Journal of Information Security and Applications 78 (2023) 103621

Contents lists available at ScienceDirect

Journal of Information Security and Applications

journal homepage: www.elsevier.com/locate/jisa

Insights into security and privacy issues in smart healthcare systems based on
medical images
Fei Yan a , Nianqiao Li a , Abdullah M. Iliyasu b,c ,∗, Ahmed S. Salama d , Kaoru Hirota c,e
a
School of Computer Science and Technology, Changchun University of Science and Technology, Changchun 130022, China
b
College of Engineering, Prince Sattam Bin Abdulaziz University, Al-Kharj 11942, Saudi Arabia
c
School of Computing, Tokyo Institute of Technology, Yokohama 226-8502, Japan
d
Faculty of Engineering and Technology, Future University in Egypt, Cairo 11835, Egypt
e
School of Automation, Beijing Institute of Technology, Beijing 100081, China

ARTICLE INFO ABSTRACT

Keywords: The advent of the fourth industrial revolution along with developments in other emerging technologies, such as
Smart healthcare security Internet of Things, big data, artificial intelligence as well as cloud and quantum computing, smart healthcare
Medical image analysis systems (SHS) are becoming ubiquitous in our daily lives. Meanwhile, patients, doctors, and other medical
Privacy protection
personnel rely on the safe and efficient storage, transmission, and analysis of medical images and electronic
Artificial intelligence
health records for successful diagnosis, treatment, and management of different ailments. Moreover since, for
Internet of things
Information security
various reasons, medical images are always the target of different illicit criminal activities, studies to utilise
advanced information technologies to safeguard the confidentiality, integrity, and availability of such data
have become a major priority in all SHS platforms. Our study evaluates recent efforts to deploy emerging
technologies to design, secure, and enhance the efficiency of SHS that are based on medical images. It is
hoped that this work will stimulate further interest aimed at the pursuit of more advanced algorithms and
frameworks covering all aspects of security and privacy in emerging and future smart healthcare applications.

1. Introduction
Recent advances in information technology and engineering have
birthed numerous intelligent systems that are used across different
scientific disciplines, professions, entertainment, etc. Smart healthcare
is a new context-aware medical paradigm that integrates classical
sciences, such as medicine and computer science, with emerging tech-
nologies, such as artificial intelligence (AI), deep learning (DL), big
data, Internet of Things (IoT), and cloud computing. Unsurprisingly,
such smart healthcare systems (SHS) are experiencing wide ranging
interests because of their potential to transform traditional face to
face (F2F) expert-centred medicare to patient-centred, distributed, and
remote modes. This entails leveraging on advances in internet and com-
munication infrastructure including big data technologies to provide
telemedical patient-centred care.
An important part of medical and health services is the use of
medical images at different stages of disease diagnosis, treatment, and
management. In the process, medical images are stored, transmitted,
and analysed to support clinicians and other experts in distinct aspects
of patient care. This utility has made medical images and other elec-
tronic health records (EHR) the target of abuse, illicit tampering, and
other criminal activities. Consequently, since medical images are essen-
tial for efficient patient care, efforts to safeguard their confidentiality,
integrity, and availability (CIA) deserve every support [1].
More recently, developments in cloud computing and information
technologies have motivated cloud-based storage and analysis of medi-
cal images. This has encouraged the development of the digital imaging
and communication medicine (DICOM) [2] standards by the national
electrical manufactures association (NEMA) that are becoming the
default frameworks for medical images and are considered as the most
important data points in SHS. Nowadays, the DICOM standard is used
to store and transmit medical images thereby further supporting the
integration of medical imaging devices with picture archiving and
communication systems (PACS) provided by different manufacturers. A
typical DICOM file includes format definition and network communi-
cation protocols required for TCP/IP communication between systems
with the aim of guaranteeing safe and efficient transmission of medical
images between two or more entities.
To provide better, safer, and more efficient services for patients,
most hospitals have developed different medical information manage-
ment systems. Recently, as part of daily clinical practices, medical

∗ Corresponding author at: College of Engineering, Prince Sattam Bin Abdulaziz University, Al-Kharj 11942, Saudi Arabia.
E-mail addresses: yanfei@cust.edu.cn (F. Yan), nianqiaoli@yeah.net (N. Li), a.iliyasu@psau.edu.sa (A.M. Iliyasu), a.salama@fue.edu.eg (A.S. Salama),
hirota@bit.edu.cn (K. Hirota).

https://doi.org/10.1016/j.jisa.2023.103621

Available online 11 October 2023

2214-2126/© 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
F. Yan et al.
Fig. 1. Organisational outline of this review. Full names of abbreviations in the figure can be found in the text.
research, and education, many medical image data have been captured
and stored in digital formats for use in future diagnosis, research,
and long-distance transmission. However, the advances recorded in
technologies for digital multimedia processing have propelled illicit
efforts aimed at the illegal acquisition and modification of medical
images. Meanwhile, the use of IoT in different stages of healthcare
to collect, store, transmit and process medical image data is on the
increase. This makes any type of vulnerability for such platforms very
decisive since the most sensitive information of patients will be exposed
to criminals. Consequently, safeguarding the CIA of medical images
alongside other EHRs is an important aspect of smart healthcare in-
frastructure. Moreover, due to their sensitive nature, even miniscule
alterations to the content of a medical image could have implications
on the diagnosis, treatment, and management of a disease.
In terms of securing medical images, various combinations of tradi-
tional image security technologies, such as encryption, watermarking
and steganography are explored together with modern information
technology paradigms for smart medical image sharing. Meanwhile,
in the closely related area of medical image analysis, attention is
focused on the process of obtaining diagnostic results of patients by
interpreting the information in medical images. In addition to the
lengthy period that it takes to train and specialise in radiology, as with
most human endeavours, it is error-prone, time consuming, and costly.
Moreover, errors in radiological interpretations could be fatal. In re-
sponse, computer-aided diagnostic (CAD) utilises computer technology
to analyse medical data such as medical images, providing assistance
to medical professionals in diagnosing diseases by offering information
and results [3]. As a specific application of CAD technology, CAD
systems improve the accuracy and efficiency of medical diagnosis.
Meanwhile, AI, particularly deep learning models (DLM), has been
revolutionary across numerous domains and applications. Its potency
has been exploited in technologies for disease detection, diagnoses,
and treatment [4]. Integrating cloud computing into DLM-based med-
ical image analysis offers immense capabilities in smart healthcare,
telemedicine, etc. However, despite the potential benefits of applying
cloud computing and AI to SHS, the complex communication networks
result in an exponential growth of sensitive medical data collected
by medical IoT devices, which are highly vulnerable to exposure [5].
Additionally, Szegedy et al.’s research demonstrates that DLMs are ex-
tremely fragile when facing carefully crafted adversarial examples [6].
Therefore, safeguarding the integrity and security of sensitive medical
data [7], particularly medical images, emanating from cloud computing
and AI is crucial in SHS systems. As noted earlier, this impedes the
CIA properties of medical images and patient data, increases chances
of misdiagnosis, and so on. Consequently, advanced infrastructure to
safeguard these SHS are necessary.
Patients, doctors, and other medical personal rely on the safe and
efficient storage, transmission, and analysis of EHRs, particularly med-
ical images, for successful diagnosis, treatment, and management of
different ailments. The integrity of these documents is necessary be-
cause when modified, altered, or corrupted medical data can lead to
2
Journal of Information Security and Applications 78 (2023) 103621
incorrect diagnoses and create serious health issues for individuals.
ML/DL models in CAD systems and medical images as training data are
similarly susceptible to security and privacy issues, with three forms of
adversarial attacks posing the greatest threat. This survey is primarily
aimed at providing readers with an apercu highlighting the earlier
enumerated advances. In doing so, we present a concise yet compre-
hensive overview of recent studies in medical image security. This is
motivated by the prominent place of medical imaging technologies in
state-of-the-art SHS. Moreover, we note that this area has not been
extensively covered in terms of their role in securing SHS. Furthermore,
our review probes into the role of secure medical image sharing on
safeguarding SHS as well as strategies for protecting ML/DL models
in CAD systems and medical image training data for efficient medical
image analysis, especially in the context of emerging technologies. This
copious background provides us with the perspicacity to ruminate over
potential technologies and infrastructure that will shape future SHS.
Other contributions of our review include:
• Insights into architectures for smart healthcare security, including
the security required for medical-image sharing and CAD security.
• Overview on the roles of recent intelligent technologies, such as
artificial intelligence, Cloud of Things, and quantum computing
improving security issues in medical image based SHS.
• Analysis of types and risks to adversarial attacks that ML/DL mod-
els and medical image training data in CAD systems for medical
image analysis are susceptible to and strategies to mitigate them.
• Discussions on immense potential of many emerging technologies
for safe, robust, and efficient SHS. With this, several insights into
the current problems and future directions for SHS technology are
also enumerated.
We start by highlighting differences and similarities between our
study and similar ones. Table 1 enumerates literature on smart health-
care security for the period from 2018 to 2022 (i.e., recent 5 years).
We especially note that unlike the majority of the surveyed literature,
our review focuses on exploring security technologies based on medical
images (as opposed to general medical data) in SHSs. In addition, our
review is not limited to any particular type of security technology; this
study discusses not only smart security technologies based on cloud,
IoT, and other technologies, but also security strategies for ML/DL
models and prospects for the application of emerging technologies in
SHS security.
In order to deliver the outlined objectives of our study, the re-
mainder of the paper consists of an overview of the motivations for
our study in terms of the significance of medical images in SHS as
well as their architecture as introduced in Section 2. Following that,
in Section 3, we highlight recent secure SHS protocols that are based
on medical image sharing. Section 4 focuses on the prominent tasks in
medical image analysis as well as the challenges and potential solutions
offered by machine learning and deep learning (ML/DL) tools with
regards to the security of CAD systems. Section 5 discusses emerging
F. Yan et al.
Table 1
Overview of differences between our review and similar literatures.
Reference Title
Nguyen et al. [8] Federated learning for smart healthcare: A
survey
Quy et al. [9] Smart healthcare IoT applications based on fog
computing: Architecture, applications and
challenges
Ullah et al. [10] Secure healthcare data aggregation and
transmission in IoT-A survey
Hathaliya et al. [11] An exhaustive survey on security and privacy
issues in healthcare 4.0
Tariq et al. [12] Blockchain and smart healthcare security: A
survey
Panchatcharam et al. [13] IoT in healthcare-smart health and surveillance,
architectures, security analysis and data
transfer: A review
Algarni et al. [14] A survey and classification of security and
privacy research in smart healthcare systems
Masood et al. [15] Towards smart healthcare: Patient data privacy
and security in sensor-cloud infrastructure
He et al. [16] Privacy in the Internet of Things for smart
healthcare
technologies for smart healthcare security based on medical images.
Finally, in Section 6, we reflect on the advances highlighted in pre-
ceding sections and offer a few insights and concluding remarks. This
outline is further presented in Fig. 1 and it is our hope that the review
will invigorate efforts towards the pursuit of more advanced algorithms
and frameworks for security and privacy issues in medical image based
smart healthcare applications.
2. Architecture and requirements for smart healthcare security
based on medical images
Considering its important role as a medical data carrier, this sec-
tion explores the architecture and requirements of intelligent medical
security based on medical images. Today, medical imaging entails the
use of advanced equipment, techniques, and processes to probe the
interior parts of a body for clinical analysis, medical intervention, and
visual access to otherwise inhibited organs and tissues. Depending on
the equipment and methods used, medical images can be categorised
as one of the following types.
• Radiography. Radiography is an imaging technique that uses X-
rays, gamma rays or similar ionising and nonionising radiation to
view the internal details of an object. Applications of radiography
include medical radiography (diagnostic and therapeutic) and
industrial radiography. Advanced techniques such as backscatter
X-ray are used in airport security to scan the body for dangerous
objects.
• Computed Tomography (CT). CT is a medical imaging tech-
nique that uses computer-processed combinations of multiple
X-ray measurements taken from different angles to produce cross-
sectional tomographic images (i.e., virtual ‘‘slices’’) of a body,
thus allowing the user to see inside the body without cutting any
part of it.
• Magnetic Resonance Imaging (MRI). MRI is a medical imaging
technique used in radiology to unravel the anatomy and the
physiological processes of the body. MRI scanners use strong
magnetic fields, magnetic field gradients, and radio waves to
generate images of the organs in the body.
• Medical Ultrasound. Medical ultrasound is a diagnostic imaging
technique or therapeutic application used to create an image of
internal body structures such as tendons, muscles, joints, blood
vessels, and internal organs. Its aim is often to find the source of
a disease or to exclude pathology.
Journal of Information Security and Applications 78 (2023) 103621
Year Research focus Contributions
2022 Federated learning for SHS. Discusses the recent advances and security
solutions of federated learning applied in SHS.
2021 Fog computing for SHS. Investigates the applications and future research
directions of IoT-based fog computing in SHS.
2021 Medical data transmission Investigates security challenges in aggregating
security and transmitting medical data to fog servers.
2020 Medical data security in Explores advantages and limitations of various
Healthcare 4.0. security and privacy techniques in healthcare
4.0.
2020 Medical data security Highlights blockchain technology for security
based on blockchain. solutions to invulnerable SHS.
2019 Medical data security Summarises security frameworks for
based on IoT. administration and application of IoT in smart
healthcare services.
2019 Medical data transmission Investigates important security attacks and
security. countermeasures in smart healthcare security.
2018 Medical data security in Discusses security technologies used to protect
sensor-cloud infrastructure. the medical data in sensor-cloud infrastructure.
2018 Medical data security Investigates security threats in healthcare IoT
based on IoT. frameworks and cryptographic vulnerabilities.
Fig. 2. Security mechanisms applied in the field of smart healthcare security based on
medical images.
Other medical imaging methods include elastography, photoacoustic
imaging, magnetic particle imaging and so on [17].
As important carriers of medical information, digital medical images
are endowed with the function of information sharing. However, in the
process of medical image transmission, eavesdropping and tampering
that result in unauthorised access to patients’ private information or
alterations impugning the quality of the received medical image need
to be avoided. At the same time, another important role of medical
images arises during medical image analysis in diagnostics, which must
be secured to safeguard the trio of CIA of the medical image contents.
The security mechanisms applied in the field of smart healthcare
security based on medical images are presented in Fig. 2. From the
figure, it can be seen that the process of securing an SHS platform is
divided into medical image sharing security (MISS) and CAD security
(CADS). The requirements of the MISS architecture are based on the
medical image itself as well as robustness of the IoT and cloud platform
to support medical image transmission. Specific security requirements
in MISS include:
• Confidentiality. This is the set of rules that limit access to
information. In the medical image scenario of SHS, confidentiality
could be realised via securing nodes that include services, devices,
managements systems, etc. from unauthorised access.
3
F. Yan et al.
• Integrity. Integrity is the assurance that information is trustwor-
thy and accurate. In SHS, integrity is validated by proof that the
images and other EHRs are untampered. Integrity is an important
feature in securing SHS since, as observed earlier, any tampering
potentially results in misdiagnosis and other avoidable mishaps.
• Availability (or reliability). This security feature is aimed at
guaranteeing reliable access to information. In the context of
IoT and cloud platforms, this entails improving the reliability of
infrastructure, such as power, internet, etc. required for secure
and efficient information sharing. This conforms with standard
definitions that availability is a warranty that information will
be available to authenticated users whenever requested, even in
times of the systems’ failures.
• Privacy. As a sub-concept of confidentiality, privacy is more fo-
cused on the protection of patients’ personal information. Medical
images are usually annotated with patient identifiers and history.
Therefore, in the hands of unauthorised persons, access to such
images leads to leakage and misuse of such data. A secure SHS
should have safeguards to forestall this from happening.
On the other hand, the SHS platform requires secure intelligent
diagnosis based on medical images, and the diagnostic tasks related to
medical image analysis are at the core of CAD systems. Noting this, our
review of medical image analysis technologies is divided into four parts:
medical image retrieval, classification, segmentation, and registration.
The envisioned security requirements of CAD systems ensure that the
confidentiality of medical information will not be violated nor will it
impugn the integrity of the ML/DL techniques employed in medical
image analysis so that accurate diagnosis is guaranteed. Therefore, this
treatise summarises six security strategies for CAD system as outlined
in Fig. 2.
Meanwhile, since using medical image sharing to secure SHS re-
quires the combination of classical image security and modern infor-
mation technologies, recent advances in these areas are reviewed in the
next section. Similarly, the challenges and solutions for using ML/DL
tools to secure CAD systems are presented in Section 4.
3. Classical and smart technologies for MISS
In conforming with the CIA principles of information security,
various security techniques are employed. For example, in applications
where premium is attached to confining information access to selected
people, confidentiality takes centre stage. In such a scenario, from
the viewpoint of users, confidentiality is focused on authenticating
the credentials of users to ensure they are authorised to access such
content. Consequently, in this sense, [1] argues that confidentiality
loosely equates to privacy. This section will explore MISS from the
perspective of classical and smart security strategies for medical image
sharing as well as the security of IoT and cloud platforms that facilitate
medical image transmission.
3.1. Classical security for medical image sharing
• Digital watermarking. Digital watermarking is the process of
embedding a message onto a host (or carrier) media, which can
be an image, video, audio, or other multimedia. In good water-
marking strategies, the messages are well concealed, difficult to
retrieve, and impossible to modify. As presented in Fig. 3, in
medical image watermarking (MIW), a common strategy is to
embed a fragile watermark in the spatial domain and a robust
one in the frequency domain. The former serves to verify the
integrity of the image whilst the latter is used to establish the
authenticity of the image. Furthermore, to safeguard the contents
of the image, the watermarks are embedded in regions of no
or least interest, often referred to as RONI. At the receiver end,
the extracted and original watermarks as well as their evaluation
4
Journal of Information Security and Applications 78 (2023) 103621
Fig. 3. Illustration of embedding, extraction, and evaluation of digital watermarking
of medical images.
parameters are compared. This serves to establish if the image
is genuine (authenticity), and that it has not been compromised
(integrity) [18].
• Digital steganography. Digital steganography is the art and
science of imperceptibly hiding information in a host (or carrier)
media, typically, an image audio, video, etc. Unlike watermark-
ing, the primary objective of steganography is to conceal the
existence of the hidden information. In medical image steganog-
raphy (MIS), safeguarding patient details (such as name, ID,
and other EHR) is accorded top priority because access to such
information is restricted to doctors only, while other staff are
granted access on a need-to-know basis.
• Encryption. Medical image encryption (MIE) provides end-to-end
protection for images from storage to transmission. Therefore, en-
cryption is the process of scrambling data so that only authorised
parties can understand the information therein. MIE converts
the pristine images into a ciphertext that is decodable using a
secret key whose access is confined to those with the required
privileges. Good encryption schemes prevent access to such media
by unauthorised third parties.
In a recent MIW scheme, Aparna et al. proposed an efficient strategy
for use in E-healthcare applications using biometrics (specifically, fin-
gerprints) for identity authentication [19]. Specifically, they separated
the regions of interest (ROI) of the medical image using the region
growing technique and encrypted it using the SHA-265 algorithm.
Additionally, the elliptic curve cryptography algorithm is utilised to
encrypt EHR data. After the two encryption methods above, the minu-
tiae point binary information extracted from the fingerprint image is
combined with the medical image information and EHR information
to generate a binary data stream, which is then embedded within the
carrier image. In their contribution, Alshanbari et al. [20] embedded
multiple watermarks in a host medical image and used insertion of
principal component (PC) to forestall ownership attacks, and, using the
Lempel–Ziv-Welch (LZW) fragile watermarking technique, intentional
tampering attacks targeting ROI in an image can also be prevented.
Recently, Yan et al. proposed a multi-watermarking technique [21] that
verifies the authenticity and integrity of medical images. In contrast
to [19], they manually segment medical images, embed the watermark
logo encrypted using a quantum random walk scheme confined within
ROI, and subsequently embed medical text data within the RONI.
Concurrently, the brain storm optimisation (BSO) process is utilised to
balance the watermark’s imperceptibility and robustness. The SHA-256
algorithm that is used to encrypt key information in studies [19,20]
is a standard hash function algorithm issued by the National Institute
of Standards and Technology (NIST) in 2001. It generates a 256-bit
hash value for messages of any length, and this hash value is extremely
sensitive to the original message, thereby providing a high level of
security. With the advent of the quantum information era, studies such
F. Yan et al. Journal of Information Security and Applications 78 (2023) 103621

as [21,22], etc. focus on quantum-based medical image encryption ap-

proaches to overcome the hidden risk that classical encryption methods
will be rendered insecure by quantum computing resources.
Other recent contributions in MIS include [23], wherein Jeevitha
et al. used a Hidden Markov Tree (HMT) contourlet transform to
embed a secret message onto a medical carrier image. This method
uses a Particle Swarm Optimisation (PSO) algorithm to select the
optimal embedding position for the secret message, so that the carrier
image has a higher payload capacity. The PSO algorithm relies on
a group of random solutions whose optimal solution is obtained by
iteration, and its evaluation is based on its fitness. In 2004, Sun et al. Fig. 4. Smart security mechanisms applied in the field of medical image sharing.
developed a quantum version of PSO, the quantum particle swarm
optimisation (QPSO) [24], which has superior global convergence and
search capabilities. This is attributed to its quantum property that
strong correlation in such images. For instance, a genetic algorithm
permits particles to be searched throughout the entire possible solution
(GA) is a metaheuristic that is often used to generate high-quality
space. In addition to the BSO algorithm employed in [21], QPSO
solutions to optimisation and search problems by employing biologi-
is a potential method for balancing a watermark’s invisibility and
cally inspired operators such as mutation, crossover, and selection. As
robustness in actual applications [25]. More recently, Stoyanov et al.
far back as 2005, in [29], Shih et al. proposed the use of a GA to
proposed an MIS scheme that they named ‘‘BOOST’’ in [26], which uses
embed watermarks in the frequency domain of RONI areas of medical
a pseudorandom byte generation algorithm that is based on the nuclear
images. In this method, the GA is used to reduce rounding errors in
spin function. They reported excellent peak signal-to-noise ratio, strong
the conversion process between the spatial and frequency domains.
collision resistance, and desirable security properties that can withstand
Similarly, in [30], Pareek et al. proposed the use of GA in greyscale MIE
most of the common theoretical and statistical attacks.
encryption. This method uses the principle of crossover and mutation
Strategies for efficient MIE could be designed based on protocols
in GAs to improve the robustness of the MIE strategy so that medical
targeting content in the spatial or frequency domain of the image.
images can be secured during real-time transmission. The differential
In spatial domain MIE, pixel level details focused on either or both
evolution (DE) algorithm, which is an improved GA, was used by Lei
location and colour details of the image are used to modify the images.
et al. to control the strength of the embedded watermark in [31]. In
In [27], Nazeer et al. proposed an MIE method based on Fresnel
contrast to the binary encoding method utilised by GA, DE employs real
transform and Arnold scrambling, followed by the use of digital holog-
number encoding. Therefore, population iteration does not require the
raphy to quickly reconstruct the medical image. They claimed that this
parent generation to generate offspring, but rather relies on the parent
encryption method can preserve the true resolution of medical imaging
generation’s self-evolution and extinction. Consequently, DE is superior
systems and effectively improve the assessment of size and morphology
to GA in terms of robustness, convergence speed, and search capability
of organs. Frequency domain MIE use DCT (Discrete Cosine Transform),
for global optimisation.
DWT (Discrete Wavelet Transform) and DFT (Discrete Fourier Trans-
Meanwhile, as described in Section 3.1, recent MIW techniques are
form) algorithms to encrypt medical images. In [28], Zhang et al. used
focused on embedding watermarks in RONI areas of medical images.
DWT and DCT in a homomorphic encryption algorithm. Furthermore,
For example, Rai et al. proposed an MIW method that can be applied
since the frequency domain of a medical image is considered stable,
they claim that this approach leads to high resistance to interference, in LTE (long-term evolution) networks [32]. To eliminate errors in the
and because of the characteristics of homomorphic encryption algo- watermark embedding, they used a support vector machine (SVM) to
rithm, the method could also be used for medical image retrieval, classify ROI and RONI of images. SVM is a supervised learning model
similarity comparison and other operations without decryption. used to analyse data for classification and regression. It maps each
example in a particular training set to a point in space such that the
3.2. Smart security technology of medical image sharing distance between the two categories is maximised. Subsequently, new
examples are mapped to the same space with anticipation of which side
Smart security technologies for medical image sharing leverage the of the gap they fall on. In a similar application in [33], a trained SVM
effectiveness of medical image sharing and the potency of AI, IoT, cloud model was used to classify medical image blocks (natural class and
computing and big data. In this section, we review such technologies encrypted class) and then the correct decrypted version of the image
from the perspective of modern information technology as mapped in block were identified in order to decrypt the encrypted medical image.
Fig. 4. For their part, Ma et al. utilised ML to develop a reversible data
hiding strategy for medical images in [34]. Their method uses inte-
3.2.1. Machine learning algorithms for medical image security ger wavelet transform to convert the image to the frequency domain
AI encompasses numerous technologies that imitate cognitive and and then embed the secret image in medium frequency sub-bands.
intelligent behaviours of humans. In machine learning (ML), algorithms Subsequently, neural networks are employed to optimise the choice
to model complex characteristics and patterns of empirical data are of embedding coefficients, which the authors claim leads to improved
understood and exploited across wide ranging applications. Within ML, embedding capacity.
neural networks, and their many variants (convolutional, deep, recur-
rent, etc.) have proven revolutionary in the landscape of intelligent 3.2.2. Security of medical image sharing based on cloud of things
systems engineering [4]. In particular, the immense adeptness of con- Cloud of Things (CoT) refers to the complementary integration of
volutional neural networks (CNN) is harnessed in the transformational cloud computing technology into IoT infrastructure which is credited
DL paradigm whose utility and potency have been exploited in almost to Aazam et al. in [35]. Cloud computing and IoT are new information
every field of science and engineering and new emerging applications technologies born as part of the march towards the fourth industrial
in almost every other area of our lives. Furthermore, ML and DLMs revolution (4iR or Industry 4.0). Separately, whereas cloud computing
have emerged as veritable tools to enhance and improve technologies provides users with improvement in computing, storage, and network
in disease prognosis, diagnosis, treatment, and cure. resources on the Internet, it does not require active management of
In medical imaging, ML and particularly DLM have proven lifesav- users and is oftentimes available in the form of a central server or as
ing in their use to circumvent issues of data capacity, redundancy, and the data centre of the Internet. In contrast, IoT technology refers to the

5
F. Yan et al. Journal of Information Security and Applications 78 (2023) 103621

interconnection of embedding sensors, software, and other technologies

as nodes or ‘‘things’’ on the Internet as platforms for exchanging data
with other devices and systems through the Internet.
The schematic diagram in Fig. 5 illustrates the concept of CoT.
As the data centre of the Internet, cloud computing servers exchange
information with user terminals through Wi-Fi, 5G and other commu-
nication channels. At the same time, the cloud computing server also
serves as the hub for massive data processing for many IoTs.
As pointed out earlier, modern smart healthcare incorporates AI as
its core technology with CoT as the base of its architecture. The use
of CoT for the storage, transmission, and analysis of medical images
provides effective platforms for remote diagnostics as well as sharing
Fig. 5. Schematic illustration of a typical CoT network.
and scheduling of resources. For example, using the vital signs’ sensor
in an intelligent IoT-ready medical facility, the health status of patients
can be monitored remotely. Doctors and medical staff can monitor and
easily oversee the whole medical process remotely using large computer
monitors or even video clips on smartphone Apps.
Studies on MISS are often based on CoT (that have integrated cloud
computing and IoT), including those focused on securing medical im-
ages that are stored on cloud servers and transmitted on IoT platforms.
Some of the available studies analyse the security of medical images
in the framework of IoT from the perspective of digital watermark-
ing [36], digital steganography [37] and encryption [38]. Further,
in [39], using IoT, Rajagopalan et al. proposed a secure medical image
transmission scheme and developed a graphical user interface (GUI) for
it. They reported that this scheme is suitable for establishing a DICOM
medical image communication system in hospitals. Fig. 6. MISS components in a typical CoT network.
Meanwhile, from the CoT architecture perspective, security is ac-
corded top priority because a cloud computing server is not only
responsible for the storage and analysis of medical images, but it
sensors is large, the frequency of information transmission to the cloud
also controls access in terms of uploading and downloading medical
server increases, and the cloud-only medical architectures, including
images. However, MISS technology are found to mainly focus on the
CloudSec, will exhibit higher power consumption and associated costs.
frameworks of cloud computing environments. For instance, in [40],
This is counterproductive when real-time emergencies (such as sporadic
Ukis et al. built a cloud framework for advanced medical image visu-
monitoring of patients’ vital signs [46] or sudden falls by elderly
alisation that they claim encrypts and anonymises DICOM data, and
people [47]) where swift response times from medical professionals is
strictly controls the access to uploaded data to guarantee the security
required. These tasks require lower latency and higher computational
of patient health information. In [41], a service-oriented medical image
efficiency to allow for the timely arrival of emergency medical help.
processing cloud framework to ensure the security and stability of the
service by using the configuration files that perform normal security Meanwhile, the purple nodes in Fig. 5, depict network edges where
protocols or self-defined principles was implemented by Chiang et al. data can be processed (i.e., edge computing [48]) which have been
The multi-could architecture [42] is one of the most common cloud found to decrease network latency and improve energy efficiency. Com-
computing frameworks. Therefore, as illustrated in Fig. 6, MISS com- pared to cloud computing, edge computing offers location awareness
ponents in a typical CoT network would have a CloudSec core to and improved security. In secure medical image sharing, edge comput-
which clients upload their images using Secure Sockets Layer (SSL) ing can be used to process, transmit, and monitor real-time medical
connection. In such a scenario, the image is divided into several parts data such as electrocardiographs (ECG), electroencephalographs (EEG),
and sent to different clouds (in a multi-cloud architecture). The distinct electromyography (EMG), etc. [45]. In this regard, in [49], Alshehri
parts of the images are subsequently processed and results sent back to et al. assessed security of edge computing using real-time medical
the CloudSec. Since authorised clients can obtain simultaneous access image monitoring. Meanwhile, whereas deep learning necessitates a
to the different clouds, then it is possible for them to recover the large number of datasets and powerful computing resources, edge
complete image. In addition to securing different stages of processing computing devices are only provided with limited computing resources.
and transmitting the images, the outlined procedure also increases Deep migration learning can efficiently utilise the knowledge of the
the efficiency of the system since information is processed in parallel. learned model, since it requires fewer computing resources to retrain
Furthermore, using ML techniques, such as k-means clustering [43] and refine the current model for the current task. In [50], Susian
and GAs [44], the process of segmenting the image into multiple parts et al. integrated edge computing into deep migration learning as an
could be further optimised. Among the methods, k-means clustering architecture for secure and real-time monitoring of COVID-19 cases.
uses Euclidean distance to distinguish the object of interest from the In the evaluation metrics of the above medical image security shar-
background area. In that manner, [43] decomposes the medical image ing studies, the imperceptibility of watermarking and steganography
into a predefined number of parts using k-means clustering so that methods can be measured using peak signal-to-noise ratio (PSNR) [26],
various sections can be processed on separate nodes to decrease the structural similarity index measure (SSIM) and mean square error
danger of sensitive data leakage. Similarly, [44] employs a genetic (MSE) [27]. The embedding capacity is often measured in bits per
method to establish the appropriate entropy threshold, which is then pixel (bpp) values [26]. Normalised cross correlation (NCC) [23] is
utilised to calculate the size of segmentation region of the medical frequently used to evaluate the robustness of medical images to var-
image. That study demonstrated the significant impact of segmentation ious attacks, such as rotation, cropping, histogram equalisation, salt
on the quality of distributed data encryption. & pepper, blurring, and resizing. Typically, the encryption effects of
While the paradigm of IoT is generally considered as a novel ap- medical images are measured by the correlation coefficient of adjacent
proach, it has been reported in [45] that when data generated by IoT pixels [27] and the information entropy [30].

6
F. Yan et al.
4. Medical image analysis and CADS
As outlined earlier in our review, secure medical image sharing
allows users to send medical images to numerous locations and re-
cipients without fear of impugning their confidentiality, integrity, and
authenticity. Additionally, considering that in modern medical image
analysis, the security of CAD systems cannot be overlooked. This sec-
tion highlights some prominent tasks in medical image analysis as well
as the challenges and solutions offered by ML/DL in securing CAD
systems.
4.1. Prominent tasks of medical image analysis and security challenges of
CAD systems
Medical image analysis is aimed at helping doctors in providing
efficient diagnosis and prognosis of diseases [3]. In this subsection, we
present an overview of some prominent tasks in medical image analysis,
and the ML/DL technologies widely used to support such tasks.
• Medical image retrieval. PACS can realise the filmless of medi-
cal institutions, i.e., generate digital medical images that can be
shared through the network to improve the efficiency of medical
operations. However, in order to process these massive medical
image data effectively, an automatic medical image classification
and retrieval system is needed to help doctors make decisions
on the prognosis and diagnosis of diseases. By using the disease
stages from similar cases retrieved by the medical image retrieval
system, doctors can make timely and accurate diagnosis of dis-
eases as well as get the needed help to determine the stages of a
disease by comparing with retrieved cases.
• Medical image classification. Medical image classification is an
important application in CAD systems but its execution is also a
big challenge in image analysis. According to the selected image
classification features, medical image classification technology is
used to classify images in a medical image dataset. The results
can then be used to confirm doctors’ initial diagnosis and ideas
and can also be used to develop CAD systems based on medical
image data.
• Medical image segmentation. Medical image segmentation is
aimed at dividing medical images into several non-overlapping re-
gions for delineation into meaningful and background regions. By
focusing the analysis and available resources on the meaningful
information, the search area in the image can be reduced, which
impacts on the efficiency of medical image analysis. Common ap-
proaches employed in medical image segmentation include those
focused on the shape, volume, relative position, and abnormal
conditions of organs.
• Medical image registration. Medical image registration is an
important pre-processing step used in medical image analysis
to enhance analysis, prognosis, and diagnosis. Multiple medical
images of the same organ will produce slight dislocation due
to different imaging conditions during their image acquisition
which will affect the efficiency of monitoring changes in le-
sions. However, using medical image registration, two images can
be accurately aligned. In other applications, image registration
could be used to measure similarity between a moving and fixed
image where an optimisation algorithm constantly updates the
parameters required to maximise similarity [51].
Computer-aided diagnosis has long been an active area of study
in medical image analysis [52]. Since the early 1970s, developments
in CAD systems have attempted to improve the accuracy of medical
image analysis by providing ‘‘auxiliary opinions’’ to doctors [53], which
leads to reduced workload for doctors and medical costs for patients. By
exploiting the potency of ML/DL techniques, the efficiency and accu-
racy of an array of tasks in medical image analysis including retrieval,
7
Journal of Information Security and Applications 78 (2023) 103621
classification, segmentation, and registration have been improved. For
example, in [54], Cao et al. used the deep Boltzmann machine based
multimodal learning model to understand a joint probability density
model from the visual and text information of multimodal medical
images required to bridge the semantic gap in medical image retrieval.
In [55], a CNN classifier was used to classify chest X-rays according
to the direction of lung in the chest image. In [56], Milletari et al.
used 3-dimensional (3D) CNN to enhance segmentation of prostate MRI
volumes. Similarly, in [57], Haskins et al. proposed a DLM to register
transrectal ultrasound (TRUS) and MRI images. They further designed
a CNN with skip connection and trained it to learn target registration
error (TRE) between the two images as the similarity evaluation score.
Despite the successful application of ML/DL models in medical
image based disease diagnosis, the ML/DL systems are also susceptible
to challenges related to security and privacy. On the one hand, the ML
systems must prevent unauthorised users from accessing confidential
information, i.e., when the medical images are used as training data
in the CAD systems which involves safeguarding the CIA of contents
of the medical image as mentioned earlier in Section 2. On the other
hand, the discovery of the so-called ‘‘adversarial examples’’ has exposed
the vulnerability of most ML techniques. In their study, Szegedy et al.
demonstrated the sensitivity of modern ML/DL models to adversarial
attacks in the form of slightly perturbed images resembling original
images [6]. In such vulnerable scenarios, attackers could gain access
to patients’ examination reports, which have been attributed among
causes of insurance fraud and misdiagnosis by clinicians.
In adversarial attacks [7], the target of hackers is to generate
adversarial examples by adding imperceptibly small disturbances to the
original input samples so as to evade the integrity of ML/DL techniques.
In general, there are three types of adversarial attacks.
• Evasion attacks. In this type of attack, criminals modify the
samples to evade detection so that the malicious data is passed on
as legitimate or spoofing attacks targeting healthcare biometric
systems [58]. Evasion attacks are particularly challenging for
ML/DL algorithms in CAD systems, because there are intelligent
and adaptive adversaries who can carefully manipulate the in-
put medical image data to circumvent the performance of the
detection system and violate the medical image data [59].
• Poisoning attacks. These attacks manipulate training data and
inject malicious samples to mislead the learning process of the
ML/DL model, thus disrupting retraining efforts [60]. Compared
with direct modifications to the original training datasets in CAD
systems, attackers prefer to add carefully crafted malicious sam-
ples to the medical image datasets [7]. This class of poisoning
attacks are difficult to detect and will lead to wrong diagnoses
by CAD systems, thereby delaying and comprising treatment. For
example, hinderance of a hypothyroid diagnosis may have life-
threatening consequences [61] or a false positive classification of
COVID-19 could cause unnecessary panic [62].
• Model stealing. In this type of attack, criminals reconstruct
the confidential model or they extract sensitive training data by
probing the black-box ML/DL system [63]. For example, in 2020,
Zhang et al. cloned a CNN model for predicting lung cancer that is
complicated with pulmonary embolism using only a small number
of labelled samples by the target CNN [64]. The success of the
attack indicates that the trained ML/DL models in the CAD system
as well as the patients’ private data are at risk.
4.2. Security defence strategies for CAD systems
In this section, we outline some security defence strategies to safe-
guard CAD systems and ensure accuracy of medical image based di-
agnosis and privacy of patients’ training datasets. Table 2 summarises
some of the most advanced defence strategies that safeguard medical
images against adversarial attacks.
F. Yan et al. Journal of Information Security and Applications 78 (2023) 103621

Table 2
Summary of security defence strategies for CAD systems.
Reference Security Methodology Applications Advantages Environment & parameters
strategy
Ohrimenko et al. [65] Encryption Controlling access to Implementing Significantly lower Quad-core Intel Skylake processor, 8
strategy ‘‘enclave’’ memory to privacy-preserving overhead than common GB RAM, 256 GB solid state drive,
protect ML/DL tasks multi-party machine cryptographic approaches and Windows 10 enterprise
learning
Ziller et al. [66] Differential Protecting privacy by using Protecting medical image Better compatibility and GNU Linux on the 5.4.0-72 kernel,
privacy the 𝑑𝑒𝑒𝑝𝑒𝑒 framework classification and rigorous privacy guarantees macOS 11.2, Backbone: VGG-11, and
segmentation tasks 𝛿-parameter: 10−5
Sun et al. [67] Differential Differential privacy, Preventing attackers from Improving the availability Platform: TensorFlow and PyTorch,
privacy decision tree, and a weight invading personal privacy of medical data and activation function: ReLU, and
calculation system with the training data ensuring the accuracy of normalisation parameters: batch size
the ML model = 20, dropout = 0.4, and learning
rate = 0.01
Kumar et al. [68] Federated Training a global model Recognising lung CT Protecting multi-party Dataset: CC-19 (self-made), case
learning using blockchain-based images of COVID-19 medical data sharing number: 182 Training/45 Testing
federated learning patients among hospitals (patients per hospital)
Sui et al. [69] Federated Training the central model Preventing privacy Decoupling ML model NVIDIA GeForce RTX 2080 Ti, BERT
learning utilising ensemble local disclosure in the process of training from privacy model parameters: batch size = 4,
models medical text sharing protection dropout = 0.1, epochs = 2 and
learning rate = 2.5𝑒−5
Celik et al. [70] Knowledge Privacy distillation method Predicting personalised Allowing patients to Dataset: IWPC [71], 1221 training
distillation using DNN warfarin dose participate in privacy and 656 validation samples
protection
Li et al. [72] Adversarial Unsupervised learning Detecting adversarial Easy embedding onto White-box Attack:
attack approach attacks on X-ray datasets DLM-based CAD systems classifier: DenseNet-121, maximum
detection perturbation: 0.02 Black-box Attack:
classifier: ResNet-50, maximum
perturbation: 0.05
Watson et al. [73] Adversarial Explainable ML detection Detecting adversarial Easy to gain the trust of Train/Test split: 80/20, optimiser:
attack technique samples on EHRs and chest medical institutions Adam, and learning rate = 0.01
detection X-ray datasets
Lal et al. [74] Adversarial Adversarial training-based Protecting retinal fundus Fewer parameters and Dataset: 1000 instances from Kaggle,
training defencive model against image datasets higher robustness augment the dataset to 6497 images
adverse disruptions and resize them to 224 × 224 × 3
Xu et al. [75] Adversarial Misclassification-aware and Performing the single-label Improving the robustness GAP attack:
training multi-perturbations and multi-label of CAD systems 𝐿∞ norm of component 1: 7 and 11,
adversarial training classification tasks 𝐿∞ norm of component 2: 13

4.2.1. Encryption strategy
As training datasets supporting multi-party access, medical images
in CAD systems are often stored directly on cloud computing servers.
Therefore, using image encryption to provide cloud to client protec-
tion for medical images from storage to transmission are the most
direct security strategies of CAD systems. Earlier in Section 3, we
highlighted some MIE strategies that can resist poisoning attacks that
leak patients’ privacy. In addition to such strategies, sensitive data in
the memory can be further encrypted using hardware such as secure
processors. Recently, Intel developed a Software Guard Extension (SGX)
security processor [65] that adds a new set of instructions and changes
memory access in traditional Intel Architecture. This way, sensitive
data can only be dynamically decrypted in the CPU itself to prevent
unauthorised access irrespective of the user’s privilege level.
4.2.2. Differential privacy
Differential privacy refers to the mechanism of adding perturbation
to the datasets to protect classified data [76]. The idea behind differen-
tial privacy is that if the effect of any single substitution in the datasets
is small enough, the query results cannot be used to infer information
about any single individual. Therefore, applying the differential privacy
algorithm to ML/DL models in CAD systems can effectively prevent
attackers from accessing private data by repeatedly querying the mod-
els [66]. While differential privacy provides strong privacy guarantees
for training medical image data, the added perturbation will reduce the
availability of data and models. This motivated Sun et al. to propose
a method that combines differential privacy and decision tree (DPDT)
to publish data in [67], where the decision tree is used to determine
the extent of additional noise to the dataset that is required to balance
protection strength and data availability.
4.2.3. Federated learning
Federated learning can train an algorithm across multiple decen-
tralised edge devices or servers holding local data samples [69]. Each
device uses its local data to train the model, and then shares model
parameters with the central model without sharing its actual data.
Building a data network among organisations based on federated mech-
anisms allows knowledge sharing without compromising the privacy
of users in local data and thus establishes a united model so that
enterprises can work together on data security. In such a context,
in [68], Kumar et al. proposed a framework to collaboratively train
a global model using federated learning with blockchain technology
while preserving privacy. The framework uses datasets from multiple
hospitals to establish an accurate collaborative model to recognise lung
CT images from COVID-19 patients. As such, each hospital can keep its
data private and share only weights and gradients of trained models.
4.2.4. Knowledge distillation
In ML, knowledge distillation transfers knowledge from a large
model to a small one without loss of validity. Since the small model
uses a small dataset, it can be deployed even on CAD systems with
weak computing power, and it can also be used to prevent the risk of
losing patient privacy that is associated with large datasets. Motivated
by this intuition, Papernot et al. put forward the concept of defencive
distillation where the knowledge training model is extracted from deep
neural networks (DNN) to reduce the amplitude of network gradients

8
F. Yan et al.
exploited by adversaries to craft adversarial samples, and significantly
reduce the success of adversarial sample crafting to rates with little
overhead. In addition, Celik et al. implemented privacy distillation
using DNN [70] wherein patients are allowed to control the type and
amount of information they wish to disclose to CAD systems. Accuracy
of the diagnosis reported was almost as good as having all the data of
patients.
4.2.5. Adversarial attack detection
Adversarial attack detection [77] was proposed by Metzen et al.
to detect adversarial samples by training a simple binary classifier.
Furthermore, Ma et al. applied this method to detect adversarial ex-
amples in medical image analysis in [78]. They reported that using
this method, adversarial attacks against CAD systems can be easily
detected, and more than 98% of the detection AUC (area under curve)
can be achieved even in the most advanced attacks. Based on this, Li
et al. proposed an unsupervised learning approach to detect adversarial
attacks on medical images in [72]. This method can be easily embedded
into any CAD system using ML/DL as a module without sacrificing clas-
sification performance. In addition, unsupervised learning can classify
unlabelled training samples, allowing the scheme in [72] to detect a
variety of adversarial attacks without knowing the attacker. Addition-
ally, Watson et al. used explainable ML/DL to detect adversarial attacks
in chest X-ray datasets [73]. Explainable ML/DL is considered vital
for ethical, trusted, and automated decision making, especially in CAD
systems.
4.2.6. Adversarial training
Adversarial training is one of the defences against adversarial at-
tacks where adversarial examples in each training loop are used to
augment training data. Therefore, the adversarially trained models per-
form more effectively when facing adversarial examples than standard
trained models. Adversarial training is a min–max problem to find the
best solution in the worst case albeit at high computational cost. In their
contribution, Shafahi et al. reduced the overhead cost associated with
generating adversarial examples by recycling the gradient information
computed when updating model parameters [79]. In medical image
analysis, Lal et al. proposed strategies including defencive adversarial
training and feature fusion against adversarial attacks [74], which
reportedly improved the robustness of CAD system on the retinal fundus
images used in diabetic retinopathy recognition.
5. Application prospects for emerging medical image based smart
healthcare security technologies
This section showcases some emerging technologies with potential
applications in the field of medical image security. We cover paradigms
and technologies such as quantum information processing, blockchain,
and compressed sensing, as shown in Fig. 7. Furthermore, Table 3
presents an overview of these technologies and their emerging roles
in SHS.
5.1. Smart healthcare security based on quantum technology
Quantum information processing (QINP) is a paradigm where the
superposition of entangled states is harnessed to enhance speed, ver-
satility, and security [109]. Quantum hardware is reputed to exhibit
potentials and capabilities that rival their best-known classical (i.e., dig-
ital) counterparts. Two areas that QINP outperforms digital information
processing are those of security and efficiency, properties that make
it attractive in communication, computing, cryptography, and image
processing [110].
In efforts to utilise the huge potentials of QINP whilst circumventing
encumbrances related to information processing in the classical and
digital worlds, quantum-inspired frameworks have been deployed to
9
Journal of Information Security and Applications 78 (2023) 103621
Fig. 7. Some general emerging technologies applied in medical image based SHS.
enhance the performance of traditional counterparts of many applica-
tions, particularly, in the emerging sub-discipline of quantum machine
learning (QML). For example, the quantum particle swarm optimi-
sation (QPSO) strategically sheds off the moving direction attribute
in its classical counterpart, i.e., the PSO algorithm [82]. Therefore,
the update of particles is independent of the previous position of the
particle, which increases the randomness of the particle’s position.
On their part, the quantum evolutionary algorithms (QEA) [83] and
quantum GAs (QGAs) [81] (analogues of the EA and GA in the classical
world) make use of qubit coding to represent chromosomes and qubit
gates to update them. This ensures strong search optimisation in both
quantum variants of the ML methods. Furthermore, the two algorithms
exhibit good population dispersion, strong global search capability, fast
convergence, and easy integration with other algorithms.
Similar to ML, QML seems attractive for applications in SHS where
the use of QPSO and QEA in MIW have been explored in [82,83],
respectively. Both schemes report good imperceptibility and high ro-
bustness. Similar uses for QPSO and QGA have also been reported to
improve efficiency of medical image segmentation in [80,81], respec-
tively.
Meanwhile, by integrating QINP into SHS, El-Latif et al. explored
ways to enhance security of MISS in [84] where they presented a
quantum steganography algorithm for applications within a smart city
context as depicted in Fig. 8. As seen from the figure, the framework is
designed to facilitate image sharing between different hospitals where
medical staff can encode secret information (including other medical
images) as quantum information and embed them onto the images
captured in the other hospitals (also encoded as quantum informa-
tion). The scheme also proposed a quantum image processing-based
steganography scheme that uses the XOR operation targeted at the
least and most significant qubits as well as logistic chaotic mapping
and a watermarking strategy to safeguard the image prior to sending
it to the cloud server. At the other end, authorised medical staff of
the other hospital can recover the secret information using a quantum
extraction procedure. The framework utilises the parallelism inherent
to quantum computing to improve the efficiency of steganographic
protection for medical images. Additionally, the no-cloning precept of
the quantum states provides security for medical image sharing under
the framework. In yet another study, in [85], El-Latif et al. proposed
an encryption technique for quantum medical images. The highlighted
studies indicate a growing interest in utilising QINP to enhance the
security of SHS.
Meanwhile the use of quantum computing in image reconstruction
has also been reported in [111]. The need for image reconstruction
arises when different medical devices (such as CT, MRI, PET scanners)
are used to collect images. Specifically, Kiani et al. proposed a recon-
struction strategy for medical images [86], where the collated data
are considered as input quantum states, which can be processed using
quantum resources. This way, the exponential speedup could be gained
compared to the classical algorithm. Moreover, due to the improvement
F. Yan et al. Journal of Information Security and Applications 78 (2023) 103621

Table 3
Summary of emerging smart healthcare security technologies based on medical image.
Emerging technologies Classification References Applications Advantages
CAD system [80,81] Realising multi-objective segmentation of Improving the performance and
enhancement based on medical images and optimising protection of the training dataset using
QML parameters QML
Quantum
information Medical image security [82,83] Determining optimal scaling factors for Providing good visual fidelity for the
processing based on QML embedding medical images using QML watermark scheme using QML
Encoding medical [84,85] [86] Protecting medical image data using Improving security and reconstruction
images into quantum QINP-based encryption and performance of the medical images
states steganography
Access control based on [87,88] Access management for medical data in Access control is reasonably
blockchains [89,90] CAD system using blockchain technology decentralised, and integrity of data is
enhanced

Blockchain Medical image sharing [91,92] Tracking medical images during Reducing the network congestion in the
based on blockchains [93,94] transmission to guarantee privacy process of medical image sharing
protection
Secure healthcare [95,96] [97] Secure and efficient medical data Ensuring efficient and secure data
blockchain for IoT sharing based on the IoT and blockchain communication while saving computing
medical devices resources of IoT devices
Optimised sampling [98,99] Combining intelligent technology to High decryption accuracy is maintained
matrix [100,101] realise full sampling with fewer in cases of large compression ratio
Compressive sampling points
sensing
Combined with other [102,103] Integrating other image security Reducing pressure on information
image security [104,105] technologies to expand functions and transmission channel and improved
technologies [106,107] improve security capacity in steganography
[108]

in efficiency of medical image reconstruction, patents could be exposed

to less radiation that would otherwise be needed to acquire new or
additional images.
It is worth noting that the highlighted studies on quantum tech-
nology are generally based on MATLAB platform to simulate quan-
tum operations. These simulations are based on linear algebra, where
complex vectors are used as quantum states and unitary matrices as
quantum transformations. Other commonly used quantum simulators
include Qiskit [112], PennyLane [113], etc.

5.2. Smart healthcare security based on blockchain

In late 2019, the world faced an existential threat in COVID-19, Fig. 8. Framework for hiding medical quantum images proposed by El-Latif et al.
which exposed the inadequacies of medical facilities in even the most in [84].

advanced economies and nations. In all this, privacy of patients’ med-

ical data was exposed to different risks. Meanwhile, considered as
a network of centralised databases managed by peer-based networks
in compliance with clearly set protocols for inter-node communica-
tion [114], the utility of blockchain technology was exploited to safe-
guard patient data in the midst of the COVID-19 crisis. In blockchains,
data blocks are efficiently linked and secured to resist unauthorised
access. This tamper-proof security offered by blockchains is among the
many attractions for its use across different fields and applications. Its
most popular application is Bitcoins, which is an online distributed
ledger for cryptocurrencies.
As an important type of EHR, research on blockchain-based secu-
rity for medical images is also gaining traction. Moreover, as earlier
reviewed, the data sharing mechanism employed in most IoT-based SHS
are vulnerable to different types of attacks, such as privacy leakage,
tampering, forgery, etc. Therefore, blockchain technology offers a safe
and seamless platform to access these EHRs across different facilities.
For example, studies in [89,90] focused specifically on MRI and X-ray
images where the underlying principle in both methods is that when
a host requires access to the medical image, each co-host must first
sign-off on the request and then others that need the same resources
undertake an integrity test to ascertain the authenticity of the data.
Meanwhile, being at the centre of such blockchain frameworks, patients
are directly involved in signing-off before access to their data is given to
any third party [87]. In [88], Kim et al. proposed a medical question-
naire sharing method to safely collate patient data using blockchain.
They reported a semi-permanent data storage that grants access control
to patients themselves.
More recently, applications for blockchain technology in medical
image sharing have been explored. Some of these techniques have
combined blockchains with AI (particularly, ML and DL techniques) to
enhance the capability of SHS in the fight against the COVID-19 pan-
demic. For example, in [91], Nguyen et al. proposed a framework that
combines blockchain with AI for contact tracing, privacy protection and
medical supply chain management. In a slightly different contribution,
Sultana et al. proposed a medical image sharing system integrating
zero trust principle and blockchain technology in [93]. Therein, the
blockchain is used to audit and track the medical image data trans-
mission to determine whether the medical image is leaked and/or to
log when the leakage occurs in the future. They argue that their use
of the zero-trust principle is to ensure strict verification of users and
devices trying to access the resources on the network, which is aimed
at ensuring the security of medical image data during the transmission
process. Similarly, in [94], Tang et al. proposed a secure medical image
sharing scheme based on blockchain and smart contract of credit scores.
Their scheme is built on the premise that by establishing a smart con-
tract based on the credit scores of patients and medical institutions, the
process of medical image sharing can be trusted and supervised, which
also improves the robustness of PACS system. Meanwhile, in [92], Fan
et al. proposed an information management system (called MedBlock)

10
F. Yan et al.
to process patients’ EHR. They demonstrated that the use of blockchain
enhanced the security of EHR and reduced network congestion that
usually accompanies the EHR retrieval process.
Applications of blockchains in the field of medical image analysis,
specifically, medical image segmentation, have also been explored.
However, most studies reported the need for substantial amounts of
resources for training and computing power. Motivated by the need to
overcome this, Li et al. proposed a strategy to harness the computing
power of blockchain miners to perform medical image segmentation
and took the workload of miners to demonstrate Proof-of-Useful-Work
(PoUW) [115]. This method repurposes the computing power gener-
ated by miners, which the study reports led to improvement in social
responsibility and efficiency of medical image segmentation.
Meanwhile, blockchain technology can be combined with IoT to
explore efficient and secure medical data sharing technology based on
IoT devices. In this regard, Singh et al. proposed an IoT and blockchain
ledger technology based on improved zero-knowledge protocols [95].
This technology allows multiple range-proofs to be aggregated into a
single range-proof, thereby reducing verification size and improving the
efficiency of range-proof. Similarly, Dhar et al. proposed a distributed
system framework for sharing and exchanging multimedia files on
wireless IoT platforms [96]. The framework is based on blockchain
and InterPlanetary File System and uses a strong secure authentication
protocol based on zero-knowledge proofs, which provides high security
without compromising latency. In their contribution, Srivastava et al.
proposed a novel blockchain-based IoT model that comprises overlay
network, cloud storage, healthcare providers, smart contracts, and IoT-
equipped patients [97]. This model employs lightweight encryption
technology to offer secure network data communication and cloud
storage, while eliminating the concept of PoUW leading to conservation
of computing resources for IoT devices.
5.3. Medical image security based on compressive sensing
Compressive sensing is an emerging technology that helps to com-
press data at rates higher than traditional methods [116]. Due to the
independent energy requirement of sensors, their computing power
and information transmission capacity are often limited. As such, the
application of compressed sensing in wireless sensor networks (WSN)
can significantly reduce the power consumption and traffic demand
of each sensor. In healthcare, medical images often contain a lot of
visual information, which leads to the slow sampling speed in medical
imaging, such as MRI. As a new signal sampling paradigm, compressed
sensing technology (CST) can produce the same results as full sampling
with fewer sampling points, that is, only a small fraction of the data
needs to be sampled to provide high fidelity reconstruction of the
original image [98].
CST itself has the property of encryption because a sampling matrix
is required to reconstruct the original signal. As a result, many studies
on medical image security focus on the compressive sensing algorithm
since it can compress and encrypt medical images synchronously [99].
Moreover, the sampling matrix can be learned from the training image
using CNN, so that the measured value of compressed sensing can retain
more information on the image structure and reconstruct the image
better [100]. Accordingly, in [101], Wang et al. proposed an algorithm
based on tensor compressed sensing, which the authors claim is suitable
for compressing and encrypting 3D medical images represented as ten-
sors. They further asserted that their algorithm records high decryption
accuracy even at large compression ratios.
Due to its combination of security features and compression capa-
bility, CST is often used alongside other image security technologies.
For example, in [104], Zhang et al. proposed a MIE and compression
scheme based on CST and chaotic encryption. Similarly, in [102],
Sreedhanya et al. embedded the encrypted image generated by CST
into another image using steganography models. They further reported
an increase in the capacity of the secret file embedded in the carrier
11
Journal of Information Security and Applications 78 (2023) 103621
image while also satisfying the requirements for security and robust-
ness. In their contribution, Wang et al. proposed a medical image
acquisition and transmission solution based on CST in [103]. Applying
this scheme, they reported that a medical image can be compressed to
20%, which greatly alleviates pressure on the information transmission
channel, while the encryption attribute of compressed sensing gives
the medical image high security. More importantly, the scheme inte-
grates homomorphic encryption technology and meets the aggregation
requirements, which makes it conducive for fusion and aggregation
of medical data, such as accurate collection of public medical signals
through collaborative sensing. Furthermore, [105] presents an image
encryption technique based on two-dimensional CST that divides the
original image into non-overlapping blocks, compresses them using
CST, and then encrypts them using chaotic systems. It is outlined that
the scheme increases the transmission volume and efficiency of key
distribution.
In the highlighted applications of CST, the Gaussian distribution of
the measurement matrix leads to the non-uniform distribution of the
cryptographic image, which hackers could exploit. To mitigate this,
in [117], Huang et al. used Arnold scrambling to hide the Gaussian
distribution of cipher images. They reported enhanced the robustness of
medical image transmission against continuous packet loss or malicious
clipping attacks.
Meanwhile, parallel compressed sensing (PCS), which is an im-
proved version of the traditional CST, rearranges the one-dimensional
signals to be operated into a two-dimensional matrix and builds the
measurement matrix on the initial values and control parameters of
the chaotic system. Noting that PCS can lower the required stor-
age space and computational complexity in comparison to traditional
CST, Wang et al. proposed two image encryption algorithms for se-
cure communication [106,107] and demonstrated that PCS as a key
step substantially improves the efficiency of image encryption and
transmission. In another related study, Xiao et al. stated that the two-
dimensional Kronecker CST minimises the computational cost of the
encoding end and is suitable for use in devices with limited computing
capabilities [108].
6. Concluding remarks and overview of challenges facing SHS
A confluence between developments in information technology and
engineering on one hand and medicine on another have led to the
growth in the popularity of smart healthcare systems (SHS). Among
others, the main advantages of SHS include:
• Patient-centred care. Modern and progressive high-tech health-
care environments can meet the healthcare needs of patients
and accord due respect to their autonomy, so that they can be
empowered to become active participants in their care process.
• Medical informatisation. SHS uses a new generation of infor-
mation technologies, such as IoT, big data, cloud computing, and
AI, to make healthcare delivery more efficient, convenient, and
secure.
• Personalised management. SHS can coordinate and integrate
the medical systems to realise the personalised management for
patient care to ensure patients get the services they need most
and facilitate the rational allocation of medical resources.
• Preventive healthcare. SHS consists of multiple stakeholders
comprising of doctors and patients, hospitals, and research in-
stitutions. The health of patients can be monitored at any time
through wearable devices to reduce the likelihood and severity
of potentially life-threatening illnesses.
• Potential for new tech integration. Emerging technologies such
as quantum information processing, blockchain, and compres-
sive sensing demonstrate exciting prospects for more secure and
efficient storage, sharing, and analysis of medical images in SHS.
F. Yan et al.
As the world continues its march towards globalisation, where local
facilities are expected to operate at international levels, it is pertinent
we look in the direction of intelligent and emerging technologies to
safeguard privacy and enhance efficiency in all areas of healthcare
access and delivery. This study primarily focused on providing an
overview of current state-of-the-art in that regard. First, this study
proposed an overall architecture for security in SHS comprising of
MISS and CADS. Second, it elaborated on how the most advanced
intelligent technologies, including artificial intelligence, the Internet of
Things, and cloud computing, are used in the medical image based SHS.
Third, the study enumerates the risk posed by adversarial attacks to
the ML/DL model and medical image training data in the CAD system
and highlights the security defense strategies aimed at reducing such
risks. Finally, the study highlighted the significance of various emerging
technologies (including quantum information processing, blockchain,
and compressed sensing) in the future medical image based SHS.
While the technologies covered in this review have highlighted var-
ious levels of success, our review indicates that the immense potentials
of many emerging technologies have not been fully exploited in the
long road to safe, robust, and efficient SHS. Consequently, in the sequel,
we enumerate a few areas that would definitely help to expedite this
journey.
• Challenges in lack of standardised schemes for medical image
security. Current research on secure medical image sharing have
overlooked the difference between classical and medical images.
In many of the cases that classical image sharing techniques are
directly applied to medical images, great security risks have been
encountered [118]. Whereas many are aware of the need for well-
designed security strategies to meet the security requirements of
the medical environment, due to the lack of unified standards,
the required strategies cannot be applied in practice. In addition,
due to the requirement for high precision and computing time for
medical image analysis, the smart technologies used to analyse
classical images cannot be directly applied to medical images. In
fact, Zhou [119] and Kumar [120] emphasised this peculiarity
of medical images in the field of medical image segmentation
and detection, respectively. Similarly, Zanty et al. [121] com-
pared the performance and security of key classical and clustering
technology used for classical medical image segmentation and
determined three suitable segmentation technologies for medical
images. Although the birth of the DICOM framework [2] signifies
efforts towards unification of standards for medical image storage
and transmission, there is still a lack of standardised schemes for
security, which is very essential for the realisation of efficient of
SHS.
• Challenges in securing information collection terminals. As
highlighted in Section 3.2.2, whereas IoT-based SHS use sensors
to acquire several types of medical and health records, the in-
formation collection methods for IoT frameworks often require a
variety of WSNs. These sensors use CoT architecture to transmit,
and little consideration is given to the vulnerabilities both the
sensors and wireless networks are exposed to [122], particularly
when the networks are designed for medical infrastructure. First,
the channel of the WSN is open to anyone with appropriate
devices, so it is vulnerable to radio jamming, spoofing, eaves-
dropping, etc. Second, oftentimes, choice of the manufacturers of
WSN can often be predicted, which is guided by cost. Potential at-
tackers can use real-world WSN hardware to prepare and develop
attacks in advance, so as to increase the chances of attackers’
success. Third, a large number of WSNs are easily damaged due
to the changes in environment and weather. Consequently, the
unpredictable sensor ‘‘death’’ will affect the network performance
and system reliability. Therefore, it is very important to design
a secure network topology. Presently, studies on sensor-level
security protection for information collection terminals (which
12
Journal of Information Security and Applications 78 (2023) 103621
moves privacy protection closer to the data source) need to be
enhanced [123]. Early studies mainly focus on integrating the
watermark function on the sensor and embedding the watermark
onto the collected image data in real-time [124]. Meanwhile,
with continuous improvement in performance, sensors can be
used to implement more complex security schemes. For example,
Pittaluga et al. integrated the sensor with privacy preserving
optics to filter or block sensitive information directly from the
incident light-field [125]. In [126], Prabhu et al. secured the high
frequency video streams collected by using memory and computer
constrained hardware security extensions.
• Challenges of building a unified conceptual framework for
explainable ML. Despite their promise and potential, ML models
play the role of a ‘‘black box’’ in many applications in medical
image analysis as well as the larger SHS frameworks. With the
ever increasing application of ML models in clinical practice, the
trust of clinicians needs to be established. ‘Explainability’ or the
ability to justify its outcomes and assist clinicians in rationalising
the model prediction, is often considered to be the key to building
such trust. However, since everyone has different conditions for
trust in ML, only a unified definition of a reliable ML system can
provide continuous usage and have a positive clinical impact. In
the field of clinical healthcare, the general agreement regarding
explainability is fairly low. Several studies have attempted to
provide a basis for rigorous evaluation of explainability methods,
including application types [127], user suggestions [128] and
statistical methods [129]. However, there is still a lack of unified
conceptual framework covering all kinds of ML to support the
explainability needs of the medical and health infrastructure.
• Challenges in other emerging technologies. In the field of
quantum information, the realisation of physical quantum hard-
ware is still a challenge in terms of qubit technology, control
electronics, microarchitecture for the execution of quantum cir-
cuits, and efficient quantum error correction. Meanwhile, the
vulnerability of qubit to decoherance implies that, upon inter-
action with environment, information stored in a qubit will be
lost leading to gate error rates of around 10−2 [130]. However,
if the gate error rates are below a certain threshold, the quantum
system can be protected and/or recovered from such errors by
using quantum error correction (QEC) and fault tolerant (FT)
computations [131], but they will also greatly increase the num-
ber of qubits required for computation. Therefore, the challenge
for advanced SHS based on quantum technology comes from
increasing the fidelity of quantum operations and restricting the
number of qubits [132]. In blockchain technology, large blocks
come with large storage space, but their network propagation
speed is slow, resulting in block centralisation, which potentially
increases security risks. Therefore, the trade-off between block
size and security has become a challenge [133]. Another issue
worth noting is that many blockchain specific solutions based on
medical IoT are accompanied by high computational and band-
width costs, while many edge devices do not have the computing
power required for blockchain functionality [134]. Therefore,
implementing the original functionality of these solutions can
be very difficult. In addition, although compressive sensing can
reduce the amount of data in medical images and also relieve
the pressure on transmission channels, it may also lead to a
decrease in medical image quality, which can affect diagnostic
results. Therefore, researchers need to maintain compromise be-
tween medical image quality while pursuing a high compression
rate [135]. At the same time, many medical images are non-
sparse in time domain and other transform domains, so designing
more efficient schemes to recover compressed medical images is
a challenge for compressive sensing technologies [136].
F. Yan et al.
To recap, our review has uncovered that, while medical image based
SHS has a lot of prospects in terms of providing patients with timely and
efficient medical services (when needed), every aspect of the medical
and health ecosystem could benefit from recent advances in informa-
tion and communication technologies and engineering. Moreover, a
unified security management for medical information will help medical
facilities reduce running costs, doctors’ efficiency via reduced work-
loads, and secure patients’ privacy for the benefit of all stakeholders. In
future work, we will focus on the design of smart healthcare security
technology suitable for edge computing nodes in the CoT as well as the
pursuit of greater medical data encryption efficiency and rapid response
to medical tasks with minimal computing resources. Moreover, in view
of the difficulties faced by the current quantum information technol-
ogy, we are more inclined to applying quantum-inspired algorithms to
medical image based SHS in order to develop medical image security
algorithms built on quantum computing principles that can run on
traditional computers.
Declaration of competing interest
The authors declare that they have no conflict of interest.
Data availability
Data will be made available on request.
Acknowledgements
This study is sponsored by the Prince Sattam Bin Abdulaziz Univer-
sity, Saudi Arabia via funding for the Project Number 2023/RV/0005
while support from the Future University in Egypt is also acknowl-
edged.
Funding
This study is sponsored in full by the Prince Sattam Bin Abdulaziz
University via funding for the Project Number 2023/RV/0005.
References
[1] Iliyasu AM. Cellular-automated protocol to safeguard confidentiality of QR
codes. IEEE Access 2019;7:144451–71.
[2] Al-Qershi OM. Authentication and data hiding using a hybrid ROI-based
watermarking scheme for DICOM images. J Digit Imaging 2009;24(1):114–25.
[3] Doi K. Computer-aided diagnosis in medical imaging: Historical review, current
status and future potential. Comput Med Imaging Graph 2007;31(4–5).
[4] Sedik A, Iliyasu AM, Abd El-Rahiem B, Abdel Samea ME, Abdel-Raheem A,
Hammad M, Peng J, Abd El-Samie FE, Abd El-Latif AA. Deploying machine
and deep learning models for efficient data-augmented detection of COVID-19
infections. Viruses 2020;12:769.
[5] Sun W, Cai Z, Li Y, Liu F, Fang S, Wang G. Security and privacy in the medical
internet of things: A review. Secur Commun Netw 2018;2018:1–9.
[6] Szegedy C, Zaremba W, Sutskever I, Bruna J, Erhan D, Goodfellow I. Intriguing
properties of neural networks. In: 2nd international conference on learning
representations. 2014.
[7] Qayyum A, Qadir J, Bilal M, Al-Fuqaha A. Secure and robust machine learning
for healthcare: A survey. IEEE Rev Biomed Eng 2020;14:156–80.
[8] Nguyen DC, Pham QV, Pathirana PN, Ding M, Seneviratne A, Lin Z, Dobre O,
Hwang WJ. Federated learning for smart healthcare: A survey. ACM Comput
Surv 2022;55(3):1–37.
[9] Quy VK, Hau NV, Anh DV, Ngoc LA. Smart healthcare IoT applications based on
fog computing: Architecture, applications and challenges. Complex Intell Syst
2021;8(5):3805–15.
[10] Ullah A, Azeem M, Ashraf H, Alaboudi AA, Humayun M, Jhanjhi NZ. Secure
healthcare data aggregation and transmission in IoT—A survey. IEEE Access
2021;9:16849–65.
[11] Hathaliya JJ, Tanwar S. An exhaustive survey on security and privacy issues
in healthcare 4.0. Comput Commun 2020;153:311–35.
[12] Tariq N, Qamar A, Asim M, Khan FA. Blockchain and smart healthcare security:
A survey. Procedia Comput Sci 2020;175:615–20.
13
Journal of Information Security and Applications 78 (2023) 103621
[13] Panchatcharam P, Vivekanandan S. Internet of things (IOT) in healthcare—
Smart health and surveillance, architectures, security analysis and data transfer:
A review. Int J Softw Innov 2019;7(2):21–40.
[14] Algarni A. A survey and classification of security and privacy research in smart
healthcare systems. IEEE Access 2019;7:101879–94.
[15] Masood I, Wang Y, Duad A, Aljohani NR, Dawood H. Towards smart healthcare:
Patient data privacy and security in sensor-cloud infrastructure. Wirel Commun
Mob Comput 2018;2018:2143897.
[16] He D, Ye R, Chan S, Guizani M, Xu Y. Privacy in the internet of things for
smart healthcare. IEEE Commun Mag 2018;56(4):38–44.
[17] Theek B, Nolte T, Pantke D, Schrank F, Gremse F, Schulz V, Kiessling F.
Emerging methods in radiology. JDer Radiol 2020;60:S41–53.
[18] Abhinav S, Chandan S. Medical image authentication through watermarking.
Int J Adv Res Comput Sci Technol 2014.
[19] Aparna P, Kishore P. Bio-metric based efficient medical image watermarking in
E-healthcare application. IET Image Process 2018;13(3):421–8.
[20] Alshanbari H. Medical image watermarking for ownership & tamper detection.
Multimedia Tools Appl 2021;80(11):16549–64.
[21] Yan F, Huang H, Yu X. A multiwatermarking scheme for verifying medical
image integrity and authenticity in the internet of medical things. IEEE Trans
Ind Inf 2022;18(12):8885–94.
[22] Abd-El-Atty B, Iliyasu AM, Alaskar H, El-Latif A, Ahmed A. A robust quasi-
quantum walks-based steganography protocol for secure transmission of images
on cloud-based E-healthcare platforms. Sensors 2020;20(11):3108–27.
[23] Jeevitha S, Prabha NA. Effective payload and improved security using
HMT contourlet transform in medical image steganography. Health Technol
2019;10(1):217–29.
[24] Sun J, Xu W, Feng B. A global search strategy of quantum-behaved particle
swarm optimization. In: Proceedings of IEEE conference on cybernetics and
intelligent systems. 2004, p. 111–6.
[25] Soliman M, Hassanien A, Onsi H. An adaptive watermarking approach based
on weighted quantum particle swarm optimization. Neural Comput Appl
2016;27(2):469–81.
[26] Stoyanov B, Stoyanov B. BOOST: Medical image steganography using nuclear
spin generator. Entropy 2020;22(5):501–13.
[27] Nazeer M, Nargis B, Malik YM, Kim D. A fresnelet-based encryption of medical
images using Arnold transform. Int J Adv Comput Sci Appl 2013;4(3):131–40.
[28] Zhang C, Li J, Wang S, Wang Z. An encrypted medical image retrieval algorithm
based on DWT-DCT frequency domain. In: IEEE 15th international conference
on software engineering research, management and applications. 2017, p.
135–41.
[29] Shih FY, Wu Y. Robust watermarking and compression for medical images based
on genetic algorithms. Inform Sci 2005;173(3):200–16.
[30] Pareek NK, V. Patidar medical image protection using genetic algorithm
operations. Soft Comput 2014;20(2):763–72.
[31] Lei B, Tan E, Chen S, Ni D, Wang T, Lei H. Reversible watermarking
scheme for medical image based on differential evolution. Expert Syst Appl
2014;41(7):3178–88.
[32] Rai A, Singh HV. Machine learning-based robust watermarking technique for
medical image transmitted over LTE network. J Intell Syst 2017;27(1):105–14.
[33] Manikandan VM, Masilamani V. Reversible data hiding scheme during
encryption using machine learning. Proc Comput Sci 2018;133:348–56.
[34] Ma B, Li B, Wang X, Wang C, Li J, Shi Y. Code division multiplexing and
machine learning based reversible data hiding scheme for medical image. Secur
Commun Netw 2019;1–9.
[35] Aazam M, Khan I, Alsaffar AA, Huh E. Cloud of things: integrating Internet of
Things and cloud computing and the issues involved. In: Proceedings of 11th
international bhurban conference on applied sciences & technology. 2014, p.
414–9.
[36] Al-Shayea TK, Batalla JM, Mavromoustakis CX, Mastorakis G. Embedded
dynamic modification for efficient watermarking using different medical inputs
in IoT. In: IEEE 24th international workshop on computer aided modeling and
design of communication links and networks. 2019, p. 1–6.
[37] Hashim MM, Rhaif SH, Abdulrazzaq AA, Ali AH, Taha MS. Based on
IoT healthcare application for medical data authentication: towards a new
secure framework using steganography. IOP Conf Ser: Mater Sci Eng
2020;881(1):012120.
[38] Khan J, Li J, Haq AU, Parveen S, Khan GA, Shahid M, Monday HN, Ullah S,
Ruinan S. Medical image encryption into smart healthcare IOT system. In:
16th international computer conference on wavelet active media technology
and information processing. 2019, p. 378–82.
[39] Rajagopalan S, Janakiraman S, Rengarajan A, Rethinam S, Arumugham S,
Saravanan G. IoT framework for secure medical image transmission. In: IEEE
international conference on computer communication and informatics. 2018, p.
1–5.
[40] Ukis V, Rajamani ST, Balachandran B, Friese T. Architecture of cloud-based ad-
vanced medical image visualization solution. In: IEEE international conference
on cloud computing in emerging markets. 2013, p. 1–5.
[41] Chiang W, Lin H, Wu T, Chen C. Bulding a cloud service for medical
image processing based on service-orient archtecture. In: IEEE 4th international
conference on biomedical engineering and informatics. 2011, p. 1459–65.
F. Yan et al.
[42] Marwan M, Kartit A, Ouahmane H. A secure framework for medical image
storage based on multi-cloud. In: IEEE 2nd international conference on cloud
computing technologies and applications, Vol. 41. 2016, p. 88–94, (7).
[43] Marwan M, Kartit A, Ouahmane H. A new medical image processing approach
for the security of cloud services. In: RTIS 2017: Lecture notes in real-time
intelligent systems. 2018, p. 378–92.
[44] Marwan M, Kartit A, Ouahmane H. A cloud-based framework to secure medical
image processing. J Mob Multimedia 2018;14(3):319–44.
[45] Hartmann M, Hashmi US, Imran A. Edge computing in smart health care
systems: Review, challenges, and research directions. Trans Emerg Telecommun
Technol 2019;e3710.
[46] Pustokhina I, Pustokhin D, Gupta D, Khanna A, Shankar K, Nguyen G. An
effective training scheme for deep neural network in edge computing enabled
internet of medical things (IoMT) systems. IEEE Access 2020;8:107112–23.
[47] El Attaoui A, Largo S, Kaissari S, Benba A, Jilbab A, Bourouhou A. Machine
learning-based edge-computing on a multi-level architecture of WSN and IoT
for real-time fall detection. IET Wirel Sens Syst 2020;10(6):320–32.
[48] Khan WZ, Ahmed E, Hakak S, Yaqoob I, Ahmed A. Edge computing: A survey.
Future Gener Comput Syst 2019;97:219–35.
[49] Alshehri F, Muhammad G. A comprehensive survey of the Internet of Things
(IoT) and AI-based smart healthcare. IEEE Access 2021;9:3660–78.
[50] Sufian A, Ghosh A, Sadiq AS, Smarandache F. A survey on deep transfer
learning to edge computing for mitigating the COVID-19 pandemic. J Syst
Archit 2020;108:101830.
[51] Oliveira F, Tavares J. Medical image registration: A review. Comput Methods
Biomech Biomed Eng 2012;17(2):73–93.
[52] Kubota K, Kuroda J, Yoshida M, Ohta K, Kitajima M. Medical image analysis:
Computer-aided diagnosis of gastric cancer invasion on endoscopic images. Surg
Endosc 2012;26(5):1485–9.
[53] Duncan JS, Ayache N. Medical image analysis: Progress over two decades and
the challenges ahead. IEEE Trans Pattern Anal Mach Intell 2000;22:85–106.
[54] Cao Y, Steffey S, He J, Xiao D, Tao C, Chen P, Muller H. Medical image
retrieval: a multimodal approach. Cancer Inform 2015;13(S3):125–36.
[55] Rajkomar A, Lingam S, Taylor AG, Blum M, Mongan J. High-throughput
classification of radiographs using deep convolutional neural networks. J Digit
Imaging 2016;30(1):95–101.
[56] Milletari F, Navab N, Ahmadi S. V-net: Fully convolutional neural networks for
volumetric medical image segmentation. In: Fourth international conference on
3D vision (3DV). 2016, p. 565–71.
[57] Haskins G, Kruecker J, Kruger U, Xu S, Pinto PA, Wood BJ, Yan P. Learning
deep similarity metric for 3D MR–TRUS image registration. Int J Comput Assist
Radiol Surg 2018;14(3):417–25.
[58] Biggio B, Corona I, Maiorca D, Nelson B, Srndic N, Laskov P, Giacinto G,
Roli F. Evasion attacks against machine learning at test time. In: Proceedings
of joint European conference on machine learning and knowledge discovery in
databases. 2013, p. 387–402.
[59] Zhang F, Chan P, Biggio B, Yeung D, Roli F. Adversarial feature selection against
evasion attacks. IEEE Trans Cybern 2015;46(3):766–77.
[60] Kwon H, Yoon H, Park K. Selective poisoning attack on deep neural networks.
Symmetry 2019;11(7):892.
[61] Mozaffari-Kermani M, Sur-Kolay S, Raghunathan A, Jha N. Systematic poisoning
attacks on and defenses for machine learning in healthcare. IEEE J Biomed
Health Inf 2014;19(6):1893–905.
[62] Sakib S, Tazrin T, Fouda M, Fadlullah Z, Guizani M. DL-CRC: Deep learning-
based chest radiograph classification for COVID-19 detection: A novel approach.
IEEE Access 2020;8:171575–89.
[63] Lee T, Edwards B, Molloy I, Su D. Defending against neural network model
stealing attacks using deceptive perturbations. In: IEEE security and privacy
workshops. 2019, p. 43–9.
[64] Zhang L, Lin G, Gao B, Qin Z, Tai Y, Zhang J. Neural model stealing attack
to smart mobile device on intelligent medical platform. Wirel Commun Mob
Comput 2020;2020:8859489.
[65] Ohrimenko O, Schuster F, Fournet C, Mehta A, Nowozin S, Vaswani K, Costa M.
Oblivious multi-party machine learning on trusted processors. In: Proceedings
of in 25th security symposium. 2016, p. 619–36.
[66] Ziller A, Usynin D, Braren R, Makowski M, Rueckert D, Kaissis G. Medical
imaging deep learning with differential privacy. Sci Rep 2021;11(1):1–8.
[67] Sun Z, Wang Y, Shu M, Liu R, Zhao H. Differential privacy for data and model
publishing of medical data. IEEE Access 2019;7:152103–14.
[68] Kumar R, Khan A, Zhang S, Kumar J, Yang T, Golilarz N, Ali I, Shafiq S,
Wang W. Blockchain-federated-learning and deep learning models for COVID-19
detection using CT imaging. IEEE Sens J 2021;21(14):16301–14.
[69] Sui D, Chen Y, Jia Y, Xie Y, Sun W. Feded: Federated learning via ensemble
distillation for medical relation extraction. In: Proceedings of the conference on
empirical methods in natural language processing. 2020, p. 2118–28.
[70] Celik Z, Lopez-Paz D, McDaniel P. Patient-driven privacy control through
generalized distillation. In: Proceedings of IEEE symposium on privacy-aware
computing. 2017, p. 1–12.
[71] International Warfarin Pharmacogenetics Consortium. Estimation of the
warfarin dose with clinical and pharmacogenetic data. N Engl J Med
2009;360(8):753–64.
14
Journal of Information Security and Applications 78 (2023) 103621
[72] Li X, Zhu D. Robust detection of adversarial attacks on medical images. In:
Proceedings of 17th international symposium on biomedical imaging. 2020, p.
1154–8.
[73] Watson M, Moubayed N. Attack-agnostic adversarial detection on medical
data using explainable machine learning. In: Proceedings of 25th international
conference on pattern recognition. 2021, p. 8180–7.
[74] Lal S, Rehman S, Shah J, Meraj T, Rauf H, Damasevicius R, Mohammed M,
Abdulkareem K. Adversarial attack and defence through adversarial training and
feature fusion for diabetic retinopathy recognition. Sensors 2021;21(11):3922.
[75] Xu M, Zhang T, Li Z, Liu M, Zhang D. Towards evaluating the robustness of
deep diagnostic models by adversarial attack. Med Image Anal 2021;69:101977.
[76] Abadi M, Chu A, Goodfellow I, McMahan H, Mironov I, Talwar K, Zhang L. Deep
learning with differential privacy. In: Proceedings of the 2016 ACM SIGSAC
conference on computer and communications security. 2016, p. 308–18.
[77] Metzen J, Genewein T, Fischer V, Bischoff B. On detecting adversarial
perturbations. In: Proceedings of 5th international conference on learning
representations. 2017, p. 24–6.
[78] Ma X, Niu Y, Gu L, Wang Y, Zhao Y, Bailey J, Lu F. Understanding adversarial
attacks on deep learning based medical image analysis systems. Pattern Recognit
2021;110:107332.
[79] Shafahi A, Najibi M, Ghiasi A, Xu Z, Dickerson J, Studer C, Davis L, Tay-
lor G, Goldstein T. Adversarial training for free!. In: Proceedings of the 33rd
international conference on neural information processing systems. 2019, p.
3358–69.
[80] Li Y, Shang R, Stolkin R. Dynamic-context cooperative quantum-behaved
particle swarm optimization based on multilevel thresholding applied to medical
image segmentation. Inform Sci 2015;294:408–22.
[81] Hilali-Jaghdam I, Ishak AB, Abdel-Khalek S, Jamal A. Quantum and classical ge-
netic algorithms for multilevel segmentation of medical images: A comparative
study. Comput Commun 2020;162:83–93.
[82] Soliman MM, Hassanien AE, Onsi HM. An adaptive medical images water-
marking using quantum particle swarm optimization. In: 35th international
conference on telecommunications and signal processing. 2012, p. 735–9.
[83] Samantaa S, Choudhurya A, Deyb N, Ashourc AS, Balasd VE. Quantum-inspired
evolutionary algorithm for scaling factor optimization during manifold medical
information embedding. Quantum Inspir Comput Intell 2017;285–326.
[84] Abd El-Latif AA, Abd-El-Atty B, Hossain MS, Rahman A, Alamri A, Gupta BB.
Efficient quantum information hiding for remote medical image sharing. IEEE
Access 2018;6:21075–83.
[85] Abd El-Latif AA, Abd-El-Atty B, Talha M. Robust encryption of quantum medical
images. IEEE Access 2017;6:1073–81.
[86] Kiani BT, Villanyi A, Lloyd S. Quantum medical imaging algorithms. 2020,
arXiv:2004.02036.
[87] Yue X, Wang H, Jin D, Li M, Jiang W. Healthcare data gateways: Found
healthcare intelligence on blockchain with novel privacy risk control. J Med
Syst 2016;40(10):1–8.
[88] Kim MG, Lee AR, Kwon HJ, Kim JW, Kim IK. Sharing medical questionnaries
based on blockchain. In: IEEE international conference on bioinformatics and
biomedicine. 2018, p. 2767–9.
[89] Liu B, Liu M, Jiang X, Zhao F, Wang R. A blockchain-based scheme for secure
sharing of X-ray medical images. In: International conference on security with
intelligent computing and big-data services. 2019, p. 29–42.
[90] Brunesea L, Mercaldob F, Reginellic A, Santoned A. A blockchain based proposal
for protecting healthcare systems through formal methods. Procedia Comput Sci
2019;159:1787–94.
[91] Ding M, Pathirana PN, Seneviratne A. Blockchain and AI-based solutions to
combat coronavirus (COVID-19)-like epidemics: A survey. TechRxiv 2020.
[92] Fan K, Wang S, Ren Y, Li H, Yang Y. Medblock: Efficient and secure medical
data sharing via blockchain. J Med Syst 2018;42(8):1–11.
[93] Sultana M, Hossain A, Laila F, Taher KA, Islam MN. Towards developing
a secure medical image sharing system based on zero trust principles and
blockchain technology. BMC Med Inform Decis Mak 2020;20(1):1–10.
[94] Tang H, Tong N, Ouyang J. Medical images sharing system based on blockchain
and smart contract of credit scores. In: The 1st IEEE international conference
on hot information-centric networking. 2018, p. 240–1.
[95] Singh R, Dwivedi AD, Mukkamala RR, Alnumay WS. Privacy-preserving ledger
for blockchain and Internet of Things-enabled cyber–physical systems. Comput
Electr Eng 2022;103:108290.
[96] Dhar S, Khare A, Singh R. Advanced security model for multimedia data sharing
in Internet of Things. Trans Emerg Telecommun Technol 2022;e4621.
[97] Srivastava G, Crichigno J, Dhar S. A light and secure healthcare blockchain for
IoT medical devices. In: Proceedings of IEEE Canadian conference of electrical
and computer engineering. 2019, p. 1–5.
[98] Wang G, Bresler Y, Ntziachristos V. Guest editorial compressive sensing for
biomedical imaging. IEEE Trans Med Imaging 2011;30(5):1013–6.
[99] Thanki R, Borra S, Dwivedi V, Borisagar K. A steganographic approach for
secure communication of medical images based on the DCT-SVD and the
compressed sensing (CS) theory. J Imaging Sci 2017;65(8):457–67.
[100] Shi W, Jiang F, Liu S, Zhao D. Image compressed sensing using convolutional
neural network. IEEE Trans Image Process 2019;29:375–88.
F. Yan et al.
[101] Wang Q, Chen X, Wei M, Miao Z. Simultaneous encryption and compression of
medical images based on optimized tensor compressed sensing with 3D lorenz.
Biomed Eng online 2016;15(1):1–20.
[102] Sreedhanya AV, Soman KP. Ensuring security to the compressed sensing
data using a steganographic approach. Bonfring Int J Adv Image Process
2013;3(1):1–7.
[103] Wang L, Li L, Li J, Li J, Gupta BB, Liu X. Compressive sensing of medical
images with confidentially homomorphic aggregations. IEEE Internet Things J
2018;6(2):1402–9.
[104] Zhang L, Zhu Z, Yang B, Liu W, Zhu H, Zou M. Medical image encryption
and compression scheme using compressive sensing and pixel swapping based
permutation approach. Math Probl Eng 2015;2015:1–9.
[105] Wang K, Wu X, Gao T. Double color images compression-encryption via
compressive sensing. Neural Comput Appl 2021;33(19):12755–76.
[106] Wang X, Liu C, Jiang D. A novel visually meaningful image encryption
algorithm based on parallel compressive sensing and adaptive embedding.
Expert Syst Appl 2022;209:118426.
[107] Wang X, Liu C, Jiang D. An efficient double-image encryption and hiding
algorithm using a newly designed chaotic system and parallel compressive
sensing. Inform Sci 2022;610:300–25.
[108] Xiao D, Zhao A, Li F. Robust watermarking scheme for encrypted images based
on scrambling and Kronecker compressed sensing. IEEE Signal Process Lett
2022;29:484–8.
[109] Yan F, Iliyasu AM, Venegas-Andraca S. A survey of quantum image
representations. Quantum Inf Process 2015;15(1):1–35.
[110] Yan F, Li N, Hirota K. QHSL: A quantum hue, saturation, and lightness color
model. Inform Sci 2021;577:196–213.
[111] Koshka Y, Novotny M. Comparison of use of a 2000 qubit D-wave quantum
annealer and MCMC for samplingm, image reconstruction, and classification.
IEEE Trans Emerg Top Comput Intell 2018;5(1):119–29.
[112] Alexander T, Kanazawa N, Egger D, Capelluto L, Wood C, Javadi-Abhari A,
McKay D. Qiskit pulse: Programming quantum computers through the cloud
with pulses. Quantum Sci Technol 2020;5(4):044006.
[113] Bergholm V, Izaac J, Schuld M, Gogolin C, Alam M, Ahmed S, Killoran N.
Pennylane: Automatic differentiation of hybrid quantum–classical computations.
2018, arXiv:1811.04968.
[114] Bodkhe U, Tanwar S, Parekh K, Khanpara P, Tyagi S, Kumar N,
Alazab M. Blockchain for industry 4.0: A comprehensive review. IEEE Access
2020;8:79764–800.
[115] Li B, Chenli C, Xu X, Jung T, Shi Y. Exploiting computation power of blockchain
for biomedical image segmentation. In: Proceedings of the IEEE/CVF conference
on computer vision and pattern recognition workshops. 2019, p. 4321–30.
[116] Donoho DL. Compressed sensing. IEEE Trans Inform Theory 2006;52(4):1289–
306.
[117] Huang R, Sakurai K. A robust and compression-combined digital image encryp-
tion method based on compressive sensing. In: Seventh international conference
on intelligent information hiding and multimedia signal processing. 2011, p.
105–8.
15
Journal of Information Security and Applications 78 (2023) 103621
[118] Li X, Kim S. Optical 3D watermark based digital image watermarking for
telemedicine. Opt Lasers Eng 2013;51(12):1310–20.
[119] Zhou Z, Siddiquee M, Tajbakhsh N, Liang J. Unet++: A nested U-net archi-
tecture for medical image segmentation. In: Deep Learning in Medical Image
Analysis and Multimodal Learning for Clinical Decision Support. 2018, p. 3–11.
[120] Kumar S, Amutha R. Edge detection of angiogram images using the classical
image processing techniques. In: Proceedings of IEEE-international conference
on advances in engineering, science and management. 2012, p. 55–60.
[121] Zanaty E, Abdelhafiz W. A performance study of classical techniques for medical
image segmentation. Int J Inform Med Data Process 2016;1(2):33–47.
[122] Zhou Y, Fang Y, Zhang Y. Securing wireless sensor networks: A survey. IEEE
Commun Surv Tutor 2008;10(3):6–28.
[123] Pittaluga F, Zivkovic A, Koppal S. Sensor-level privacy for thermal cameras. In:
Proceedings of IEEE international conference on computational photography.
2016, p. 1–12.
[124] Winkler T, Rinner B. Sensor-level security and privacy protection by embedding
video content analysis. In: Proceedings of international conference on digital
signal processing. 2013, p. 1–6.
[125] Pittaluga F, Koppal S. Pre-capture privacy for small vision sensors. IEEE Trans
Pattern Anal Mach Intell 2016;39(11):2215–26.
[126] Prabhu N, Naik D, Anwar F. Trusted video streaming on edge devices. In:
IEEE international conference on pervasive computing and communications
workshops and other affiliated events. 2021, p. 655–60.
[127] Du M, Liu N, Hu X. Techniques for interpretable machine learning. Commun
ACM 2019;63(1):68–77.
[128] Poursabzi-Sangdeh F, Goldstein D, Hofman J, Vaughan J, Wallach H. Manipulat-
ing and measuring model interpretability. In: Proceedings of the CHI conference
on human factors in computing systems. 2021, p. 1–52.
[129] Adebayo J, Gilmer J, Muelly M, Goodfellow I, Hardt M, Kim B. Sanity checks for
saliency maps. In: Proceedings of the 32nd international conference on neural
information processing systems. 2018, p. 9525–36.
[130] Nielsen M, Chuang I. Quantum computation and quantum information.
Cambridge: Cambridge University Press; 2000.
[131] Yan F, Iliyasu AM, Le P. Quantum image processing: A review of advances in
its security technologies. Int J Quantum Inf 2017;15(03):1730001.
[132] Almudever C, Lao L, Fu X, Khammassi N, Ashraf I, Bertels K. The engineering
challenges in quantum computing. In: Design, automation and test in europe
conference and exhibition. 2017, p. 836–45.
[133] Zheng Z, Xie S, Dai H, Chen X, Wang H. Blockchain challenges and
opportunities: A survey. Int J Web Grid Serv 2018;14(4):352–75.
[134] McGhin T, Choo KKR, Liu CZ, He D. Blockchain in healthcare applications:
Research challenges and opportunities. J Netw Comput Appl 2019;135:62–75.
[135] Novamizant L, Ramatryana INA, Magdalena R, Pratama IPAE, Rahim T, Shin SY.
Compressive sampling of color retinal image using spread spectrum Fourier
sampling and total variant. IEEE Access 2022;10:42198–207.
[136] Zhang Z, Rao B, Jung T. Compressed sensing for energy-efficient wireless
telemonitoring: Challenges and opportunities. In: Proceedings of Asilomar
conference on signals, systems and computers. 2013, p. 80–5.