.org/zigbee-for-developers/network-specifications/ [114] Cellular Networks for Massive IoT, 1st ed., Ericsson, Stockholm,
zigbeeip/ Sweden, 2016. [Online]. Available: https://www.ericsson.com/assets/
[87] Daintree Networks. (2017). ZigBee Security. [Online]. Available: local/publications/white-papers/wp_iot.pdf
http://www.daintree.net/resources/zigbee-security/ [115] S. Bera, S. Misra, and J. J. P. C. Rodrigues, ‘‘Cloud computing appli-
[88] T. Zillner, ZigBee Exploited, 1st ed. Vienna, Austria: Cognosec, 2015. cations for smart grid: A survey,’’ IEEE Trans. Parallel Distrib. Syst.,
[Online]. Available: https://www.blackhat.com/docs/us-15/materials/us- vol. 26, no. 5, pp. 1477–1494, May 2015.
15-Zillner-ZigBee-Exploited-The-Good-The-Bad-And-The-Ugly- [116] P. Mach and Z. Becvar, ‘‘Mobile edge computing: A survey on architec-
wp.pdf ture and computation offloading,’’ IEEE Commun. Surveys Tuts., vol. 19,
[89] D. M. Taub, S. B. Leeb, E. C. Lupton, R. T. Hinman, J. Ziesel, and no. 3, pp. 1628–1656, 3rd Quart., 2017.
S. Blackler, ‘‘The escort system: A safety monitor for people living with [117] A. R. Khan, M. Othman, S. A. Madani, and S. U. Khan, ‘‘A survey of
Alzheimer’s disease,’’ IEEE Pervasive Comput., vol. 10, no. 2, pp. 68–77, mobile cloud computing application models,’’ IEEE Commun. Surveys
Apr./Jun. 2011. Tuts., vol. 16, no. 1, pp. 393–413, Feb. 2014.
VOLUME 5, 2017 26543

[118] N. Sultan, ‘‘Making use of cloud computing for healthcare provision: WEI XIANG (S’00–M’04–SM’10) received the
Opportunities and challenges,’’ Int. J. Inf. Manage., vol. 34, no. 2, B.Eng. and M.Eng. degrees in electronic engi-
pp. 177–184, Apr. 2014. [Online]. Available: http://www.sciencedirect. neering from the University of Electronic Science
com/science/article/pii/S0268401213001680 and Technology of China, Chengdu, China, in
[119] L. A. Tawalbeh, R. Mehmood, E. Benkhlifa, and H. Song, ‘‘Mobile cloud 1997 and 2000, respectively, and the Ph.D. degree
computing model and big data analysis for healthcare applications,’’ IEEE in telecommunications engineering from the Uni-
Access, vol. 4, pp. 6171–6180, 2016. versity of South Australia, Adelaide, Australia,
[120] S. Ghanavati, J. Abawajy, and D. Izadi, ‘‘An alternative sensor cloud
in 2004.
architecture for vital signs monitoring,’’ in Proc. Int. Joint Conf. Neural
During 2004 and 2015, he was with the School
Netw. (IJCNN), Jul. 2016, pp. 2827–2833.
[121] K. Lin, F. Xia, W. Wang, D. Tian, and J. Song, ‘‘System design for of Mechanical and Electrical Engineering, Univer-
big data application in emotion-aware healthcare,’’ IEEE Access, vol. 4, sity of Southern Queensland, Toowoomba, Australia. He is currently a Foun-
pp. 6901–6909, 2016. dation Professor and the Head of Discipline Internet of Things Engineering
[122] P. K. Sahoo, S. K. Mohapatra, and S.-L. Wu, ‘‘Analyzing healthcare big with the College of Science and Engineering, James Cook University, Cairns,
data with prediction for future health condition,’’ IEEE Access, vol. 4, Australia. He has published over 200 papers in peer-reviewed journal and
pp. 9786–9799, 2016. conference papers. His research interests are in the broad area of communi-
[123] J. H. Abawajy and M. M. Hassan, ‘‘Federated Internet of Things and cations and information theory, particularly coding, and signal processing
cloud computing pervasive patient health monitoring system,’’ IEEE for multimedia communications systems. He is a Fellow of the IET and
Commun. Mag., vol. 55, no. 1, pp. 48–53, Jan. 2017. an elected Fellow of Engineers Australia. He received the TNQ Innovation
[124] Australian Government. (2017). What is My Health Record? [Online]. Award in 2016. He was a co-recipient of the three Best Paper Awards at the
Available: http://myhealthrecord.gov.au/internet/mhr/publishing.nsf/- 2015 WCSP, the 2011 IEEE WCNC, and the 2009 ICWMC. He has been
Content/find-out awarded several prestigious fellowship titles. He was named a Queensland
[125] Managing Access, Privacy and Security, Australian Government, Sydney, International Fellow by the Queensland Government of Australia from 2010
NSW, Australia, 2017.
to 2011, an Endeavour Research Fellow by the Commonwealth Government
[126] C.-Y. Hung, W.-C. Chen, P.-T. Lai, C.-H. Lin, and C.-C. Lee, ‘‘Com-
of Australia from 2012 to 2013, a Smart Futures Fellow by the Queensland
paring deep neural network and other machine learning algorithms for
stroke prediction in a large-scale population-based electronic medical Government of Australia from 2012 to 2015, and a JSPS Invitational Fellow
claims database,’’ in Proc. 39th Annu. Int. Conf. IEEE Eng. Med. Biol. jointly by the Australian Academy of Science and Japanese Society for Pro-
Soc. (EMBC), Jul. 2017, pp. 3110–3113. motion of Science from 2014 to 2015. He is the Vice Chair of the IEEE North-
[127] J. Park, K.-Y. Kim, and O. Kwon, ‘‘Comparison of machine learning algo- ern Australia Section. He is an Editor for the IEEE COMMUNICATIONS LETTERS,
rithms to predict psychological wellness indices for ubiquitous healthcare and an Associate Editor for the Telecommunications Systems (Springer).
system design,’’ in Proc. Int. Conf. Innov. Design Manuf. (ICIDM),
Aug. 2014, pp. 263–269.
[128] D. Thilakanathan, S. Chen, S. Nepal, and R. Calvo, ‘‘SafeProtect: Con-
trolled data sharing with user-defined policies in cloud-based collabora-
tive environment,’’ IEEE Trans. Emerg. Topics Comput., vol. 4, no. 2,
pp. 301–315, Apr./Jun. 2016.
[129] J. Pacheco, C. Tunc, P. Satam, and S. Hariri, ‘‘Secure and resilient cloud
services for enhanced living environments,’’ IEEE Cloud Comput., vol. 3,
no. 6, pp. 44–52, Nov./Dec. 2016.
[130] S.-D. Bao, M. Chen, and G.-Z. Yang, ‘‘A method of signal scrambling to
secure data storage for healthcare applications,’’ IEEE J. Biomed. Health
Informat., vol. 21, no. 6, pp. 1487–1494, Nov. 2017.
[131] U. Premarathne et al., ‘‘Hybrid cryptographic access control for cloud-
based EHR systems,’’ IEEE Cloud Comput., vol. 3, no. 4, pp. 58–64,
Jul./Aug. 2016.
[132] A. Khedr and G. Gulak, ‘‘SecureMed: Secure medical computation using
GPU-accelerated homomorphic encryption scheme,’’ IEEE J. Biomed.
Health Informat., to be published.
[133] O. Kocabas, T. Soyata, and M. K. Aktas, ‘‘Emerging security mechanisms
for medical cyber physical systems,’’ IEEE/ACM Trans. Comput. Biol.
Bioinf., vol. 13, no. 3, pp. 401–416, May/Jun. 2016.
[134] C. Gentry, S. Halevi, and N. P. Smart, ‘‘Homomorphic evaluation of the
AES circuit,’’ in Proc. 32nd Annu. Cryptol. Conf., Santa Barbara, CA,
USA, 2012, pp. 850–867.
STEPHANIE B. BAKER received the B.Eng.

degree (Hons.) in computer systems engineering
from James Cook University, Townsville, QLD,
Australia, in 2016, where she is currently pursu- IAN ATKINSON is a Professor and the Director of
ing the Ph.D. degree in engineering and related the eResearch Centre, James Cook University. His
technologies. research interests include environmental monitor-
Previously, she was a summer intern with ing, sensors, data management, and supramolec-
Cochlear Ltd., where she was involved in an ular chemistry. He has involved in various
individual research project in biomedical field. professional/corporate IT roles and is a recipient
Her research interests include developing accurate of several major research grants.
biomedical sensors, wireless body area networks, machine learning for medi-
cal diagnostics, and the use of Internet of Things technology for the provision
of rural and emergency healthcare.
26544 VOLUME 5, 2017

Internet of Things For Smart Healthcare: Technologies, Challenges, and Opportunities

Uploaded by

Copyright:

Available Formats

Internet of Things For Smart Healthcare: Technologies, Challenges, and Opportunities

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Internet of Things For Smart Healthcare: Technologies, Challenges, and Opportunities

Uploaded by

Copyright:

Available Formats

Received October 2, 2017, accepted November 6, 2017, date of publication November 29, 2017,

date of current version December 22, 2017.

Internet of Things for Smart Healthcare:

Corresponding author: Stephanie B. Baker ([email protected])

I. INTRODUCTION in [12], with some focus placed on communications.

26522 VOLUME 5, 2017

VOLUME 5, 2017 26523

FIGURE 1. Overview of the proposed model.

26524 VOLUME 5, 2017

VOLUME 5, 2017 26525

Pressure sensors and PPG sensors are combined

26526 VOLUME 5, 2017

VOLUME 5, 2017 26527

blood density [54]. In ideal conditions, where devices had

26528 VOLUME 5, 2017

VOLUME 5, 2017 26529

TABLE 1. Comparison of short-range communication standards.

26530 VOLUME 5, 2017

VOLUME 5, 2017 26531

TABLE 2. Comparison of long-range communication standards.

1) SIGFOX technologies operating in the sub-GHz spectrum, no globally

26532 VOLUME 5, 2017

VOLUME 5, 2017 26533

26534 VOLUME 5, 2017

VOLUME 5, 2017 26535

26536 VOLUME 5, 2017

VOLUME 5, 2017 26537

26538 VOLUME 5, 2017

VOLUME 5, 2017 26539

26540 VOLUME 5, 2017

VOLUME 5, 2017 26541

26542 VOLUME 5, 2017

VOLUME 5, 2017 26543

STEPHANIE B. BAKER received the B.Eng.

26544 VOLUME 5, 2017

You might also like