Journal Pre-proof

The Applications of Machine Learning Techniques in Medical Data

Processing based on Distributed Computing and the Internet of
Things

Sarina Aminizadeh , Arash Heidari , Shiva Toumaj ,

Mehdi Darbandi , Nima Jafari Navimipour , Mahsa Rezaei ,
Samira Talebi , Poupak Azad , Mehmet Unal

PII: S0169-2607(23)00411-X
DOI: [Link]
Reference: COMM 107745

To appear in: Computer Methods and Programs in Biomedicine

Received date: 31 March 2023

Revised date: 15 July 2023
Accepted date: 2 August 2023

Please cite this article as: Sarina Aminizadeh , Arash Heidari , Shiva Toumaj , Mehdi Darbandi ,
Nima Jafari Navimipour , Mahsa Rezaei , Samira Talebi , Poupak Azad , Mehmet Unal , The Ap-
plications of Machine Learning Techniques in Medical Data Processing based on Distributed Com-
puting and the Internet of Things, Computer Methods and Programs in Biomedicine (2023), doi:
[Link]

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.

© 2023 Elsevier B.V. All rights reserved.

Highlights

 Providing a comprehensive analysis of the most current innovations in medical data

processing;

 Proposing a systematic review of the available platforms for medical data processing;

 Providing an overview of the most basic ML/DL-based methodologies in medical data

processing;

 Presenting a summary of distributed computing methods for health-care data analysis with
classifying them based on practical characteristics of the techniques;

 Evaluating each method that is associated with numerous aspects such as advantages,
challenges, databases, implementations, privacy, and security matters;

 Outlining the vital aspects where the preceding strategies may be improved soon.
 The Applications of Machine Learning Techniques in Medical Data
Processing based on Distributed Computing and the Internet of Things
Sarina Aminizadeh1, Arash Heidari2*, Shiva Toumaj3, Mehdi Darbandi4, Nima Jafari Navimipour 5,10*, Mahsa Rezaei6, Samira Talebi7,
Poupak Azad8, Mehmet Unal9
1
Medical Faculty of Islamic Azad University of Tabriz, Tabriz, Iran
2
Department of Computer Engineering, Tabriz Branch, Islamic Azad University, Tabriz, Iran
3
Urmia University of Medical Sciences, Urmia, Iran
4
Department of Electrical and Electronic Engineering, Eastern Mediterranean University, Gazimagusa 99628, Turkiye
5
Department of Computer Engineering, Kadir Has University, Istanbul, Turkiye
6
Tabriz University of Medical Sciences, Faculty of Surgery, Tabriz, Iran
7
Department of Computer Science, University of Texas at San Antonio, TX, USA
8
Department of Computer Science, University of Manitoba, Winnipeg, Canada
9
Department of Computer Engineering, Nisantasi University, Istanbul, Turkiye
10
Future Technology Research Center, National Yunlin University of Science and Technology, Douliou, Yunlin 64002, Taiwan

*Email: Arash_Heidari@[Link] (Arash Heidari), jnnima@[Link], [Link]@[Link] (Nima Jafari

Navimipour)

Abstract
Medical data processing has grown into a prominent topic in the latest decades with the primary goal of maintaining
patient data via new information technologies, including the Internet of Things (IoT) and sensor technologies, which
generate patient indexes in hospital data networks. Innovations like distributed computing, Machine Learning (ML),
blockchain, chatbots, wearables, and pattern recognition can adequately enable the collection and processing of
medical data for decision-making in the healthcare era. Particularly, to assist experts in the disease diagnostic
process, distributed computing is beneficial by digesting huge volumes of data swiftly and producing personalized
smart suggestions. On the other side, the current globe is confronting an outbreak of COVID-19, so an early
diagnosis technique is crucial to lowering the fatality rate. ML systems are beneficial in aiding radiologists in
examining the incredible amount of medical images. Nevertheless, they demand a huge quantity of training data that
must be unified for processing. Hence, developing Deep Learning (DL) confronts multiple issues, such as
conventional data collection, quality assurance, knowledge exchange, privacy preservation, administrative laws, and
ethical considerations. In this research, we intend to convey an inclusive analysis of the most recent studies in
distributed computing platform applications based on five categorized platforms, including cloud computing, edge,
fog, IoT, and hybrid platforms. So, we evaluated 27 articles regarding the usage of the proposed framework,
deployed methods, and applications, noting the advantages, drawbacks, and the applied dataset and screening the
security mechanism and the presence of the Transfer Learning (TL) method. As a result, it was proved that most
recent research (about 43%) used the IoT platform as the environment for the proposed architecture, and most of the
studies (about 46%) were done in 2021. In addition, the most popular utilized DL algorithm was the Convolutional
Neural Network (CNN), with a percentage of 19.4%. Hence, despite how technology changes, delivering appropriate
therapy for patients is the primary aim of healthcare-associated departments. Therefore, further studies are
recommended to develop more functional architectures based on DL and distributed environments and better
evaluate the present healthcare data analysis models.

Keywords: Medical Data Processing, Healthcare Data Analysis, Deep Learning, Distributed Computing

1. Introduction
Health and medicine are receiving tremendous attention worldwide, and many new techniques and
scientific reports are being implemented [1-3]. In recent decades, distributed computing, signal
processing, and healthcare have been the main topics of utmost interest [4-6]. These two subjects have
been integrated, and researchers have already been developing methods to seek beneficial novel trends in
data supplied by medical equipment and utilized in training courses and examinations [7, 8]. The results
of medical measurements, experiments, and inspections would ultimately be incorporated into models
using novel methods [9, 10]. The aim of healthcare systems, which is to restore patients' health, involves
the application of effective medical data enabling evidence-based intervention [11-13]. Due to records
management, compliance and regulatory obligations, and patient care, the healthcare industry produces a
lot of data [14, 15]. Also, the current tendency is for these massive volumes of data to be rapidly
digitalized and assembled with data gathered from individual and mobility devices [16-18]. In terms of
patient information, treatment options, pharmaceutical details, or other aspects of healthcare, everything
must be done fast, correctly, and, in some situations, transparently enough to meet severe industry rules
[19, 20]. Medical data processing is one of the essential duties of Healthcare Providers (HCP) [21].
Computerized physician order input apps featuring decision support fields prevent excess medical faults
using integrated memory help [22]. These automated notification alarm signals allow suitable and prompt
response that provides better and more efficient healthcare coverage [23]. The design principles of
computer technologies must follow pre-stated rules and principles to preserve secrecy [24]. So, user-
friendly technology provides the effective and timely transmission of healthcare data for excellent patient
care fulfilling the demands of the individuals and the institution [25].
Gathering data without any distortion is costly and time-consuming [26]. A vast volume of medical
data is now available because of recording devices and technological advancements. These medical
records are chaotic and must be eliminated to obtain an accurate diagnosis [27]. The use of Information
Technologies (IT) for medicine, including Artificial Intelligence (AI), has recently increased; meanwhile,
all versions of such systems are inadequate, lacking actual data, particularly register-based models [27].
In general, AI has already been acknowledged as a valuable strategy for assistance in various medical
concepts, including the diagnosis of diseases, management and remote monitoring of patients, and
effective treatment options [28-30]. Machine Learning (ML) is one of the most frequently employed AI
methodologies, corresponding to the intelligence revealed by computers [31]. Deep Learning (DL) is an
improved and scalable ML extension that aids in enhancing the design and use of learning algorithms [32-
34]. DL, a subtype of ML, has been used frequently to build intricate models using enormous amounts of
data and in a streamlined setting [35, 36]. The cloud platform has gradually become one of the most
popular topics in computer science [37]. High-Performance Computing (HPC), virtualization, and utility
computing are some of the various research fields that cloud computing is centered on [38]. Also, edge
computing provides the capacity to deal with concerns, including reaction time, power consumption,
bandwidth cost reduction, data security, and privacy [39]. Plus, fog computing covers the accompanying
topics by presenting elastic services and resources to target customers at the network's edge, while cloud
computing focuses on distributing resources over the core network [40]. Besides, Quality of Service
(QoS) is required to support the operation of any healthcare system by aiding in the efficient and
successful management of work procedures [41]. So, tasks would be disrupted if QoS is not implemented
appropriately [42]. As a result, previous studies have stated that innovative ML techniques combined with
powerful distributed cloud computing platforms assist in the exploration of healthcare and medical big
data [43-45].
Medical data processing components, such as image processing methods, drug discovery, Facial
Emotion Recognition (FER), survival analysis, forecasting, and so on, have earned great interest in the
healthcare setting. As a result, these components were also appropriate for healthcare and medical
research. However, there is no comprehensive study of the utilization of distributed computing systems in
medical processing. This study concentrates on prioritizing the advancements, recognition, and
classification of healthcare challenges, including cancers or epidemic situations such as COVID-19,
healthcare data processing, and medical image processing methods to address medical issues, and so on,
to present a comprehensive description of the modern devices supplied to use these innovations.
Consistent with previous reviews on many medical data processing applications recently produced for
identifying and treating various disorders, this study makes a substantial contribution by targeting the
most promising field of research. This study employs a Systematic Literature Review (SLR) [46, 47] to
identify, analyze, and combine results from relevant investigations. We also classify medical data
processing platforms used for healthcare delivery into five categories: Internet of Things (IoT), cloud,
edge, fog, and hybrid platforms. Hybrid platforms integrate some platforms like IoT, edge, fog, Internet
of Drones (IoD), or Internet of Vehicles (IoV). We evaluated numerous features such as benefits,
challenges, datasets, usage, methods, and security for each classification associated with medical data
processing methods. This paper describes the methodologies and implementations of data processing in
the outlet of healthcare. We have also looked over future scopes, in particular, highlighting all of the gaps
that need to be handled. In general, the expenditures of the present paper are:

 Providing a thorough analysis of the most current innovations in medical data processing;

 Proposing a systematic review of the available platforms for medical data processing;

 Providing a summary of the most basic ML/DL-based methodologies in medical data

processing;

 Presenting a summary of distributed computing methods for healthcare data analysis by

classifying them based on practical characteristics of the techniques;

 Evaluating each method that is associated with numerous aspects such as advantages,
challenges, databases, implementations, privacy, and security matters;

 Outlining the vital aspects where the preceding strategies may be improved soon.
The study's layout is defined by the classifications mentioned below. The following section discusses
the principles of distributed computing and the terminology of applied platforms in medical data
processing. The relevant reviews are presented in Section 3. The research methodologies for study
materials and inclusion criteria are discussed in Section 4. The classifications of the selected articles are
explained in Section 5. The results and comparisons, explaining platforms, methods, dataset, privacy, and
security with a comprehensive analysis of the results, are presented in Section 6. Section 7 discusses the
topics and difficulties that are still unresolved. Finally, a brief conclusion of the study with
recommendations for further research is provided in Section 8. Table 1 illustrates the abbreviation applied
within this paper.
Table 1. Abbreviation table.

Abbreviation Definition Abbreviation Definition Abbreviation Definition

ADE Adverse Drug Effect FDA Federated Deep Learning MLPs Multilayer Perceptrons
AI Artificial Intelligence FER Facial Emotion Recognition MSE Mean Squared Error
ANN Artificial Neural Network FL Federated Learning NFV Network Function
Virtualization
AUC Area Under Curve FIS Fuzzy Inference System NLP Natural Language Processing
BERT Bidirectional Encoder Representations from GANs Generative Adversarial NN Neural Network
Transformers Networks
CAD Computer Aided Diagnosis HCP Health-Care Provider QoS Quality of Service
CDSS Clinical Decision Support Systems HEM Hybrid Ensemble Model RBFNs Radial Basis Function Networks
COVID-19 Coronavirus Disease 2019 HIoT Healthcare Internet of RBMs Restricted Boltzmann Machines
Things
CNNs Convolutional Neural Networks HPC High-Performance RL Reinforcement Learning
Computing
 CKD Chronic Kidney Disease ICT Information and RNNs Recurrent Neural Networks
Communication
Technologies
CPS Cyber-Physical Systems ICU Intensive Care Unit ROC Receiver Operating
Characteristic
CVS Cross-Validation Score IJP Inkjet-Printed RQs Research Questions

CTA CT-Angiography IoMT Internet of Medical Things SDCA Summation Discriminant

Correlation Analysis
DBNs Deep Belief Networks IoT Internet of Things SDN Software-Defined Network
DDTL Distant Domain Transfer Learning IoV Internet of Vehicles SLA Selective Learning Algorithm
DFF Distant Feature Fusion IT Information Technologies SLR Systematic Literature Review
DICOM Digital Imaging and Communications in LDP Local Differentially Private SOMs Self-Organizing Maps
Medicine
DWT-PCA Discrete Wavelet Transform and Principal LSTMs Long Short-Term Memory SVM Support Vector Machine
Component Analysis Networks
DL Deep Learning MAS Multi-Agent System TL Transfer Learning
EA Epileptiform Activity MEC Mobile Edge Computing US Ultra-Sound
EDA Exploratory Data Analysis ML Machine Learning WHO World Health Organization

2. Terminologies and Fundamental Concepts

This section addresses the primary concepts of data processing techniques in the healthcare era regarding
the association with platforms of ML and distributed computing.
2.1. Machine learning and deep learning methods
Integrating AI in the medical sector has been an increasingly prominent topic [48]. ML is a subtype of AI,
defined as a branch of science focusing on how computers learn from data [49, 50]. It emerges at the
intersection of statistics, which aims to discover connections between data, and computer science,
emphasizing efficient computing processes [51]. It is possible to establish complicated relationships
utilizing ML that could be difficult to represent using an equation [52]. For instance, Neural Networks
(NN) similarly show the data on the function of the human brain by integrating a vast number of neurons
[53]. This allows ML systems to address the issue similarly to one used by a physician, achieving justified
outcomes by investigating the facts. One of the key components of ML is the requirement to supply the
algorithm with a large amount of data to progress on the investigation. There are diverse groups of
patients followed retrospectively, but they generally contain hundreds or a few thousands of individuals.
However, ML systems could deal with massive datasets [54]. Breakthroughs in DL and ML will aid in
making healthcare decisions but will not displace the physician entirely. With the assistance of ML, the
introduction of patient records optimized the outcomes of statistical regression significantly, providing a
perspective for individualized therapy. Human mistakes in healthcare are connected with significant
financial losses, most of which can be eliminated with the assistance of ML and DL. Previous studies
reported reduced mortality rate and lower hospitalization in an Intensive Care Unit (ICU) after treatments
were administered in an ML model [55].
When designed, assessed, and distributed correctly, DL approaches can aid in acquiring, interpreting,
and combining healthcare information from diverse sources and bringing it to our fingertips—as if we had
an experienced subspecialist to rely on for every patient and medical circumstance. Also, DL could aid in
remote monitoring and analysis of patients, leading the data collection that can be forwarded to a
healthcare facility. Perhaps ML algorithms can assist in discovering novel and subtle data trends that
could influence patient care [56]. So, ML is applied in several contexts, such as deepfake detection, image
classification, spam detection, multimedia concept retrieval, video recommendation, and text mining.
Among numerous AI techniques, DL is extremely widely implemented in these apps. The unforeseen
expansion in data collection capabilities and the significant advancements made in hardware advances,
such as High-Performance Computing (HPC), are the fundamental causes of the ongoing birth of new
research in deep learning and distributed learning [57]. Also, the effectiveness of DL in several pattern
recognition chores has created enthusiasm and a strong perception that DL will bring revolutionary
advances in healthcare. With the development of DL in many ML applications in recent years, there are
high expectations that DL can bring advancements in Computer Aided Diagnosis (CAD) performance and
the prevalent use of DL-based CAD or AI in various procedures in patient care [58]. It is essential to
express the emerging prominence of ML approaches in incorporating health data [59].
2.2. Medical data processing in general
In recent decades, QoS and computation of data processing connected to healthcare have become a major
task in numerous research disciplines, for example, ML, AI, and DL utilizing medical data and
processing records stored in medical centers [60, 61]. The three primary data streams from clinical trials,
records connected to medical research, and organizational information procedures drive the collection of
data ledgers. Medical data acquisition and pre-processing design setting is a significant priority in
healthcare [62, 63]. Assessing and analyzing this recorded information for computer-based executives
helps, and the emergence of real-time platforms is now a major trend in the latest smart healthcare.
Medical data analysis has progressed rapidly due to improvements in ML and AI technology [64]. The
healthcare industry creates a vast quantity of data in the event of record-keeping, regulatory compliance,
and patient care [65]. The current tendency is for these large volumes of data to be progressively
integrated and combined with information obtained from individual and mobility sensors. Efficient data
processing and investigation may discover hidden insights for improving patient care, especially with
large datasets. They can further properly estimate a patient’s health condition if the data is obtained from
outside medical clinics. The findings might also be applied to better, less expensive real-time healthcare
[9].
In addition, the government and authorized healthcare centers are primarily responsible for data
collection. Knowledge application, which is the purpose and motivating factor behind data processing,
has become largely involved in clinical management and treatment program termination. Categorization,
clustering, association rules, and regression are just a few of the data mining technologies easily
accessible. Researchers can only make a decision and build a prediction model after carefully considering
the dataset. In understanding the performance of the model, experts must perform certain tests on it.
Furthermore, the patterns and data discovered must be examined and improved. Consequently, data
processing is an ongoing dynamic process that needs corrective input. Only in this manner will we be able
to get a comparatively superior knowledge model [66].
2.3. What is distributed computing?
Cloud computing is the strategy of storing and processing data using a network of remote computers on
the Internet or, rather, a personal computer or local server. Also, IoT is a platform in which objects are
implanted with a network connection, enabling them to collect and share data without the need for human
interaction [67-70]. Moreover, edge and fog platforms are novel storage and data analysis paradigms
providing an extra layer of processing capability between the cloud and a device, such as a medical
sensor. The additional layer is close to the data source. Such structures reduce the amount of data that
must be transported to and received from cloud-based servers and increase the accessibility of data that
must be acted upon within milliseconds of collection. To the degree that specific projects call for speedy
data analysis, fog can help emerge paradigms in healthcare data analysis and remote medical monitoring,
such as processing massive data and privacy-conscious distributed learning systems [71]. In addition to
data collection, an edge medical device serves as a gateway to a network or the Internet. The edge
computing paradigm shifts network service administration from central nodes (known as the core) to the
opposite extreme, which is the sensor component, rather than to nodes. Fog and edge are beneficial when
the aim is immediate data processing following the instantaneous creation of a rules-based command.
These two designs may enable medical sensor data to be placed quickly, resulting in efficient conclusions.
The following are three possible obstacles for distributed data analysis using edge or fog platforms with
limited computational power: (1) converting medical data from multiple sources to a popular architecture
while also protecting privacy if data are to be shared; (2) attempting to balance abstract data to allow for
restricted storage device against lesser abstraction to allow for productivity; and (3) recognizing
inaccurate data from separated deficient sensors or wireless transmitters [39, 72]. In the next section, we
will briefly analyze the relevant reviews on this topic, mostly regarding the use of IoT and the Internet of
Medical Things (IoMT) in healthcare data, DL applications in patient care, and virtual patient monitoring.

3. Relevant Reviews
In the previous section, we discussed concepts of distributed computing, ML, and DL with applications of
them deployed in medical data processing. In this section, we will have a brief review of the related works
in this field. Generally, ML and, specifically, DL techniques dominate medical image and data analysis.
In a related study, Sarhan, et al. [73] provided an overview of ML techniques recommended for
diagnosing, classification, and segmentation of multiple lesions in ophthalmic diseases, primarily diabetic
retinopathy, age-related macular degeneration, and glaucoma, as well as a brief description of public
datasets and issues associated upon each pathology and the strategy development. Color fundus imaging
and optical coherence tomography were identified as the most frequently utilized imaging modalities.
Also, Kaur and Kaur [74] presented a comprehensive understanding of DL and its implications in
healthcare obtained during the last decade. They focused on the DL image processing capabilities used to
defeat COVID-19 and an in-depth analysis of the recent structure built using this technology. Moreover,
et al. [57] evaluated the methodologies associated with the study of the head and brain from the fetal
Ultra-Sound (US) perspective. The investigated methods are further classified into five broad groups
depending on their basic theoretical approach, as described in Global intensity-based, learning-based,
deformable, registration, and active shape models. Also, the best image-processing approaches for
analyzing the head and brain were determined. The assessment of the reviewed methods enabled the
identification of the essential techniques for each sector, their benefits, and the ongoing value of the
investigation.
Kashani, et al. [75] intended to identify systematically and taxonomically categorize Healthcare
Internet of Things (HIoT) research to provide an extensive comparison with the existing publications'
limits and potentials by providing a comprehensive overview of the latest studies in HIoT that concentrate
on applied methodologies, methods, and tools. The reviewed articles were classified into five categories:
communication-based, security-based, resource-based, application-based, and sensor-based. The
advantages and drawbacks of the proposed techniques and a complete comparison of assessment
procedures, tools, and metrics were also discussed. In addition, Gatouillat, et al. [76] provided a detailed
literature evaluation of the current efforts focused on strengthening the IoMT via the application of formal
approaches supplied by the Cyber-Physical Systems (CPS) community by explaining the realistic use of
the democratization of medical supplies for both healthcare professionals and patients. They also tried to
propose undiscovered research pathways and possible developments to address unknown research
difficulties. The domain of IoMT-based systems and IoMT devices was investigated from a multi-layer
perspective.
Also, Mbunge, et al. [77] discussed the dynamic sensory web in the major trends, competencies,
innovations, apparent issues, and ethical concerns, as well as emerging technologies, including AI, IoT,
IoMT, big data, and cloud, which are critical in the progression and integration of sensory emotive Web
in virtual healthcare. They provided a summary of the expansion of the WWW in healthcare,
emphasizing capacities and technology even while explaining security and ethical concerns. Finally, they
presented a variety of solutions to the limitations and ethical concerns.
Ben-Israel, et al. [78] aimed to discover the use of AI methods in clinical settings and their capacity to
build upon the tremendous human intellect that has brought medicine to where it is now. They
emphasized the present status of ML studies in clinical medicine by characterizing the target patient
demographics, types of employed data, and the existence or absence of DL techniques. Also, Heidari, et
al. [79] concentrated on emphasizing the achievements, recognition, and identification of DL and medical
image processing approaches to combat the COVID-19 pandemic, mortality forecast, predicting health
equipment, and so on by creating an SLR to explore, interpret, and integrate data from comparable
research. In their study, the DL techniques applied in COVID-19 were divided into seven basic separate
categories. Several features like benefits, difficulties, database, utilizations, safety, and TL for each
classification and approach using DL methods were reviewed in this study.
This survey paper distinguishes itself from previous studies by comprehensively analyzing distributed
computing systems in medical data processing. It explores the utilization of various platforms such as
IoT, cloud, edge, fog, and hybrid platforms and evaluates them based on practical characteristics. The
survey also emphasizes the integration of ML/DL methodologies in medical data processing and provides
a systematic review of the available techniques. It evaluates each methodology from different
perspectives, including advantages, challenges, implementations, and privacy/security considerations.
Additionally, the paper identifies gaps and limitations in existing strategies, highlighting areas for future
improvements. Overall, this survey provides a valuable resource for researchers and practitioners in the
field, offering a comprehensive overview and identifying avenues for further research and development.
A summary of relevant reviews is provided in Table 2. In the next section, we are going to discuss
research methodologies thoroughly.

Researchers Idea Scope Advantages Challenges

Sarhan, et Reviewing most recent impactful Data Pointing to limitations and strengths The papers published before
al. [73] papers about ML methods in processing 2015 were not included
Ophthalmology.
Kaur and Illustrating the concept of DL Image Introducing some issues and Future works are not
Kaur [74] techniques using medical imaging. processing confrontations to control the health thoroughly discussed.
using DL crisis and outbreaks
Torres, et Reviewing the state-of-the-art of Image It was the first review that thoroughly Future works are not
al. [80] image-based approaches to assess processing investigated the fetal head and brain in thoroughly discussed.
fetal head/brain in the US. methods the US.
Kashani, et Attempting to establish, compare, IoT in The results are presented in great Future works are not
al. [75] and categorize available data in healthcare detail. thoroughly discussed.
HIoT systems.
Gatouillat, Describing the progression of IoMT IoMT Mentioned potential research paths There are still some
et al. [76] actions over time. and issues of relevance unexplained challenges
Mbunge, et Discussing emotive sensory Web in Virtual Outlined capabilities and innovations, The procedure for selecting
al. [77] virtual healthcare. healthcare privacy, and ethical aspects articles is not obvious
Ben-Israel, Highlighting the current status of ML on patient The results are presented in great The procedure for
et al. [78] ML research in clinical practice. care detail. selecting articles is not
obvious.
Heidari, et Reviewing the DL apps created for DL The results are presented in great No solution was
al. [79] the diagnosis and cure of COVID-19 detail recommended for ethical
disease. and technological issues
Table 2. Summary of related works.

4. Research Methodology
In the previous section, we provided a summary of the related reviews of DL-based techniques in medical
data processing. In this section, we will give the details of the methodologies employed in this study.
Also, the SLR approach is used in this section to clarify how medical data processing is done [81]. The
SLR is a critical review and investigation of all studies on a specified subject. This section concludes a
thorough examination of the performance of each distributed computing platform in the analysis of
medical data. Following that, we examine the validity of the study selection approaches. The study topics
and selection criteria are covered in detail in the following subsections, along with the search process.
4.1 Formalization of Questions
The basic targets of this research are to categorize, distinguish, and assess all related articles discovered in
medical data processing methodologies. To meet the above-stated objectives, the elements and features of
the approaches can be thoroughly studied utilizing an SLR. An additional goal of SLR is to understand
the significant concerns and problems facing this industry. A few Research Questions (RQs) that have
been formulated are as follows:
 RQ 1: What are the main issues and unanswered questions that need to be resolved in this area?

The answers to this issue will be offered in Section 5, and the unresolved issues will be presented
in Section 7.
 RQ 2: What steps should we take to find the article and choose the best data processing techniques
for the healthcare setting?
This is addressed in Sections 6 and 7.

 RQ 3: How can distributed computing mechanisms be classified in medical healthcare? What are
some of their examples?
The response to this issue is found in Section 6.

 RQ 4: What techniques are employed by the researchers to conduct their study?

This question is answered in Sections 5.1 to 5.5.

4.2 The selection procedure for articles

The technique used to find and choose the papers for this investigation consists of the following five
steps. This method is shown in Fig. 1. Table 3 displays the search phrases and keywords used to index the
articles in the first phase. These publications are the results of a search of widely used electronic
databases. Applied electronic databases include Scopus, IEEE Explore, Springer Link, Google Scholar,
Elsevier, Nature, JMIR, Soft Computing, and Computer Communications. Moreover, journals, conference
papers, books, chapters, notes, technical studies, and special issues are excluded. Stage 1 yielded 1608
papers. The publisher's distribution is illustrated in Fig. 2.

Fig. 1. The stages of the search procedure and article selection.

 Table 3. Search terms and keywords.

S# Search Terms and Keywords

S1 ―Medical data processing‖ or ―Healthcare data processing‖
S2 ―Medical data processing‖ or ―Medical data analysis‖
S3 ―Machine learning‖ or ―Deep learning‖
S4 ―Medical data processing‖ and ―Cloud computing‖
S5 ―Medical data processing‖ and ―Internet of things‖
S6 ―Medical data processing‖ and ―Edge computing‖
S7 ―Medical data processing‖ and ―Fog computing‖
S8 ―Medical data processing‖ and ―Hybrid techniques‖

Fig. 2. The distribution of papers based on the publisher. Fig. 3. The criteria for selecting papers to be included in the process.

Fig. 4. Overview of the type of publisher in the reviewed papers.

Fig. 5. The publisher`s distribution of the selected papers. Fig. 6. Journal-based distribution of the chosen articles.

In stage 2, duplicated papers are omitted. Next, in stage 3, based on the inclusion criteria, the selection
of articles is continued, with which 962 papers remain. The distribution of papers regarding the type is
represented in Fig. 4. Stage 4 eliminates the review articles; of the remaining 962 items in Stage 2, 493
(or 51.2%) were review papers. IEEE and Elsevier collectively publish 34% of all scientific articles.
BioMed and IEEE publish 18.8% of all review papers. There are now 469 articles remaining, and Fig. 3
shows how they are distributed by publication and article kind at this time. Stage 5 involved reviewing
the papers' titles and abstracts. At this stage, 249 publications have been selected for further study, with
the distribution of the papers depending on the publisher shown in Fig 4. The papers' methodology,
evaluation, discussion, and conclusion have also been examined to verify their relevance to the research.
The remaining publications in Table 4 were reviewed and examined using 27 papers that satisfied the
high requirements. In Fig. 5, the distribution of the chosen articles by their publishers is also depicted.
Elsevier and Springer publish most selected articles (36%, 13 articles). The lowest number is related to
The Optical Society, Frontiers Media S.A., Academic Press Inc., Alexandria University, and Tech
Science Press (18%, 5 articles). The year with the most papers published (46%) is 2021, whereas the year
with the fewest articles published (11%) is 2019. In Fig. 6, the journals that publish the publications are
shown. Journal of Medical Internet Research, IEEE IoT Journal, and Informatics in Medicine Unlocked
publish the most articles (22%, 6 articles).
Table 4. Details of the selected articles (updated on March 15, 2023).

Publisher Author Citation JCR Scopus Journal Name H-

index
Journal of Medical Internet Research Lee, et al. [82] 2 Q1 Q1 Journal of Medical Internet Research 142
Springer New York Egger, et al. [83] 0 Q1 Q1 Journal of Digital Imaging 58
Frontiers Media S.A. Stirling, et al. [84] 5 Q2 Q2 Frontiers in Neurology 67
Institute of Electrical and Electronics Engineers Aazam, et al. [85] 1 Q1 Q1 IEEE Transactions on Sustainable Computing 21
Inc.
Springer Netherlands Liu, et al. [86] 0 Q2 Q2 Journal of Supercomputing 68
Institute of Electrical and Electronics Engineers Ma and Pang [87] 22 Q2 Q1 IEEE Access 158
Inc.
Institute of Electrical and Electronics Engineers Niu, et al. [88] 11 Q1 Q1 IEEE Journal of Biomedical and Health 125
Inc. Informatics
Elsevier Ltd. Połap, et al. [89] 24 Q2 Q2 Journal of Information Security and Applications 40
Springer London Khan, et al. [90] 13 Q1 Q1 Neural Computing and Applications 80
Institute of Electrical and Electronics Engineers Mahawaga Arachchige, et al. 38 Q1 Q1 IEEE Internet of Things Journal 97
Inc. [91]
Springer Verlag Mansour and Althobaiti [92] 0 Q2 Q2 Soft Computing 90
Nature Warnat-Herresthal, et al. [93] 31 Q1 Q1 Nature Publishing Group 1226
Institute of Electrical and Electronics Engineers Kong, et al. [94] 16 Q1 Q1 IEEE Internet of Things Journal 97
Inc.
Alexandria University Yang, et al. [95] 2 Q1 Q1 AEJ - Alexandria Engineering Journal 58
Elsevier BV Rahman, et al. [96] 0 Q2 Q2 Smart Health 9
Tech Science Press Khan, et al. [97] 14 Q2 Q1 Computers, Materials and Continua 40
Academic Press Inc. Manocha, et al. [98] 4 Q1 Q1 Journal of Biomedical Informatics 103
Public Library of Science Shukla, et al. [99] 28 Q1 Q1 PLoS ONE 332
Elsevier Horng, et al. [100] 0 Q1 Q1 Computer Communications 105
Elsevier BV Mohamed Akram, et al. [101] 0 Q2 Q1 Biomedical Signal Processing and Control 84

Elsevier Ltd. Singh, et al. [102] 2 - Q2 Informatics in Medicine Unlocked 32

Elsevier Ltd. Fan, et al. [103] 35 Q1 Q1 Information Processing and Management 101
Elsevier Arabi and Zaidi [104] 17 Q1 Q1 Medical Image Analysis 135
The Optical Society Yuan, et al. [105] 12 Q1 Q1 Biomedical Optics Express 86
Elsevier Ireland Ltd Al-Saffar and Yildirim [106] 11 Q1 Q1 Computer Methods and Programs in Biomedicine 102

Elsevier Ltd. Momenzadeh, et al. [107] 0 - Q2 Informatics in Medicine Unlocked 32

Springer Netherlands Seba and Benifa [108] 0 Q3 Q2 Wireless Personal Communications 65

5. Medical Data Processing Methods

In section 4, we talked about the methodologies of this study based on the SLR method. This part will
review the techniques applied in medical data processing for healthcare delivery as classified into five
categories: IoT, cloud, fog, edge, and hybrid methods. Moreover, we take a brief review of some articles
published recently for each of the platforms. Fig. 7 illustrates the taxonomy of medical data processing
methods.

Fig. 7. The proposed medical data processing taxonomy separated five distinct methods

5.1 Cloud computing

For a variety of applications, cloud computing as a network approach provides diversity, reliability,
adaptability, inclusivity, flexibility, intention, unpredictability, scalability, dynamic nature, quick service
structure, lower capital costs, distribution, and resource pooling [109-111]. Cloud computing allows
content providers to get internet access for sharing flexible computer resources [112]. Additionally, it
offers various features for processing data on a widespread, decentralized basis [113]. A cloud
environment can offer clients a variety of services and computational resources, including databases,
networking, storage, and applications. [114]. While giving excellent cooperation opportunities, it offers
the organization, businesses, and individuals considerable financial benefits. So, the cloud-based platform
helps the medical and healthcare groups by facilitating equal access to clinical information. [115].
In a relevant study, Lee, et al. [82] created a DL-based Natural Language Processing (NLP) algorithm
for medical specialty categorization by gathering data that adequately represented patients' symptoms in a
real-world context. They built several DL-based NLP systems, compared their results, and then chose the
best model for an AI chatbot that was deployed on Google Cloud. First of all, data collection and
sanitation were done, and an Exploratory Data Analysis (EDA) was conducted to uncover the
interpretable aspects of the data before creating the DL–based NLP models. Four different Long Short-
Term Memory (LSTM) frameworks with or without recognition, with or without a pre-trained FastText
embedding layer, and bidirectional encoder ideas from NLP transformers were trained and evaluated
using a randomly chosen test data set. Precision, recall, F1-score, and Area Under Curve (AUC) were
utilized to assess the models' performance. An AI chatbot was created to make it simple for patients to use
this specialty referral system. The AI chatbot was made using the "Alpha" open-source framework.
In another study, Egger, et al. [83] introduced Studierfenster, a client-server platform for (bio-)
medical image analysis that is also free and open-source. It has several features, including the ability to
display medical data in popular web browsers. Other features include calculating standard medical scores,
manually outlining structures in medical images slice by slice, placing landmarks in medical imaging
data, visualizing medical data in Virtual Reality (VR), reconstructing faces, and registering medical data
for Augmented Reality (AR). Convolutional Neural Network (CNN)-based automated cranial implant
design, Generative Adversarial Network (GAN)-based in-painting of aortic dissections, and CNN-based
automatic aortic landmark recognition in CT-Angiography (CTA) pictures are more advanced features. To
assess Studierfenster's accessibility and practical functions, a user study was conducted with medical and
non-medical professionals in medical image processing. A mean of 6.3 out of 7.0 potential points was
obtained from participants who were asked to rate their overall opinion of Studierfenster in an ISO
standard (ISO-Norm) questionnaire.
Besides, Stirling, et al. [84] created a minimally invasive Electroencephalography (EEG) recording
device which is implanted under the scalp. An ML algorithm guided the identification of Epileptiform
Activity (EA) developed to identify significant EEG signals and predict the chance of a seizure hourly.
The approach comprised a Random Forest (RF) regressor with a Logistic Regression (LR) classifier. The
system's ability to identify five participants' Interictal Epileptiform Activity (EA) and seizures was
discussed. EA is characterized as both interictal and ictal epileptic activity, including electrographic
episodes and interictal discharges. Additionally, the recordings were evaluated qualitatively in contrast to
a reference-standard 7-day ambulatory video-EEG monitoring. The potential for seizure forecasting using
sub-scalp EEG was also shown through a case study that acted as a proof-of-concept for how cycles may
be produced from event detection systems in the EEG and how these cycles may be used to estimate
epileptic episodes. The ability of a sub-scalp device to precisely record ultra-long-term EEG (>12
months) and identify focal seizure activity was successfully shown in the end. All five subjects handled
the device well, and there were no major adverse effects, indicating that a discreet, minimally invasive
device might be used to monitor EA over an extended period continuously.
Aazam, Islam [64] provided a three-tier cloud-fog-IoT framework, a term created as the CoT model,
and a task offloading mathematical system for managing incoming tasks from an underlying IoT and
selecting whether to offload to fog, a cloud, or cooperatively to fog-cloud. As a proof-of-concept, the CoT
architecture was constructed based on datasets suited for IoT in general and DL approaches in particular.
The architecture of the model was assessed by taking into consideration computational workloads for
different applications in IoT linked to medical, multimedia, location-based, and text, and by utilizing
actual datasets, based on three scenarios: fog-only (F), cloud-only (C), and fog-cloud (F/C) computing
referred primarily from hereon as (F/FC/C). Their research examined the factors that may impact the
performance and power of F/C/FC settings. The outcome indicated how much a variable might affect the
speed and power usage, the drawbacks, and what might aid in reducing these parameters.
Additionally, to utilize distributed parallel classification methods in hospital intelligent guidance, Liu,
et al. [86] developed the Parallel Random Forest (PRF) classification technique. The main usage of the
proposed model was for the detection and classifying colon cancer. The model was designed based on the
Apache Spark cloud computing and with the Bilayer Parallel Training-CNN (BPT-CNN) platform
application. An Adaptive Domain Density Peak Clustering (ADDPC) method was put forth in light of
sparse cluster loss in data sets with variable density distribution. Additionally, a case study evaluation of
the model's performance was conducted. The outcomes showed that a distributed cloud computing
platform-based PRF algorithm might independently construct data-parallel tasks, maximizing the cost and
efficiency of data communication.
Finally, Ma and Pang [87] systematically evaluated the drawbacks of the present convolution Neural
Network (NN) algorithm and sports medical data to develop the CNN algorithm-based on the resampling
approach with a self-adjusting function. To analyze vast volumes of data more properly, they also
innovatively presented the supplementary model tensor convolution self-coding NN framework to realize
the analysis of multi-dimensional input of sports medicine. Therefore, a cloud-based fusion hardware-in-
the-loop simulation model was developed to advance the creation of an intelligent healthcare data
platform for sports medicine. This model underwent analysis and research to provide technical support
and experience for the actual cloud-based fusion hardware-in-the-loop system.
Cloud computing is a versatile and advantageous network approach used in various applications. It
provides benefits such as diversity, reliability, adaptability, flexibility, and scalability while reducing
capital costs. By offering access to flexible computer resources and decentralized data processing, cloud
computing enables content providers to share information and services effectively. In the medical and
healthcare sector, cloud-based platforms facilitate equal access to clinical information and provide
financial benefits. Several studies have demonstrated the practicality and benefits of cloud computing in
different domains. For example, researchers developed a DL-based NLP algorithm for medical specialty
categorization and deployed it on Google Cloud. Another study introduced a free and open-source
platform for (bio-) medical image analysis, which received positive ratings in a user study. A machine
learning-guided minimally invasive EEG recording device was created to predict seizures accurately.
Additionally, researchers proposed a three-tier cloud-fog-IoT framework for task offloading management
and developed a Parallel Random Forest classification technique for colon cancer detection. They also
evaluated the drawbacks of existing convolutional neural network algorithms and proposed a cloud-based
fusion hardware-in-the-loop model for intelligent healthcare data processing in sports medicine. These
advancements highlight the potential of cloud computing in revolutionizing medical data processing. The
cloud platforms and their characteristics are covered in Table 5 for medical data processing.
Table 5. The methods, properties, and features of cloud computing platforms in medical data processing

Researchers Idea Advantages Challenges Security? Dataset Using Method Usage?

TL?

Lee, et al. Developing an AI -High -High No A Korean web- No LSTM Giving

[82] chatbot with the accessibility complexity based platform medical
based NLP -Easy-to-use called HiDoc advice in
models. application the
-Low delay COVID-
19
pandemic
Egger, et al. Introducing a -High -High energy No Digital Imaging No CNN Image
[83] framework for accuracy consumption and +GAN analysis
medical image -Can be Communications
analysis. extended to in Medicine
other fields of
research
Stirling, et al. Presenting a DL -High -Significant No ACTRN No RF+LR Identificat
[84] method for sensitivity false 12619001587190 ion of
seizure detection. positives seizure
activity
Aazam, et al. Providing an -High -High delay No IoT jobs/tuples No Handling
[85] implementation of scalability, DNN of
the three-tier CoT -Low energy incoming
architecture. consumption, tasks
Liu, et al. Establishing a -High -High No UCI (UC Irvine No BPT- Detect
[86] reliable medical accuracy, complexity ML Repository) CNN and
data management -High -No solution classify
system. robustness, for the colon
-Low delay, defects. cancer
Ma and Pang Proposing a -Efficient -High Yes Not mentioned No CNN Sports
[87] cloud-based prediction complexity medical
hardware-in-the- -Lower risk data
loop simulation assessment analysis
model for
information
mining.

5.2 IoT platform

The IoT is a constantly growing ecosystem that combines hardware, apps, items, and computational
devices for data collection, transfer, and communication. The IoT offers a smooth platform to enable
communication between people and various physical and digital objects, including fields of personalized
healthcare. A method called Adaptive Domain Density Peak Clustering (ADDPC) was suggested due to
sparse cluster loss in data sets with changing density distribution. Case analysis was also used to assess
how well the model performed. So, the IoT can reduce the burden on sanitary systems even while offering
personalized health services to enhance the quality of life. As a system of interconnected objects, IoT
enables automation in several fields, including remote and intelligent healthcare systems [116]. The IoT
utilizes common technologies like as Wireless Body Area Networks (WBANs), Wireless Sensor Networks
(WSNs), and Radio Frequency Identification (RFID) to transfer data to the cloud for analysis and
extraction of meaningful data for valuable decision-making [117-120].
In a relevant study, Niu, et al. [88] created a COVID-19 classification framework based on lung CT
scans by investigating a difficult topic called Distant Domain Transfer Learning (DDTL), which tries to
address the drawbacks of classic ML and traditional TL. The suggested approach is divided into two
components: semantic segmentation and Distant Feature Fusion (DFF), and the chosen algorithm was
just one instance-based DDTL method: Selective Learning Algorithm (SLA), which may transmit
information across apparently unrelated disciplines. Otherwise, DDTL is a recently established TL
approach that primarily tries to solve the problem of negative transfer induced by loose relationships
between the source and target domains. In contrast to conventional TL methods, their DDTL algorithm
gains by combining distant features obtained from distant domains.
Besides, Połap, et al. [89] suggested an agent architecture for IoMT based on the usage of ML systems
relying on soft set theory, with the policy of sharing chosen data or trained classifier models and the
protection of user data in the blockchain. The architecture was built on a Multi-Agent System (MAS) that
separated security and data processing for medical reasons, as well as the ability to assign particular
duties to agent units. Because of the number of threads, the agent concept is paired with Federated
Learning (FL), allowing for concurrent training of classifiers and aggregating their weights to form a
single classifier architecture. As a result of the division of jobs into agents capable of completing distinct
activities, MAS may be highly effective in a real-world application.
Plus, Khan, et al. [90] suggested a multi-modal DL data fusion system for skin lesion segmentation
and classification by collecting two-segmented pictures from two separate CNN models in the first phase,
which were afterward unified using Joint Probability Distribution (JPD) and Marginal Distribution
Function (MDF). These two CNN designs were created from scratch and had 20 and 17 layers to pinpoint
the lesion locations. In addition, a 30-layered CNN architecture was developed to learn features from the
HAM10000, ISIC2018, and ISIC2019 datasets. This design has eight convolutional layers, three FC
layers, and more. Summation Discriminant Correlation Analysis (SDCA) was used to combine a set of
features from different FC layers, and the RF approach was used to choose the best discriminating
features to improve the feature space. The findings show that the suggested fusion framework
outperforms numerous current approaches with a larger margin of error.
In addition, Mahawaga Arachchige, et al. [91] suggested a Local Differentially Private (LDP)
approach for training a DNN named LATENT, which is both private and accurate. In their study, the
privacy difficulties of DL were investigated, and a distributed privacy-preserving method based on
Differential Privacy (DP) was devised to manage and restrict privacy breaches. The spread of the CNN
structure across data owners and servers also improves data processing flexibility in the big data setting.
This distribution also makes it easier for LATENT to react to breakthroughs, such as combining SDN and
NFV in edge-cloud interaction. LATENT generated a huge feature space during the randomization phase,
which resulted in greater accuracy for the CIFAR-10 dataset than the baseline CNN model without any
privacy. The findings revealed that the novel model obtained 95- 96% testing accuracy for the MNIST
dataset and 90% -91% testing accuracy for the CIFAR-10 dataset, with a high degree of privacy.
Finally, Mansour and Althobaiti [92] introduced an FDL-COVID approach for detecting and
classifying COVID-19 on IoT-enabled Mobile Edge Computing (MEC) platforms. The suggested FDL-
COVID approach used an IoT-based data collection procedure to acquire patients' CXR images.
Furthermore, the SqueezeNet method detected and classified COVID-19 using CXR pictures. The IoT
devices uploaded the encrypted variables into the cloud server, which then conducts FL on significant
variables employing the SqueezeNet model to generate a global cloud model. Finally, a GSO algorithm-
based hyperparameter optimizer is used to optimize the selection of hyperparameters in the SqueezeNet
model, significantly improving COVID-19 identification results.
The IoT offers personalized health services, remote monitoring, and improved access to medical
resources in the healthcare sector. It utilizes technologies such as WBANs, WSNs, and RFID to transfer
data for analysis in the cloud. Several studies have explored the application of IoT in healthcare. For
example, a COVID-19 classification framework based on lung CT scans was developed using DDTL,
which combines semantic segmentation and DFF. Another study proposed an agent architecture for IoT in
IoMT, incorporating ML systems, soft set theory, and blockchain for data sharing and protection.
Additionally, a multi-modal deep learning data fusion system was proposed for skin lesion segmentation
and classification, achieving better performance than existing methods. Privacy-preserving techniques
were also explored, such as an LDP approach for training a DNN and an IoT-enabled MEC platform for
COVID-19 detection and classification. These advancements highlight the potential of IoT in
revolutionizing healthcare systems and improving patient outcomes. Table 6 discusses the IoT platforms
used in medical data processing and their properties.
Table 6. The methods, properties, and features of IoT platforms in medical data processing.

Researchers Idea Advantages Challenges Security? Dataset Using TL? Method Usage?

Niu, et al. Proposing a -Low delay -High No -A small set of Yes FL COVID-19
[88] TL system -Low energy complexity labeled COVID-19 diagnosis
for medical consumption lung CT as the target based on
image data. lung CT
classification.
Połap, et al. Proposing an -High -High Yes Skin Cancer Yes FL Organizing
[89] architecture of accuracy complexity MNIST:HAM10000 medical tasks
IoMT that -High
ensures the reliability
security of -High
private data. flexibility
Khan, et al. Proposing a -High -High delay No The ISBI2016, the No CNN Skin lesion
[90] multi-modal accuracy ISIC2017, and localization
fusion system. the ISBI2018 and
classification
Mahawaga Presenting an -High -Low No CIFAR-10, MNIST No CNN Big data
Arachchige, algorithm for accuracy scalability analysis
et al. [91] the -High
redesigning Complexity Limiting
training privacy leaks
process.
Mansour Presenting an -High -Only No Benchmark CXR No FL Detect and
and FDL-COVID accessibility available in Dataset classify
Althobaiti detection -High online COVID-19
[92] model. accuracy mode

5.3 Edge computing

Edge is a term for an open platform offering the closest service with the principle functions of a network,
processing, storage, and apps on the network's edge, closest to the sources of people, devices, or data
[121, 122]. In a relevant work, Warnat-Herresthal, et al. [93] proposed Swarm Learning (SL), a
decentralized ML system that combines edge, blockchain-based Peer-to-Peer (P2P) networking and
coordination while retaining anonymity without the necessity of a central coordinator, hence extending
beyond FL. Their method can promote introducing medical data from any data owner worldwide without
violating privacy laws. Also, four utilization cases of heterogeneous diseases were assessed in their study
to demonstrate the feasibility of using SL in the development of disease classifiers using distributed data.
Finally, SL is designed to offer confidentiality-preserving ML and may inherit future breakthroughs in
differential privacy techniques, functional encryption, or encoded TL.
Besides, Kong, et al. [94] developed an edge-based mask system to identify mask-wearing using DL,
integrating VR, face recognition, and mask identification, capable of preventing COVID-19 infection
while providing real-time public health precaution notifications. Through Intel NCS, an edge computing
strategy was used in the video analysis process to improve the efficacy of detection and recognition at the
edge devices. Moreover, many trials were designed to validate the ECMask's enhanced performance. The
results demonstrated that VR can increase detection reliability based on the actual bus drive monitoring
information. The edge computing-based method outperforms the entire video analysis regarding inference
time efficiency.
In addition, Yang, et al. [95] employed AI image recognition to retrieve a Digestive endoscopic photo
and then processed the image using a 5G DL edge algorithm to determine the patient's illness type and
treatment strategy. The picture of the patients was obtained using a fluorescent probe and merged into the
AI image recognition result for blood vessel identification. The illness was then matched based on the
picture target recognition findings by performing image segmentation and AI image recognition feature
extraction. Regarding feature identification accuracy, this article's 5G DL edge method outperformed the
basic Yolo approach by 68 percent while being identical in speed. Compared to the RCNN algorithm, the
accuracy and speed were boosted by 21% and 85%, respectively.
Finally, Rahman, et al. [96] primarily focused on the feasibility of Inkjet-Printed (IJP) sensors for
biomedical applications, which included developing layout guidelines, wireless data transfer, on-device
data processing, user interaction assessment, and so on for the connected community, and displayed a
multi-modal framework for on-device data processing and inference, as well as analyzing the impact on
system performance. The findings suggested that daily wearables and other sensors might be used to
capture Events of Interest (EoI), such as biomarkers for illnesses, and the potential of on-device
processing and ML at edge devices without significantly affecting device performance.
Edge computing provides network functions, processing, storage, and applications at the network's
edge, close to data sources, devices, or people. SL is a decentralized ML system that combines edge
computing, blockchain-based P2P networking, and coordination. SL allows the introduction of medical
data from various owners globally while ensuring privacy and anonymity. An edge-based mask system
was developed to identify mask-wearing using DL, VR, face recognition, and mask identification. This
system prevents COVID-19 infections and provides real-time public health notifications. AI image
recognition and 5G DL edge algorithms were employed for digestive endoscopic photo analysis,
achieving higher accuracy and speed compared to baseline approaches. IJP sensors were explored for
biomedical applications, including on-device data processing, wireless data transfer, and user interaction
assessment. On-device processing and ML at the edge were shown to have potential without significantly
affecting performance. These advancements showcase the use of edge platforms in medical data
processing for improved efficiency and real-time analysis. Also, Table 7 discusses the edge platforms
used in medical data processing and their properties.
Table 7. The methods, properties, and features of edge platforms in the medical data processing

Researchers Idea Advantages Challenges Security? Dataset Using Method Usage?

TL?

Warnat- Introducing a -High data -Limited Yes PBMC Yes DNN Detection of
Herresthal, method for data privacy and dataset, transcriptome heterogeneous
et al. [93] integration security, -Lower dataset, Whole- illnesses
without -Low data performance. blood-derived
violating privacy traffic and transcriptome
laws. duplication. datasets, and X-
ray dataset
Kong, et al. Suggesting an -High -Limited Yes Bus Drive No Auxiliary COVID-19
[94] edge-based accuracy dataset Monitoring model prevention
mask -Low energy Dataset training
identification consumption
architecture. -Low cost.

Yang, et al. Proposing a 5G -High -Low data No MSTAR No RCNN Disease

[95] DL edge accuracy coverage +yolov3 recognition
algorithm. -Lower Delay related to
digestive
endoscopy
Rahman, et Implementing -High -High energy Yes 9 participants No RF Early
al. [96] pre-trained ML scalability consumption conducted 480 +MLP monitoring of
in the data collection a disease
smartphone app sessions in the outbreak
for computing ―living lab‖
disease-related environment
EoI.

5.4 Fog computing

The fog platform is a modified subset of the cloud architecture from the center to the network's edge. It is
a highly scalable platform that aims to connect end devices to conventional cloud servers by integrating
computations, storage, and networking [123]. Various studies have investigated novel features of fog
computing in medical analysis uses by proposing diverse frameworks. For example, because
cardiovascular disease is one of the main causes of high mortality rates in numerous countries, Khan, et
al. [97] suggested a smart fog-based heart disease model architecture equipped with a supervised ML
system—a supervised ML technique known as SVM (SVM is a set of supervised learning approaches
utilized for classification, and regression through the training data set [124]) was studied to predict heart
disease output to obtain maximum accuracy. Data was gathered via IoMT-enabled devices. The model
was separated into two phases: training and validation, following the evaluation of the training forecast
layer's accuracy using various statistical techniques such as accuracy, missing ratio, sensitivity, and so on.
If the necessary training learning conditions were not fulfilled, the prediction layer was retrained, and its
performance was assessed. A successful training model was stored in the cloud after meeting the requisite
threshold learning criterion or the appropriate number of cycles. It was shown that the suggested system
generated 93.33% accuracy.
In another study, Manocha, et al. [98] offered a framework for assessing an individual's physical
postures in real-time to forecast health issues using superior fog computing and video analytics. The goal
of merging fog characteristics with video analytics was to improve hardware competency by lowering the
cost of data transfer to faraway servers. The primary goal of the proposed research's fog computing was to
deliver higher computational capabilities by minimizing the latency and computational cost of distant IoT
nodes. The framework's structure was divided into 4 stages: 1) user subsystem: data gathering and pre-
processing, 2) fog analytics: health condition prediction, 3) cloud subsystem: determination score
recording, and 4) two-phased decision-making mechanism. All stages operated independently, providing
an excellent operating environment for the next phase. In addition, the suggested framework's execution
performance in determining the answer for each frame segment has to be evaluated. The suggested system
was shown to be very efficient in live monitoring by merging many cutting-edge technologies such as
Cloud-of-Things, Fog-of-Things, and IoT.
Besides, a hybrid fuzzy-based Reinforcement Learning (RL) method using NN evolutions techniques
was suggested by Shukla, et al. [99]. The suggested technique was employed for allocating and selecting
IoT data packets for the healthcare industry in a fog environment using fog nodes. The healthcare IoT
data were categorized using the Fuzzy Inference System (FIS) and the linear Support Vector Machine
(SVM). The following metrics were used to analyze the high latency problem: network latency,
communication latency, compute latency, network utilization, and RAM consumption. The suggested
technique considerably decreased network utilization, RAM consumption, computational latency,
communication latency, and network latency for healthcare IoTs. The findings showed improved
execution of the suggested strategy for FC-based latency reduction. This methodology was determined to
be ideal, demonstrating its usefulness in healthcare IoTs.
Plus, Horng, et al. [100] created an anomaly detection approach utilizing DL with precise data to allow
the NN to be trained, in addition to pre-processing the original file and effectively keeping just the portion
of the brain region that has to be studied, normalizing the brain image to fulfill the parameters of model
training. Furthermore, they employed Variational AutoEncoder Wasserstein Generative Adversarial
Network with Gradient Penalty (VAE WGAN-GP) to create 3D medical pictures to address the issue of
insufficient training data. The authors recommended adopting the generation model to overcome the issue
of too little data. The model produced accurate brain images that were equivalent to the raw data, and the
probability density of the visual data set suggests that the model had a typical amount of training.
Automatically locating outliers in images and acquiring the input data characteristic were done using an
AutoEncoder. With the model's input and output similarity of more than 50%, 76 percent of brain
bleeding and non-bleeding accuracy was thus attained.
 In addition, Mohamed Akram, et al. [101] presented a system for COVID-19 detection by attempting
to minimize the spread of the pandemic via early identification of the infected people. Also, the suggested
approach permitted the detection of COVID-19 anywhere and anytime. The system was simple to use and
easily accessible. Furthermore, they concentrated on the QoS of the cloud by establishing an interstitial
layer between both the user and the cloud to decrease the latency and enable real-time response. For
improved categorization and identification results, the combination of Discrete Wavelet Transform and
Principal Component Analysis (DWT-PCA) was applied to decrease the dimension and extract the best
features. Plus, the method reduced the complexity by decreasing trainable parameters and time. The
assessment of this CNN model in the classification demonstrated effective performance in categorizing
Covid-19, pneumonia, and healthy cases with a high accuracy rate.
Finally, Singh, et al. [102] presented a fog-centric intelligent medical support service for controlling
and monitoring the Swine Flu virus epidemic, which employs the concept of fog computing for delay-
sensitive programs and a hybrid classifier to identify swine flu patients early on by sending alerts to
health authorities and patients' caregivers. In the experimental context, they focused on obtaining low
latency and improved energy efficiency compared to the current technologies, allowing the suggested
scheme to provide real-time alerts with little delay. Compared with the previous cloud-only model, the
results reveal the advantages of merging clouds and fog computing services by offering optimized
network bandwidth stability, greater performance, and a faster response time despite generating real-time
notifications.
The fog platform is a modified subset of cloud architecture that brings computations, storage, and
networking closer to the edge devices. It enables connecting end devices to cloud servers and has been
explored for medical analysis purposes. A smart fog-based heart disease model architecture was proposed
using supervised ML techniques to predict heart disease accurately. Real-time assessment of physical
postures was achieved through the integration of fog computing and video analytics, enhancing
computational capabilities and reducing latency. A hybrid fuzzy-based reinforcement learning method
was suggested for allocating and selecting IoT data packets in healthcare, improving latency and resource
consumption. An anomaly detection approach using DL was developed to detect brain anomalies
accurately. A system for COVID-19 detection was introduced, leveraging discrete wavelet transform and
principal component analysis for feature extraction and achieving effective classification performance. A
fog-centric intelligent medical support service was presented for monitoring and controlling the Swine
Flu epidemic, combining fog computing and cloud services to deliver real-time alerts with improved
performance. These studies demonstrate the potential of fog computing in medical applications, offering
enhanced efficiency and real-time capabilities. Table 8 discusses the fog platforms used in medical data
processing and their properties.
Table 8. The methods, properties, and features of fog platforms in the medical data processing

Researchers Idea Advantages Challenges Security? Dataset Using Method Usage?

TL?

Khan, et al. Proposing an -High -High delay No UCI repository No SVM Heart
[97] intelligent- accuracy section of Cleveland diseases
based heart -Lower miss Heart Disease diagnosis
disease rates dataset
prediction
system.
Manocha, Presenting an -Low delay -There are still No NTU RGB + No 3D CNN- Monitoring
et al. [98] analytic- some false D dataset LSTM anomalous
assisted results physical
physical stance- activities
 based
irregularity
recognition.

Shukla, et Proposing a -High -Low No Obtained from the No RL Healthcare

al. [99] hybrid robustness reliability online UCI ML +NN IoT
fuzzy-based RL -Low repository +SVM Devices
algorithm in an latency analysis
FC
environment.
Horng, et Proposing a 3D -Ultra-low -Poor model Yes Digital Imaging and No GAN Prediction
al. [100] deep latency performance Communications in +AutoEncoder the
convolution -High for few Medicine (DICOM) possibility
AutoEncoder accuracy training data of bleeding
model. after
treating
brain
ischemia
Mohamed Proposing a DL -Lower -Less Yes Kaggle repository No CNN Covid-19
Akram, et framework for delay complexity detection
al. [101] COVID-19 -Better
detection. privacy
Singh, et al. Monitoring the -High -No deep Yes Data from various No Ensemble Swine Flu
[102] health accuracy analysis of the sensors like classifier detection
of citizens -Power model due to location, health,
concerning saving lack of time. climate, and
Swine Flu. -Low prescription sensors
latency

5.5 Hybrid systems

The term "hybrid approaches" might apply to simultaneous acquisitions in platforms using two different
methodologies. [125]. In relevant research, Fan, et al. [103] innovated the use of social networking health
discussion data for Adverse Drug Effect (ADE) extraction from websites such as WebMD and [Link],
while previous research has used data from Twitter, EHR, and medical case reports. The suggested model
produced fresh state-of-the-art F1 outcomes that may serve as the foundation for future improvements. By
comparing ADE detection findings with the proposed model, They achieved an AUC of 0.94,
demonstrating the viability of Bidirectional Encoder Representations from Transformers (BERT) word
embeddings over other forms of non-pretrained word embeddings and pretrained embeddings.
In another study, Arabi and Zaidi [104] introduced a method to anticipate patient-specific Attenuation
Correction Factors (ACFs) from PET emission data utilizing DL methods. In this instance, the data from
the PET emission sinogram was used to train a deep CNN to forecast the associated attenuation map or,
more precisely, the ACF sinograms. The deep CNN was trained and assessed using ACF sinograms
produced from reference CT images. The reference CT-based attenuation correction was compared to the
predicted ACF sinograms utilized for attenuation correction within PET image reconstruction. The DL-
EM method outperformed the segmentation-based strategy typical of clinic algorithms in quantitative
PET analysis. The DL-EM technique resulted in an absolute SUV bias of less than 8% in all brain areas,
while the SEG method resulted in an absolute SUV bias of up to 14%.
Besides, Yuan, et al. [105] suggested a Hybrid Network (Hy-Net) for blood vessel segmentation that
incorporated FCN and U-net through a voting system on PA vascular images. They mostly discussed four
techniques: threshold segmentation, region growth, Maximum entropy, K-means clustering, and three
DL methods such as FCN, U-net, and Hy-Net. The four metrics, Dice Coefficient (DC), Intersection over
Union (IoU), Sensitivity, and Accuracy, were used for each test trial to measure the performance of tests
on different segmentation approaches. The maximum values of FCN in terms of DC, IoU, sensitivity, and
accuracy were 84.07%, 72.52%, 87.43%, and 99.71%.
 Plus, Al-Saffar and Yildirim [106] proposed a variety of techniques to enhance the results and reduce
the challenges associated with the process of medical image analysis, including a brain tumor
segmentation method based on Local Difference in Intensity-Means (LDI-Means), a feature selection
method based on Mutual Information (MI), a dimensionality reduction method based on Singular Value
Decomposition (SVD), and a brain tumor classification method based on both SVM and Multi-Layer
Perceptron (MLP). They also introduced an innovative approach termed Multiple Eigenvalues Selection
(MES) to select the most significant features as classifier inputs. It was used as a completely automated
brain tumor identification, segmentation, and grading system. The most significant characteristics are
found, which lead to good tumor grade detection and time savings. According to experimental findings,
the suggested strategy performed well in terms of accuracy, recall, specificity, precision, and error rate
with percentages of 91.02%, 86.52%, 94.26%, 87.07%, and 0.0897%, respectively.
In addition, Momenzadeh, et al. [107] suggested a hybrid platform for estimating the survival rate of
patients with prostate cancer by using the Factor Analysis of Mixed Data (FAMD) algorithm, along with
undersampling techniques for the SEER dataset as a pre-processing stage ahead of the leading
frameworks, namely XGBoost, RF, SVM, and Logistic Regression (LR), using a cross-validation
approach for parameter modification to forecast both binaries labeled and multi-class labels, which has
been rarely studied in previous relevant papers. Compared with similar studies, this technique
successfully distinguished the patients based on their mortality status and whether the reason for death
was prostate cancer. This technique can be beneficial for making clinical decisions or assessing whether
medical doctors need to modify their treatment plans.
Finally, Seba and Benifa [108] established a useful classification algorithm for the accurate prediction
of Chronic Kidney Disease (CKD) stages regarding the patient health profile as well as the laboratory test
results by presenting a Hybrid Ensemble Model (HEM), which assists in minimizing the bias and
deviation. HEM model efficiency was examined using the performance measures such as Cross-
Validation Score (CVS), accuracy, specificity, recall, F1 measure, Mean Squared Error (MSE), bias, and
deviation. It is evaluated with state-of-the-art classification schemes.
The term "hybrid approaches" refers to simultaneous acquisitions in platforms using two different
methodologies. In one study, social networking health discussion data was utilized for extracting
ADEsusing a model based on BERT word embeddings. The proposed model achieved state-of-the-art
results, demonstrating the effectiveness of BERT embeddings. Another study introduced a deep learning
method to predict patient-specific ACFs from PET emission data, outperforming clinic algorithms in
quantitative PET analysis. A hybrid network incorporating FCN and U-net was suggested for blood vessel
segmentation, achieving high-performance in terms of Dice Coefficient, Intersection over Union,
sensitivity, and accuracy. Various techniques were proposed to enhance medical image analysis,
including brain tumor segmentation, feature selection, dimensionality reduction, and brain tumor
classification, resulting in accurate tumor identification and grading. A hybrid platform utilizing the
FAMD algorithm was proposed for estimating the survival rate of patients with prostate cancer, enabling
better patient stratification. Lastly, a HEM was developed to accurately predict CKD stages, achieving
high accuracy and performance compared to other classification schemes. These studies highlight the
effectiveness of hybrid approaches in various medical analysis tasks, demonstrating improved
performance and accuracy. Table 9 discusses the hybrid platforms used in medical data processing and
their properties.
Table 9. The methods, properties, and features of hybrid platforms in medical data processing.
 Researchers Idea Advantages Challenges Security? Dataset Using Method Usage?

TL?

Fan, et al. Investigating the -Large drug data -High No .Two Yes SVM Drug
[103] use of the source complexity primary adverse
developments in -Strong datasets, events
DL as accurate robustness WebMD and discovery
models. [Link]
Arabi and Proposing a -Low delay -Remarkable No CTAC No Deep CNN Brain PET
Zaidi [104] framework to overfitting analysis
estimate ACF -High
from TOF PET complexity
emission data.
Yuan, et al. Proposing a -High accuracy -Not trained No A Swiss No FCN PA vessel
[105] hybrid system -High for capillary Webster images
consisting of robustness segmentation mouse using
both applied on an OR-PAM
PA vessel system
images.
Al-Saffar and Proposing MES -Improving the -High No TCIA No SVM Brain
Yildirim [106] methods. generalization complexity +MLP tumor in
ability of the magnetic
classifiers MRI
Momenzadeh, Suggesting a -High accuracy -Applied only No SEER No RF Prostate
et al. [107] hybrid method to -Better recall for low +SVM cancer
predict the amounts of +LR
mortality of variables
prostate cancer. - High
complexity
Seba and Proposing a -High accuracy -High No Benchmark No RF Prediction
Benifa [108] HEM model for -Less bias complexity +Ensemble of CKD
estimation of classifier stages
CKD stages.

6. Results and Comparisons

In the previous section, we systematically analyzed 27 papers selected based on the specific criteria which
are provided in Fig 2. Here in this part, we will discuss the article's main concept by summarizing the data
analysis in the healthcare domain. The following subsections discuss the categorized platforms of
distributed computing, DL and ML methods in medical data processing, security and privacy, datasets,
big medical data processing, and an analysis of the results.
6.1. Platforms in the medical data processing
The present healthcare services idea aims to innovate in the medical business by utilizing fog, IoT, big
data, and ML technologies. This section discusses several flexible applications in healthcare data
processing by focusing on the practical function, their advantages, and their drawbacks, which
significantly affect the process and the outcome. In the previous section, we investigated various
applications of distributed computing platforms in healthcare data management based on the most recent
research. This section briefly reviews and analyzes the platforms regarding their definition, practical
usage, advantages, drawbacks, identified disparities, and applications.

 Cloud computing platform

Widely considered the future IT revolution, cloud computing has recently emerged as among the most
researched topics among IT experts. Hence, in the field of medicine, the cloud has become increasingly
significant as a trigger for providing storage and computational capacity for massive quantities of data.
The primary characteristics of the cloud in medical data processing include on-demand self-service, wide-
ranging access to the network, dynamic resources, quick flexibility, and quantified service. In the review
studies in section 5, the main purpose of the cloud application in medical data management systems was
to assist HCP in to fast recognition of disorders which can enhance the accurate and effective treatment
plans and following up of the patients.
Furthermore, cloud computing in EHR enables patients to access, record, and transmit their private health
data. Several researchers have proposed the cloud as a cutting-edge and significant commercial model for
the exchange of biological data. Institutions in the healthcare sector may use cloud computing to support
efficient big data management and dissemination processes for the industry's well-organized information
systems. On the other hand, the dynamic capabilities of the cloud, together with the assistance from good
use of its analytical and data interpretation skills and organizational resources like data governance, may
enhance healthcare performance. This suggests that dynamic talents must be combined with other
company resources to provide a sustained competitive advantage.
Undoubtedly, cloud computing is a highly practical method for the healthcare issue. Still, it must
concentrate on privacy protection, data security, data control, trust, and lawful access to patients’ sensitive
data. Although cloud computing can collect and store large amounts of medical data, the key challenges
are network disruption, security, and the privacy of patient records that could be misused by users,
hackers, spyware, and other third parties. Future studies should take into account other issues, such as
fierce competition, patient satisfaction, confidentiality, data volume, environmental concerns, investment
value, and dependability.

 IoT platform
Due to high data creation velocity, IoT devices in the medical sector are also more stressed in research
articles. Besides, when associated with cloud computing, the accessibility and flexibility of equipment are
boosted. In particular, IoT-based healthcare systems can be used to identify patients' physical and
chemical changes through various sensors. These have also contributed to establishing medical services
for patient management in actual environments by conveying information to hospitals and healthcare-
related medical organizations. In particular, technology for constantly monitoring a patient's well-being
and wireless communication technology has been created.
IoT is transforming the healthcare sector by enhancing proficiency, decreasing expenditures, and
preserving stability. It aggregates various intelligent services and equipment that affect daily functioning
via e-learning, remote access, and assisted surveillance systems. IoT is a vital aspect of medical research
and for increasing the effectiveness and quality of healthcare delivery. Because of this, the idea of
healthcare has changed from being institution-centric to customized. We reviewed some applications of
IoT in the healthcare domain in section 5. Due to the COVID-19 pandemic, one of the most repeated
usages of IoT in the most recent research was the fast and accurate detection of various diseases,
including COVID-19, skin lesions, skin pigmentations, etc.
Although, for a more comprehensive insight into the healthcare domain, the DL applications for IoT
can be categorized into four categories: applications for medical detection and classifying, personal and
home-based health services, applications for disease prediction, and applications for human behavior
assessment. Several technical guidelines for healthcare systems are provided in the IoT, but the key
emphasis should be on assessing the appropriate technological development based on specific limits and
digital devices. Besides, the generation of meaningful data from the IoT for bioengineering is a
substantial research issue in this domain.

 Edge computing platform

Edge improves both the quality of care for patients and the effectiveness of medical staff by operating like
a decentralized distributed system. Medical staff is not required to send patient health data to a distant
data center for analysis and results waiting. They may build an edge data center to analyze, compute, and
store patient health data acquired by edge devices locally, bringing significant advances to the medical
field. Edge medical equipment gathers data and also offers an access point to a network connection. The
edge computing paradigm distributes the management of a network’s services away from local nodes to
the opposite extreme: the sensor itself instead of servers or nodes. Edge has indeed been described as dew
computing. In an edge computing framework, devices may conduct specialized calculations and interact
with each other. They may also communicate data to a node for specified aggregating and responses,
which can then be delivered to the cloud. Besides diabetes systems, the edge is currently utilized in plenty
of other medical closed-loop systems in which the sensor input is designed to impact the actuator output,
such as (1) cardiac pacemakers; (2) cardiac defibrillators; (3) systems interventional closed-loop
mechanical ventilation that is still not on the market; (4) the Sedasys closed-loop anesthesia delivery
service that was approved by the FDA but removed from the market due to poor sales; and (5) Brain
activity which can be altered with closed-loop stimulation. Edge is well adapted for closed-loop systems
which employ intelligent sensors to preserve physiologic homeostasis; however, the method is presently
being used only to a limited degree due to the FDA's approval of only a few independent closed-loop
systems. This method would be appropriate for an Intensive Care Unit (ICU) or an emergency medical
department, where critically ill patients need quick solutions for changes in their condition. Edge is
accomplished by linking sensors in a system to tiny, local control units that handle processing and
communication. Edge computing can result in quick Machine 2 Machine (M2M) connectivity or
machine-to-human interaction. Edge computing enables each web device to contribute to the data
processing rather than having most of it done by a central or decentralized local server.

 Fog computing platform

To the degree that certain projects need quick data analysis, fog may help new concepts in big data
analysis, privacy-sensitive distributed learning, citizen science applications, remote health monitoring,
and medical data analysis. A cloud that is closer to the ground is called fog. In contrast to virtually totally
up in the cloud, fog computing for medical devices focuses on processing "down to earth" near the
patient. Fog computing distributes processing power to a local area network level where data is processed
inside a hub, node, router, or gateway and then transferred to the relevant devices. This function can be
performed using a smartphone.
Fog often uses open standard technologies, but the edge can employ open or private technology even
though there is some overlapping between the two platforms. Under a fog computing paradigm, a device
may broadcast vital data to a nearby fog node for quick analysis and reaction to the unit within
milliseconds or seconds. The remainder of the data can be transmitted to the cloud for long-term archival
and future processing. A device can also transfer data that can wait a few minutes to a larger aggregation
fog node that supports multiple IoT devices. Data from the fog or aggregation nodes may be stored and
analyzed in the cloud. Wearable and portable diabetics devices provide increased Bluetooth connectivity
with a smartphone or tablet, which functions as a transportable fog gateway by conducting basic analytic
procedures and transferring monitored data to the cloud. A fog node may send data from wireless portable
diabetic devices to the cloud. These tools include insulin pumps, continuous glucose monitors, insulin
pens, and continuous glucose meters. Closed-loop artificial pancreas systems are also included.
Between the edge nodes and the cloud layer is an intermediary layer called the fog computing layer.
Local computing and storage are provided for local authorities by the fog computing layer, which expands
the computation notion geographically. Cloud computing is not distributed via fog computing. To enable
a new breed of programs and capabilities with an efficient interface with the cloud layer, it aims to bring a
processing and storage architecture that is significantly closer to the final nodes. A better QoS for
applications that require low latency is the anticipated benefit. When data is carefully processed at the fog
computing layer, latency is reduced. Fog computing's data analysis is more limited; thus, the cloud
computing layer should conduct more in-depth analyses and procedures. Naturally, certain activities do
not demand real-time calculation or may require significant computational power; they are conducted at
the cloud layer. For instance, in the scenario of a smart community, where residents are connected to the
provision of local services, limited latency is projected for producing immediate decisions, and as a result
computation is done locally rather than on a cloud layer that might be located in another location.

 Hybrid platforms
In medical imaging, a broad scope of visual input from other categories is already accessible to scientists
globally. Several networks gather and distribute extremely diverse data from multiple modalities,
diseases, and body zones. Consequently, hybrid platforms as a paradigm are becoming more popular. The
physical integration of complementary imaging techniques, such as CT, PET/CT, and MRI, all of which
give "Renato-metabolic" image information based on inherently synchronized morphological and
functional data, is referred to as hybrid imaging. Routine hybrid imaging diagnosis relies mostly on data
interpretation. However, this data has a lot more information that can be transformed into knowledge.
The acquisition and analysis of simple to elevated radiometric characteristics represent a paradigm
shift in illness in vivo characterization. This notion looks to be very promising in cancer research. On
both structural and metabolic levels, hybrid imaging may assist in depicting the total heterogeneity of
malignancies. As a result, hybrid methods look to be a critical tool for developing precise tumor
classification in vivo models. Per the research, advanced PET/CT and PET/MRI systems generate
terabytes of data. Furthermore, hybrid platforms mix modalities like cloud, edge, fog, IoV, etc., reflecting
varied image sizes and quality levels, resulting in essentially diversified datasets. All of these
characteristics combine to form hybrid platform data, a significant aspect of medical big data. Several
reports have indicated promising findings for disease characterization using sophisticated ML approaches
integrated into Vivo evaluation in light of hybrid imaging. Real-time remote monitoring is now possible
because of the rise of the MIoT. By collecting and integrating MIoT data with hybrid methods, predictive
analytics accuracy may be significantly increased, potentially leading to automated Clinical Decision
Support Systems (CDSS). As part of a completely patient-centric, individualized treatment, this leads to
healthcare practices that help increase patient comfort while lowering healthcare costs. As a key
component of this approach, hybrid imaging data plays a critical role in customized medicine.
6.2. DL & ML methods in the medical data processing
AI is defined as computer systems capable of carrying out tasks that people, such as decision-making,
voice recognition, and visual perception, traditionally do. ML is a branch of AI that enables adequate
statistical models to be built using computer sciences on data sets that are already available. ML models
are multi-layered models with input, output, and one additional layer ("hidden" layers). These models
occasionally provide visual representations of the structure of human brain transmission, replete with
interconnected nodes and synapses. Within these models, data is fed into the input nodes and processed
through each layer until it reaches the output layer. The number of hidden layers determines the depth of
the model.
The capacity to train using backpropagation has been important to the precision of these models'
performance. DL refers to ML models that use numerous layers of processing units. The output from one
layer to the next becomes a more sophisticated and abstract composition. DL models can quickly and
efficiently analyze huge amounts of data compared to typical ML models. Many ML models include
layers; still, there is no standard definition of how many layers constitute a "deep" model. DNNs are those
having two hidden layers, while shallow neural networks have one hidden layer. DL is a kind of ML
approach that is based on ANNs. A CNN is a sort of NN capable of processing 2D image input that is
often encountered in the field of DL.
Learning approaches such as supervised, unsupervised, and semisupervised learning may be used to
generate ML models. The distinction between the three is determined by how much training data has the
labeled intended outcomes for the model. In most cases, supervised learning is used to map an input
sample to an output label for classification. It is also used in regression, which aims to develop a mapping
from continuous inputs to constant output. We want to identify the proper links between both the input
and output in categorization and regression. Indeed, we are searching for a method to create proper output
data successfully. If the training set is chaotic or has inaccurate labels, the trained model's performance
will degrade significantly. The majority of typical ML models perform best when supervised learning
approaches are used. Linear Regression, ANN, Decision Trees, RF, SVM, and Naive Bayes are among
the models used. Unsupervised learning aims to discover the internal representation of unlabeled data.
Segmentation, density estimation, and representation learning are the most prevalent unsupervised
learning challenges. Several methods, such as auto-encoders, have been developed for this task.
Exploratory analysis and dimensionality reduction are two frequent use cases utilized in unsupervised
learning. Clustering techniques like Density-Based Spatial Clustering of Applications with Noise
(DBSCAN), k-means, and Gaussian Admixture are typical examples of algorithms employed in
unsupervised learning. Building features with desirable attributes may also be accomplished through
unsupervised learning.
Also, RL is the process of learning to map conditions to appropriate behaviors to maximize a
numerical reward signal. Unlike supervised learning, the learner in RL is not given the desired action and
must experiment with alternative actions in different scenarios to determine the optimal actions leading to
the highest compensation given the observed states. It is essential to learn action sampling to maximize
long-term utility since focusing just on the immediate reward may result in inferior productivity in the
long run. The CNN, a DL algorithm intended to handle data with natural spatial invariance, has emerged
as a key player in this area. NLP is concerned with analyzing text and voice to derive meaning from
words. In this sector, RNNs_DL methods that are adept at processing sequential inputs such as language,
voice, and time series data play a significant role. Machine translation, text creation, and picture
captioning are examples of notable NLP accomplishments. Sequential DL and language capabilities are
used in healthcare to power applications such as EHRs, which are becoming widespread, and the in recent
years, a large medical organization's EHR may collect the medical operations of over 10 million patients.
A single hospitalization produces around 150,000 bits of data. The potential advantages of this data are
substantial. An EHR of this size represents 200,000 years of clinician knowledge and 100 million years of
patient outcome data, spanning a wide range of unusual diseases and disorders. As a result, applying DL
algorithms to EHR data is a fast-growing field. The determination of model parameters is a convex
optimization issue; hence the solution is always globally optimal, which is a key characteristic of SVM.
In addition, many current convex optimization methods are easily adaptable to SVM implementation. As
a result, SVM has been widely employed in medical research.
6.3. The dataset in the medical data processing
Also, Image information is vital in many steps of the patient care process, including detection,
characterization, staging, treatment response evaluation, disease recurrence monitoring, and directing
interventional treatments, surgeries, and radiation therapy. The use of multi-modality imaging increases
the quantity of picture data that must be interpreted. Because of the increased workload, radiologists and
doctors are finding it challenging to maintain workforce productivity while using all existing imaging
information to enhance accuracy and patient care. With recent improvements in ML and computing
approaches, the possibility and necessity for creating automated methods to help radiologists in image
processing and detection have been considered essential medical imaging research and development field.
Additionally, different clinical imaging modalities, such as MRI, X-ray, and CT, make DL an
appropriate choice for early and accurate disease detection. Medical imaging has developed from a purely
visual tool to a major source of analytical methods for identifying diseases in vivo. This strategy includes
hybrid imaging, which appears crucial for creating accurate tumor characterization in vivo models.
Generally, from the perspective of hybrid imaging, various researchers have presented significant insights
for disease classification using robust ML approaches for integrated into vivo analysis.
Data is the most precious property in the health industry. The vast majority of modern ML approaches
are data-driven. The dataset relies almost entirely on DL-based methods. As a result, one of the main
motivations in managing a disease epidemic is the dataset's sources. The database industry and academics
have been researching data management-oriented issues in ML for almost a decade. This has resulted in a
wide range of tools and platforms for scaled and rapid ML established by this community. Besides,
analysis of massive datasets employing statistical ML techniques is essential to current data-driven
applications in healthcare, research, and other disciplines and efforts that utilize database-inspired
techniques to make ML quicker and more user-friendly. The variety of this environment of systems and
programs might be daunting for data management academics, data analysts, and software engineers alike.
The suppliers of the data generated for the DL application of physiologic signal analysis have been
explored by various scholars. The accessible public datasets which are extensively utilized include CHB-
MIT, PhysioNet, BCI competition II, MIT-BIH, DEAP, NinaPro, and Bonn university dataset. Also, the
private dataset was obtained by authors with their lab, hospital, or institution. The private database was
gathered if the info was not accessible as a public entity.
6.4. Security and privacy in the medical data processing
DL's privacy concerns have recently been exposed, and different attacks have been considered. Modeling
privacy threats are divided into model extraction attacks and model inversion attacks. In model extraction
threats, the adversary attempts to recreate the model's parameters/hyperparameters used to deliver cloud-
based ML services, compromising the security and intellectual property of the network operator. The
adversary uses accessible records to derive sensitive data in model inversion threats. Furthermore, a high-
performing ML model needs a lot of training data, a lot of hardware, and a lot of effort to fine-tune the
parameters. Consequently, it has been recognized that the labeled training dataset, system framework, and
model parameters are instances of business intellectual property; therefore, they must be protected. Only a
few works of watermark-based ML model protection for intellectual property are available at the moment,
and it is still challenging to guarantee their efficacy. How to protect intellectual property more effectively
and securely for NN models is yet unanswered. Recent research has established numerous privacy-
preserving DL techniques to preserve the security of sensitive data. Nevertheless, much research remains
to be done before it can be implemented. The most serious flaw with privacy-preserving DL approaches is
their high computational cost. The calculation expense of DL is considerable because of its non-linear
behavior, which severely limits its availability. Reducing the cost of privacy preservation is a critical
challenge for DL techniques that protect privacy.
Also, DL security risks are divided into two categories poisoning attacks and adversarial attacks. DNN
is vulnerable to adversarial attacks in the form of perturbations by incorporating them in the original
image, which is imperceptible to the human visual system. Such assaults can cause a NN classifier to
produce incorrect forecasts with high confidence. In poisoning attacks, the adversary attempts to
contaminate the training data by inserting malicious samples or altering data so that the learner trains a
faulty classifier, which misclassifies harmful instances or actions generated by the adversary during the
testing stage. Protecting users' privacy in healthcare is crucial since it is a user-centric application and
includes the acquisition of personal data. Any violation of privacy might lead to inevitable repercussions.
Preserving privacy implies that ML model training and interpretation should not expose any new
information about the persons from whose data was acquired. In general, ML/DL needs training data kept
on a centralized database that may contain the users’ details which presents numerous vulnerabilities. To
resolve such issues, data anonymization methods are employed. However, the literature has noted that
important information may be gleaned about individuals’ private information even though the dataset is
anonymized.
6.5. Big data analysis in the medical data processing
Big data medical data processing is defined as a collection of methods and instruments used to examine
significant volumes of unstructured data. Big data in the medical sector include unstructured data, like
images, information, data, and videos. Since of its frequent alternation, unstructured data is difficult to
process because it lacks comprehensive structural data styles. Big data may profit from pattern
recognition as well as other distinct techniques that depend on computing to provide new insights. DL,
ML, and computational advancements in pattern recognition have also been merged with big data. This is
connected to comprehending computational aspects discovered in data obtained with predicted
investigative discoveries. Big data does not allow for the completion of all properties among identical
data. There are two key differences in using big data technologies for visual pattern recognition. Big data
refers to enormous volumes of data that cannot be stored on a single device. Second, since big data
requires specific tools and techniques, the conventional data notion had to be duplicated because it lacked
structure. Several programs, including basic databases, NoSQL, Data Stream Management Systems
(DSMS), and Memcached, may be used for large data, with Hadoop being the most well-liked and
suitable option.
So, Big data processing has been a critical aspect of being creative and competitive for medical
corporations. it is one of the major topics of study in the scientific world. Alongside population growth
internationally or the development of aging in certain industrialized nations, the public healthcare systems
confront a growing proportion of healthcare expenditure and a massive volume of health records.
Furthermore, cloud computing is now a concept to handle the vast data inside this big data generation.

6.6. Analysis of the results

This study identifies several innovative applications which have been popular in medical data processing
and analysis. A total of 27 articles in 5 categorical applications of medical data processing were studied,
including cloud, edge, fog, IoT, and hybrid platforms. The most studied platform was IoT, containing
43.4% of the total searched papers. Fig. 8 shows the distribution of the papers regarding the platforms.
IoT, in combination with cloud computing, aids healthcare by analyzing medical data in ultra-low delay,
whereas appropriate decisions can be made on time. In the field of HIoT, the special advantages might
demonstrate by fog and edge computing concepts. They resemble the model where confidential material
produced by body-worn clinical devices and smartphone sensing are prepared, evaluated, and extracted
close to where it is obtained, on these devices themselves, rather than sending large amounts of signal
data to the cloud. Only aggregated, context-enhanced data/events are transferred to the cloud for
additional processing and analytics. The correct balance between Edge-Cloud should enable quick
reaction to occurrences, network capacity conservation, latency reduction, security, and privacy
protection.
 Fig. 8. Utilization of DL platforms and their frequency in the investigations.
Also, DL methods can be aggregated to propose methodologies and techniques which assist in
fulfilling the gaps in the healthcare era as disease recognition, remote health monitoring, accurate and
robust results, etc. Section 5 presented several DL and ML platforms in the discussed articles, including
CNN, GAN, LSTM, RF, SVM, FL, Ensemble methods, etc. The most utilized method was CNN, with a
distribution of 19.4% among the analyzed studies, and in the second stage, SVM, with a distribution of
11.1%, was applied in the investigations. The prominent reason for the high utility of CNN and SVM can
probably be the result of high accuracy and lower delay in their usage. Moreover, by application of CNN,
there were fewer privacy leaks and better robustness and performance of the proposed framework. Fig. 9
shows the utilized DL and ML approaches in the studied articles. Also, Fig. 10 provides the geo-chart of
the authors of the reviewed papers.

Fig.9. The distribution of the studied platforms.

.
 Fig. 10. The geo-chart within the studied countries based on the investigated articles

With the development of the machines’ processing capability, several approaches were successfully
established based on ML and DL in medical fields as a classification and recognition platform. In general,
many diverse applications can be used for healthcare data analysis. The most discussed application,
according to section 5, is quick and accurate diagnosis of patients. It can help the healthcare staff better
manage diseases with more efficient treatment plans in earlier stages, decreasing morbidity and mortality
rates for economies. Also, the recent COVID-19 pandemic was a huge burden on practically all countries
in the healthcare era, resulting in an emphasized outline of early, fast, and accurate detection and
recognition of diseases. Meanwhile, DL has been used in various medical applications, including cancer
diagnosis, heat disease recognition, localization of skin lesions and pigments, assessment of bleeding
probability after brain ischemia, and digestive endoscopy.
Otherwise, various practical applications of the usage of distributed systems were discussed in section
5, such as medical image processing and analysis, COVID-19 detection, sports medical data, face and
pattern recognition, NLP, self-health monitoring, classification and staging of CKD, forecasting seizure
episodes, outlining the progress of previous studies, designing an implantable device, drug discovery and
possible adverse effects and big medical encoding. Fig. 11 shows an overview of the most popular
applications of distributed systems via DL techniques in the healthcare era. The most repeated application
in the studies was medical image analysis and COVID-19 detection (24% and 21%, respectively). Since
the pandemic of COVID-19 was a great burden for the whole globe and had huge impact on the
healthcare of all regions, it also resulted in various challenges in the medical healthcare era [126]. So, as a
result of the pandemic, the most studied ML approaches were in the field of early analysis of COVID-19
and better analysis of recorded medical images, especially CT-scan, which had a prominent effect on
efficient diagnosis and management of infected patients.
 Fig. 11. The applications of various distributed system environments in the context of medical data processing.

Perhaps one of the most significant topics for future study could be preserving the security and privacy
of individuals' identities and processed data. Fig. 12 provides a pie chart of the evaluated parameters in
the articles regarding the distribution of the parameters. Table 12 shows the overall parameters evaluated
in the investigated articles in this study. The utmost parameter evaluated in the papers is related to
accuracy and lower latency. Almost all analyzed studies had high accuracy, and about 80% had low
latency time. In the next stage, security was the most applied parameter; in the next stage was high
security and privacy setting with a distribution of about 25%. DL is becoming a trending concept due to
its remarkable accuracy when practiced with vast data like generated IoT. Therefore, high accuracy was a
remarkable parameter for selecting studied papers. Although, when trained upon more sensitive data, such
as medical data, DL algorithms tend to lose privacy, which can impact the model's efficiency, so higher
accuracy was of great importance in the selected articles for inclusion in this study.
Fig. 12. Evaluated parameters in the articles.
 Table 10. Evaluated parameters in the articles.

Delay/Qui

consumpt
Converge

Complexi

Robustne
Accuracy

Scalabilit
response

nce time

Security
Authors

Energy
Type

time
cker

ion
ty

ss
y
Lee, et al. [82]        
Egger, et al. [83]        
Cloud

Stirling, et al. [84]        

Aazam, et al. [85]        
Liu, et al. [86]        
Ma and Pang [87]        
Niu, et al. [88]        
Połap, et al. [89]        
Khan, et al. [90]        
IoT

Mahawaga Arachchige, et al.        

[91]
Mansour and Althobaiti [92]        
Warnat-Herresthal, et al. [93]        
Kong, et al. [94]        
Edge

Yang, et al. [95]        

Rahman, et al. [96]        
Park, et al. [127]        
Khan, et al. [97]       
Manocha, et al. [98]        
Shukla, et al. [99]        
Fog

Horng, et al. [100]        

Mohamed Akram, et al. [101]        
Singh, et al. [102]        
Fan, et al. [103]        
Hybrid

Arabi and Zaidi [104]        

Yuan, et al. [105]        
Al-Saffar and Yildirim [106]        
Momenzadeh, et al. [107]        
Seba and Benifa [108]        

7. Open Issues
Medical data processing and distribution have developed as an appropriate strategy for a variety of
emerging healthcare applications. However, several challenges remain throughout the setting of remotely
monitoring health and behavior, such as data collecting, modeling, and processing. Furthermore,
comprehending the regulatory environment is the only way to proceed with ongoing research and
development. There has been an increase in academic work published on topics ranging from philosophy,
concepts, and systems to deployments of large medical data platforms for ubiquitous healthcare services
as remote patient monitoring and pervasive healthcare ideas have advanced. This section provides major
challenges and concepts which have been discussed in the most recent research in this field.
7.1. The challenges and considerations in the implementation of AI applications in critical care
The use of AI applications in critical care presents a set of challenges that are similar to those faced in
other healthcare domains. These challenges encompass various aspects, including technology,
deployment concerns, and widespread acceptance. It is crucial to address these issues in-depth to ensure
the safe and effective implementation of ML models in healthcare settings. One important consideration
is the reliability and accuracy of AI models in critical care scenarios. Using ML algorithms for medical
data processing introduces potential errors or biases in the predictions or recommendations generated. The
lack of interpretability and explainability of some ML models poses challenges in understanding the
rationale behind their decisions, which is critical in the context of patient care. Additionally, the ethical
implications of using AI in critical care should be carefully examined. The responsible and ethical
deployment of AI technologies requires addressing issues such as patient privacy, data security, and
informed consent. Ensuring transparency and accountability in the decision-making process of AI systems
is vital to maintain the trust of healthcare professionals, patients, and other stakeholders.
Another aspect to consider is the potential unintended consequences of hasty implementation. Rushing
the integration of ML models into healthcare systems without thorough evaluation and validation may
lead to adverse outcomes. It is essential to conduct rigorous testing and validation processes to ensure AI
models' safety, reliability, and generalizability in critical care. Furthermore, relying on fog computing for
medical monitoring introduces its own challenges. One limitation is that patients cannot submit real-time
data for analytics if their smartphone or hub device is broken or not charged. Moreover, older medical
devices, such as blood glucose meters, often require manual data transfer through wired connections to
hubs or computers, hindering seamless and automated data analysis. Asynchronous telemedicine systems,
although not true fog devices, may delay the identification of critical patterns that necessitate immediate
response until data uploading is completed. In recent years, wireless connectivity capabilities have
superseded these outdated devices, enabling real-time data transmission. However, addressing the
limitations and challenges associated with data transfer and connectivity in critical care settings remains
crucial to ensure timely and accurate medical data analysis. In conclusion, implementing AI applications
in critical care entails various challenges and considerations, including the technology's reliability, ethical
implications, potential unintended consequences, and limitations associated with fog computing and data
transfer. A thorough examination of these issues is necessary to develop robust and safe AI systems for
medical data processing in critical care environments.
7.2. Unregulated development and deployment of ML methods and concerns in clinical
adoption
The unregulated development and deployment of ML methods in the context of medical data processing
based on distributed computing and the IoT raise several critical concerns that need to be explored in-
depth. These issues revolve around patient care, the accountability of healthcare professionals and
software licensing, ML methods' adaptability, and patient data's security.
One significant concern is the potential for overdiagnosis and excessive testing resulting from
unregulated ML methods. While ML techniques have shown promise in minimizing redundant testing
and reducing costs, their inappropriate usage can lead to unnecessary interventions and increased
healthcare expenses. It is imperative to evaluate ML methods thoroughly to ensure their effectiveness and
safety in clinical adoption. Rigorous validation processes and adherence to high-level health system
validation and regulation are essential to protect patients' lives and well-being. The responsibility for
accurate diagnosis and management becomes complex in the context of ML-assisted decision-making.
The extent to which software tools provide decision assistance raises questions about the distribution of
accountability between healthcare professionals and software licensing. Inaccurate diagnosis or
suboptimal management could potentially be attributed to either party. Clear guidelines and frameworks
need to be established to define the boundaries of responsibility, ensuring that patients receive optimal
care and appropriate measures are taken in cases of diagnostic errors or mismanagement.
The architectural design and adaptability of ML methods used in medical data processing are critical
considerations. ML algorithms should be designed to evolve and adapt to changing clinical practices,
ensuring that they stay aligned with the latest advancements and guidelines. Failure to update and
incorporate new knowledge into ML models can disrupt their delivery and impact patient care.
Establishing mechanisms for continuous monitoring, maintenance, and updating of ML algorithms is vital
to guarantee their effectiveness and relevance in evolving clinical contexts. Another crucial issue is the
security of patient data and the potential access of ML algorithms to other sources. The unexpected
exposure of patient data to unauthorized sources raises privacy concerns and the risk of data breaches.
Additionally, ML algorithms must be robust and resilient against external influences, whether
unintentional or purposeful. Rigorous testing should be conducted to assess the sensitivity of ML
algorithms to external influences and to ensure their ability to withstand potential attacks or
manipulations.
Addressing these open issues requires the introduction of properly validated ML applications into
clinical settings. The integration of validated ML methods can provide insights and solutions to the
challenges in medical data processing. Additionally, a collaboration between regulatory bodies, healthcare
institutions, and ML developers is crucial to establish guidelines and regulations that govern the
development, deployment, and usage of ML methods in medical contexts. By delving deeper into these
concerns, researchers, practitioners, and policymakers can identify potential risks, develop appropriate
safeguards, and ensure the responsible and effective implementation of ML techniques in medical data
processing. By comprehensively addressing these open issues, the potential benefits of ML applications in
healthcare can be realized while maintaining patient safety, privacy, and the highest standards of care.
7.3. Challenges in Training Data Availability and Imbalanced Data in DL for Diagnostic
Imaging
One of the key challenges in the adoption of DL techniques in diagnostic imaging, within the context of
medical data processing, based on distributed computing and the (IoT is the requirement for large training
databases to demonstrate the reliability of DL classifications. However, the availability of appropriate
data sources poses a significant barrier to achieving this goal. The reliability and performance of DL
algorithms heavily rely on the quality and diversity of the training data. In the case of diagnostic imaging,
the generation of massive datasets that accurately represent a wide range of medical conditions is highly
complex. Annotating these datasets with ground truth labels requires substantial input from medical
specialists, as their expertise is essential for accurate annotation. Multiple expert views are often
necessary to address the issue of human error and ensure the quality of annotations.
However, the process of data annotation may not always be feasible due to several factors. Firstly,
there may be a scarcity of skilled specialists available to perform the annotations, especially in the case of
rare conditions or specialized areas of medicine. Limited access to experts who can provide accurate
annotations poses a significant challenge in generating the necessary training data for DL models.
Another critical problem is the imbalanced structure of data, which is particularly prevalent in the health
industry. Uncommon illnesses or conditions are often underrepresented in the available datasets by their
nature of being infrequent. If not appropriately addressed, this class imbalance can severely impact the
performance and reliability of DL models. DL algorithms tend to be biased towards the majority class,
leading to reduced accuracy and potentially missing crucial diagnoses or insights related to less common
conditions.
To overcome these challenges, various strategies can be employed. Firstly, efforts should be made to
expand the availability of comprehensive and diverse training datasets by collaborating with medical
institutions, research organizations, and healthcare providers. Creating partnerships that enable the
collecting and sharing of anonymized medical data can help address the scarcity of training data.
Addressing the issue of data annotation requires the development of semi-automated or assisted
annotation tools that can leverage the expertise of medical specialists more efficiently. These tools can
assist in the annotation process, reducing the burden on specialists and increasing the scalability of dataset
generation. Dealing with imbalanced data necessitates employing techniques such as oversampling,
undersampling, or generating synthetic data to balance the representation of different classes. These
approaches can help mitigate the negative impact of class imbalance on DL model performance and
ensure accurate predictions for both common and rare medical conditions.
Furthermore, collaborations between the DL research community and medical experts can facilitate
the development of specialized DL architectures and techniques tailored specifically for diagnostic
imaging. These approaches can leverage domain knowledge and expert insights to improve the reliability,
interpretability, and clinical relevance of DL models in medical applications. The adoption of DL
techniques in diagnostic imaging can be enhanced by addressing the challenges related to training data
availability and imbalanced data. These efforts will contribute to developing more reliable and accurate
DL models, ultimately leading to improved diagnostic accuracy, better patient care, and advancements in
medical data processing within the distributed computing and IoT paradigm.
7.4. Overcoming Challenges for Adoption and Collaboration in DL for Medical Imaging
The potential advantages of DL in medical data processing based on distributed computing and the IoT
are immense. However, several challenges and expenses are associated with the early efforts to
implement DL in medical imaging. Major players in the industry, including Google DeepMind, IBM
Watson, academic laboratories, top hospitals, and medical equipment suppliers such as Siemens, Philips,
Hitachi, and GE Healthcare, have made significant investments and are actively working towards the
integration of DL into medical imaging solutions. These efforts indicate the potential of DL to
revolutionize disease detection and treatment. However, despite the progress made by these substantial
stakeholders and their optimistic projections for the future of DL in medical imaging, concerns and
barriers still need to be addressed before widespread adoption can occur. One concern is the fear of
replacing human expertise with automated systems. While DL provides clear benefits in terms of disease
detection and treatment, there needs to be a thoughtful discussion and consideration of how DL can
complement and enhance the capabilities of healthcare professionals rather than replace them entirely.
Extensive collaboration is crucial to overcome these challenges and ensure the successful adoption of DL
in medical imaging. Cooperation between healthcare providers, medical equipment suppliers, and ML
experts is needed to inspire and empower the implementation of DL solutions. Such collaborations will
help address the shortage of accessible data for ML researchers, as healthcare data is often fragmented
and dispersed across various institutions. Moreover, the vast quantity of healthcare data, expected to
increase as the healthcare sector transitions to body sensor network-dependent monitoring, poses a
significant challenge. Sophisticated approaches and techniques are required to effectively manage,
analyze, and interpret this large volume of data. ML experts and data scientists need to develop advanced
algorithms, data processing methods, and scalable infrastructure to handle this growing data complexity.
Additionally, ethical and regulatory considerations play a crucial role in adopting DL in medical
imaging. Patient privacy, data security, and regulatory compliance are paramount concerns that must be
addressed to ensure the responsible and ethical use of DL techniques. Clear guidelines and regulations
need to be established to govern the collection, storage, sharing, and analysis of medical data to safeguard
patient confidentiality and maintain public trust. Addressing these challenges requires strong
collaboration, interdisciplinary efforts, and shared expertise from healthcare providers, medical
equipment suppliers, ML researchers, data scientists, and regulatory bodies. By working together, these
stakeholders can collectively overcome barriers, foster innovation, and create a framework for the
successful integration of DL techniques in medical data processing based on distributed computing and
IoT. The resulting advancements will improve the effectiveness of healthcare, enhance patient outcomes,
and revolutionize medical imaging practices.
 7.5. Cloud Computing, Networking, Security, and Privacy Challenges in Medical Data
Processing
The integration of cloud computing in EHR offers numerous benefits, such as easy access, data
replication, and secure transmission of personal health information. However, there are significant
challenges associated with networking breakdown, security, and privacy of patient data, which need to be
addressed to ensure the safe and responsible use of the cloud in medical data processing. One of the main
concerns is the potential abuse of patient data by unauthorized individuals, including consumers, hackers,
and malware. The sensitive nature of health information makes it a valuable target for malicious actors,
emphasizing the importance of robust security measures. Data breaches and unauthorized access to
patient records can lead to severe consequences, including identity theft, compromised patient
confidentiality, and the misuse of personal health information. Therefore, strict security protocols,
encryption mechanisms, access controls, and regular security audits are essential to protect patient data
and maintain trust in cloud computing solutions.
Furthermore, the reliability, security, and stability of the Internet of Medical Things (IoMT) pose
significant challenges. The IoMT refers to the network of interconnected medical devices, sensors, and
systems that collect and transmit health-related data. As more medical devices become connected,
ensuring their reliability and safeguarding against cybersecurity threats becomes crucial. The potential
vulnerabilities in IoMT devices can expose patients to risks, including unauthorized access, data
manipulation, and disruptions in healthcare delivery. Efforts should be made to establish robust security
frameworks, encryption protocols, and monitoring mechanisms to mitigate these risks and maintain the
integrity of the IoMT ecosystem. Data privacy and security concerns related to ML systems have gained
attention recently. Regulations such as the Health Insurance Portability and Accountability Act (HIPAA)
play a vital role in safeguarding the accessibility and appropriate use of health data. However, as new data
patterns emerge and new data partnerships are formed, regulations must evolve to address emerging
challenges. It is essential to balance facilitating the accessibility and sharing of health data for research
and innovation purposes while ensuring the privacy and protection of sensitive patient information.
Continuous advancements in security technologies, including encryption algorithms, authentication
mechanisms, and intrusion detection systems, are necessary to address these challenges. Collaborative
efforts between healthcare organizations, cloud service providers, regulatory bodies, and cybersecurity
experts are crucial to establishing industry standards, best practices, and guidelines for the secure
implementation of cloud computing in medical data processing. Regular security assessments, audits, and
training programs should be conducted to maintain a proactive stance against emerging threats and
vulnerabilities. By addressing the networking, security, and privacy challenges associated with cloud
computing and the IoMT, the applications of Machine Learning techniques in medical data processing
can benefit from the vast amounts of data collected while ensuring the privacy and confidentiality of
patient information. This will lead to improved healthcare outcomes, enhanced research capabilities, and
greater patient trust in the use of advanced technologies for medical data processing.

8. Conclusion and Limitations

The rapid development of information technology has led to advancements in machine learning, deep
learning, big data processing, and distributed systems, which have had significant and far-reaching effects
on the structure and efficiency of traditional healthcare businesses, as well as the design and maintenance
of contemporary medical management information systems. In this study, a total of 27 papers were
selected for analysis. Due to inclusion criteria, the selected articles were original research, written in
English, and published in high-ranked journals. Conference papers, editorial letters, and book chapters
were excluded. We presented an SLR for applications and methods of distributed computing platforms in
healthcare data analysis and performed a comprehensive overview of various present applications in 5
categorical environments underlying distributed systems such as cloud, IoT, edge, fog, and hybrid
systems. Most of the papers were published in 2021 (46%). IoT was the most discussed environment
(43%). Medical image analysis and COVID-19 detection were the most discussed applications, with a
total percentage of about 45%. Otherwise, high accuracy, high security and robustness, and low security
were the remarkable parameters that were evaluated in this study. The evolution of medical DL systems
has improved the uniformity and precision of clinical decision-making while expanding the data
collection capabilities of medical knowledge-based systems (Knowledge-based systems use a data
repository known as a knowledge base to provide a method for problem-solving [128]). Examples of
these systems include EHR, medical imaging technology, medical big data, intelligent medication design,
and innovative health management systems. These advancements may also help clinicians and researchers
optimize treatment programs and make decisions regarding the best therapy alternatives. As a subset of
ML, DL is the most prominent issue with various techniques such as computer vision, NLP, voice
recognition, visual object identification, disease prognosis, drug discovery, bioinformatics, biomedicine,
etc. Healthcare and scientific health implementations are significant in these industries' growth. As with
the machines’ processing capability development, several approaches were successfully established based
on ML and DL in medical fields as a classification and recognition platform. The massive big data
expansion, the IoT, connected devices, and HPC are the major factors why DL is so popular. Based on
their unique functions, medical IoT, digital pictures, EHR data, genetic data, and centralized medical
datasets are the primary data inputs for DL systems. However, several possible concerns like privacy,
QoS optimization, and implementation emphasize the vital element of DL. In conclusion, the use of data
mining and ML techniques can offer the medical staff insight beyond the field of profession, which will
help them make more reasonable decisions faster. Hence, these methods are not intended to replace the
position of academia and medical professionals but rather to provide a suggestion system for more
effectively allocating resources.
There are some limitations to this study. First of all, several of the investigated publications did not
include comprehensive descriptions of their methodologies or had challenges in data selection, power
consumption, or security mechanisms. Plus, most of the studied methods were done quickly due to the
urgent need for DL platforms in the recent pandemic. Otherwise, non-English documents were excluded
due to inaccessibility. Further studies for more detailed method descriptions, including all published
papers in recent years, are recommended to provide better insights into DL techniques in medical data
processing. Also, the future of DL techniques in medical data processing holds significant potential for
advancements. Researchers should focus on refining and enhancing DL methodologies to overcome
previous challenges, particularly in terms of improving the accuracy, robustness, and interpretability of
DL models in healthcare applications. Exploring novel applications, such as personalized medicine,
disease prediction, and treatment optimization, can greatly enhance patient care and outcomes. Ethical
and privacy implications should also be a crucial consideration in DL techniques for medical data
processing. Future research efforts should prioritize the development of robust security mechanisms and
privacy-preserving techniques to safeguard sensitive patient information while maintaining the
effectiveness and efficiency of DL-based healthcare systems.
-Conflict of Interest: The author declares no conflict of interest.
-Availability of materials: All data are reported in the paper.

References
[1] S. Vahdat, "Clinical profile, outcome and management of kidney disease in COVID-19 patients - a
narrative review," Eur Rev Med Pharmacol Sci, vol. 26, pp. 2188-2195, Mar 2022.
[2] F. A. Zadeh, D. O. Bokov, G. Yasin, S. Vahdat, and M. Abbasalizad-Farhangi, "Central obesity
accelerates leukocyte telomere length (LTL) shortening in apparently healthy adults: A
systematic review and meta-analysis," Critical Reviews in Food Science and Nutrition, pp. 1-10,
2021.
[3] L. Sun, M. Zhang, B. Wang, and P. Tiwari, "Few-Shot Class-Incremental Learning for Medical Time
Series Classification," IEEE Journal of Biomedical and Health Informatics, 2023.
[4] M. Wang and L. Li, "Research and Implementation of Distributed Computing Management
System for College Students' Sports Health Based on Integrated Regional Collaborative Medical
Care," Occupational Therapy International, vol. 2022, 2022.
[5] A. Heidari, N. Jafari Navimipour, M. Unal, and S. Toumaj, "Machine learning applications for
COVID-19 outbreak management," Neural Computing and Applications, 2022/06/10 2022.
[6] F. Liu, X. Zhao, Z. Zhu, Z. Zhai, and Y. Liu, "Dual-microphone active noise cancellation paved with
Doppler assimilation for TADS," Mechanical Systems and Signal Processing, vol. 184, p. 109727,
2023.
[7] P. Mylonas, "Exploring the Notion of Context in Medical Data," Adv Exp Med Biol, vol. 989, pp.
39-55, 2017.
[8] E. Zakhem, S. V. Murphy, M. L. Davis, S. Raghavan, M. T. J. B. e. Lam, and c. biology, "Image and
Video Acquisitonb and Processing for Clinical Application," vol. 7, ed: SAGE Publications Sage
UK: London, England, 2016, p. BECB. S40272.
[9] S. Vitabile, M. Marks, D. Stojanovic, S. Pllana, J. Molina, M. Krzysztoo, et al., "Medical Data
Processing and Analysis for Remote Health and Activities Monitoring," ed, 2019, pp. 186-220.
[10] Y. Zhuang, N. Jiang, and Y. Xu, "Progressive Distributed and Parallel Similarity Retrieval of Large
CT Image Sequences in Mobile Telemedicine Networks," Wireless Communications and Mobile
Computing, vol. 2022, 2022.
[11] W. M. Tierney, J. E. Sidle, L. O. Diero, A. Sudoi, J. Kiplagat, S. Macharia, et al., "Assessing the
impact of a primary care electronic medical record system in three Kenyan rural health centers,"
vol. 23, pp. 544-552, 2016.
[12] D. Luan, A. Liu, X. Wang, Y. Xie, and Z. Wu, "Robust two-stage location allocation for emergency
temporary blood supply in postdisaster," Discrete dynamics in nature and society, vol. 2022, pp.
1-20, 2022.
[13] S. Lu, J. Yang, B. Yang, Z. Yin, M. Liu, L. Yin, et al., "Analysis and Design of Surgical Instrument
Localization Algorithm," CMES-Computer Modeling in Engineering & Sciences, vol. 137, 2023.
[14] A. V. Stokes, "Distributed processing: a strategy for health care computing," Computer
Communications, vol. 8, pp. 306-315, 1985.
[15] M. Liu, X. Zhang, B. Yang, Z. Yin, S. Liu, L. Yin, et al., "Three-dimensional modeling of heart soft
tissue motion," Applied Sciences, vol. 13, p. 2493, 2023.
[16] A. Dhillon, A. J. J. o. B. Singh, and T. s. World, "Machine learning in healthcare data analysis: a
survey," vol. 8, pp. 1-10, 2019.
[17] Z. Zhang, L. Wang, W. Zheng, L. Yin, R. Hu, and B. Yang, "Endoscope image mosaic based on
pyramid ORB," Biomedical Signal Processing and Control, vol. 71, p. 103261, 2022.
[18] B. Cao, J. Zhao, Z. Lv, and P. Yang, "Diversified personalized recommendation optimization based
on mobile data," IEEE Transactions on Intelligent Transportation Systems, vol. 22, pp. 2133-
2139, 2020.
[19] H. Habibzadeh, K. Dinesh, O. R. Shishvan, A. Boggio-Dandry, G. Sharma, and T. Soyata, "A Survey
of Healthcare Internet of Things (HIoT): A Clinical Perspective," IEEE Internet of Things Journal,
vol. 7, pp. 53-71, 2020.
[20] Z. Lv, D. Chen, H. Feng, H. Zhu, and H. Lv, "Digital twins in unmanned aerial vehicles for rapid
medical resource delivery in epidemics," IEEE Transactions on Intelligent Transportation
Systems, vol. 23, pp. 25106-25114, 2021.
[21] J. C. Sandberg, Y. Ge, H. T. Nguyen, T. A. Arcury, A. J. Johnson, W. Hwang, et al., "Insight into the
sharing of medical images: Physician, other health care providers, and staff experience in a
variety of medical settings," Applied Clinical Informatics, vol. 3, pp. 475-487, 2012.
[22] W. Sun, Z. Cai, Y. Li, F. Liu, S. Fang, and G. Wang, "Data processing and text mining technologies
on electronic medical records: A review," Journal of Healthcare Engineering, vol. 2018, 2018.
[23] K. Adane, D. Muluye, and M. Abebe, "Processing medical data: A systematic review," Archives of
Public Health, vol. 71, 2013.
[24] Y. Kuroda, G. Yamamoto, and T. Kuroda, "Data Processing Model for Compliance with
International Medical Research Data Processing Rules," Advanced Biomedical Engineering, vol.
11, pp. 48-57, 2022.
[25] K. Adane, M. Gizachew, and S. Kendie, "The role of medical data in efficient patient care
delivery: a review," Risk Manag Healthc Policy, vol. 12, pp. 67-73, 2019.
[26] Y. W. Chow, T. Treasure, and S. Gallivan, "Can the internet be used to collect clinical data for
research?," Health Informatics Journal, vol. 10, pp. 77-82, 2004.
[27] A. Jarynowski, D. Marchewka, and A. J. a. p. a. Buda, "Internet-assisted risk assessment of
infectious diseases in women sexual and reproductive health," 2018.
[28] K. Muhammad, S. Khan, J. Del Ser, V. H. C. J. I. T. o. N. N. De Albuquerque, and L. Systems, "Deep
learning for multigrade brain tumor classification in smart healthcare systems: A prospective
survey," vol. 32, pp. 507-522, 2020.
[29] Q. Yuan, B. Kato, K. Fan, and Y. Wang, "Phased array guided wave propagation in curved plates,"
Mechanical Systems and Signal Processing, vol. 185, p. 109821, 2023.
[30] X. Liu, J. Zhao, J. Li, B. Cao, and Z. Lv, "Federated neural architecture search for medical data
security," IEEE transactions on industrial informatics, vol. 18, pp. 5628-5636, 2022.
[31] A. Heidari, S. Toumaj, N. J. Navimipour, and M. Unal, "A privacy-aware method for COVID-19
detection in chest CT images using lightweight deep conventional neural network and
blockchain," Computers in Biology and Medicine, p. 105461, 2022/03/28/ 2022.
[32] H. Zunair, A. B. J. S. n. a. Hamza, and mining, "Synthesis of COVID-19 chest X-rays using unpaired
image-to-image translation," vol. 11, pp. 1-12, 2021.
[33] L. Li, P. Wang, X. Zheng, Q. Xie, X. Tao, and J. D. Velásquez, "Dual-interactive fusion for code-
mixed deep representation learning in tag recommendation," Information Fusion, p. 101862,
2023.
[34] X. Deng, E. Liu, S. Li, Y. Duan, and M. Xu, "Interpretable Multi-modal Image Registration Network
Based on Disentangled Convolutional Sparse Coding," IEEE Transactions on Image Processing,
vol. 32, pp. 1078-1091, 2023.
[35] M. Abbasi, A. Shahraki, and A. J. C. C. Taherkordi, "Deep learning for network traffic monitoring
and analysis (NTMA): A survey," vol. 170, pp. 19-41, 2021.
[36] Z. Gao, X. Pan, J. Shao, X. Jiang, Z. Su, K. Jin, et al., "Automatic interpretation and clinical
evaluation for fundus fluorescein angiography images of diabetic retinopathy patients by deep
learning," British Journal of Ophthalmology, 2022.
[37] F. Li, Z. Qu, and R. Li, "Medical cloud computing data processing to optimize the effect of drugs,"
Journal of Healthcare Engineering, vol. 2021, 2021.
[38] C. Gong, J. Liu, Q. Zhang, H. Chen, and Z. Gong, "The Characteristics of Cloud Computing," in
2010 39th International Conference on Parallel Processing Workshops, 2010, pp. 275-279.
[39] W. Shi, J. Cao, Q. Zhang, Y. Li, and L. Xu, "Edge Computing: Vision and Challenges," IEEE Internet
of Things Journal, vol. 3, pp. 637-646, 2016.
[40] S. Yi, C. Li, and Q. Li, "A Survey of Fog Computing: Concepts, Applications and Issues," presented
at the Proceedings of the 2015 Workshop on Mobile Big Data, Hangzhou, China, 2015.
[41] A. A. Khan, Z. A. Shaikh, L. Baitenova, L. Mutaliyeva, N. Moiseev, A. Mikhaylov, et al., "QoS-
ledger: Smart contracts and metaheuristic for secure quality-of-service and cost-efficient
scheduling of medical-data processing," Electronics (Switzerland), vol. 10, 2021.
[42] K. Adane, D. Muluye, and M. Abebe, "Processing medical data: a systematic review," Arch Public
Health, vol. 71, p. 27, Oct 9 2013.
[43] L. Papp, C. P. Spielvogel, I. Rausch, M. Hacker, and T. Beyer, "Personalizing medicine through
hybrid imaging and medical big data analysis," Frontiers in Physics, vol. 6, 2018.
[44] X. Dai, Z. Xiao, H. Jiang, M. Alazab, J. C. Lui, G. Min, et al., "Task offloading for cloud-assisted fog
computing with dynamic service caching in enterprise management systems," IEEE Transactions
on Industrial Informatics, vol. 19, pp. 662-672, 2022.
[45] F. Hu, X. Shi, H. Wang, N. Nan, K. Wang, S. Wei, et al., "Is health contagious?—Based on
empirical evidence from China family panel studies' data," Frontiers in Public Health, vol. 9, p.
691746, 2021.
[46] M. Esmailiyan, A. Amerizadeh, S. Vahdat, M. Ghodsi, R. I. Doewes, and Y. Sundram, "Effect of
different types of aerobic exercise on individuals with and without hypertension: An updated
systematic review," Current problems in cardiology, p. 101034, 2021.
[47] S. Vahdat and S. Shahidi, "D-dimer levels in chronic kidney illness: a comprehensive and
systematic literature review," proceedings of the national academy of sciences, india section b:
biological sciences, vol. 90, pp. 911-928, 2020.
[48] T. L. R. J. T. L. R. m. Medicine, "Opening the black box of machine learning," vol. 6, ed, 2018, p.
801.
[49] G. Y. Lip, R. Nieuwlaat, R. Pisters, D. A. Lane, and H. J. Crijns, "Refining clinical risk stratification
for predicting stroke and thromboembolism in atrial fibrillation using a novel risk factor-based
approach: the euro heart survey on atrial fibrillation," Chest, vol. 137, pp. 263-72, Feb 2010.
[50] C. O'Mahony, F. Jichi, M. Pavlou, L. Monserrat, A. Anastasakis, C. Rapezzi, et al., "A novel clinical
risk prediction model for sudden cardiac death in hypertrophic cardiomyopathy (HCM risk-
SCD)," Eur Heart J, vol. 35, pp. 2010-20, Aug 7 2014.
[51] R. C. Deo, "Machine Learning in Medicine," Circulation, vol. 132, pp. 1920-30, Nov 17 2015.
[52] C. Deng, X. Ji, C. Rainey, J. Zhang, and W. Lu, "Integrating Machine Learning with Human
Knowledge," iScience, vol. 23, p. 101656, 2020/11/20/ 2020.
[53] X. Li and Y. Sun, "Application of RBF neural network optimal segmentation algorithm in credit
rating," Neural Computing and Applications, vol. 33, pp. 8227-8235, 2021.
[54] A. Heidari, N. Jafari Navimipour, M. Unal, and G. Zhang, "Machine learning applications in
internet-of-drones: systematic review, recent deployments, and open issues," ACM Computing
Surveys, vol. 55, pp. 1-45, 2023.
[55] E. Mekov, M. Miravitlles, and R. Petkov, "Artificial intelligence and machine learning in
respiratory medicine," Expert Review of Respiratory Medicine, vol. 14, pp. 559-564, 2020/06/02
2020.
[56] G. Quer, R. Arnaout, M. Henne, and R. Arnaout, "Machine Learning and the Future of
Cardiovascular Care: JACC State-of-the-Art Review," J Am Coll Cardiol, vol. 77, pp. 300-313, Jan
26 2021.
[57] L. Alzubaidi, J. Zhang, A. J. Humaidi, A. Al-Dujaili, Y. Duan, O. Al-Shamma, et al., "Review of deep
learning: Concepts, CNN architectures, challenges, applications, future directions," vol. 8, pp. 1-
74, 2021.
[58] H.-P. Chan, R. K. Samala, L. M. Hadjiiski, and C. Zhou, "Deep Learning in Medical Image Analysis,"
Advances in experimental medicine and biology, vol. 1213, pp. 3-21, 2020.
[59] S. Feng, C. Hategeka, and K. A. J. P. h. m. Grépin, "Addressing missing values in routine health
information system data: an evaluation of imputation methods using data from the Democratic
Republic of the Congo during the COVID-19 pandemic," vol. 19, pp. 1-14, 2021.
[60] L. Tan, K. Yu, A. K. Bashir, X. Cheng, F. Ming, L. Zhao, et al., "Toward real-time and efficient
cardiovascular monitoring for COVID-19 patients by 5G-enabled wearable medical devices: a
deep learning approach," Neural Comput Appl, pp. 1-14, Jul 4 2021.
[61] H. Zhang, H. Zhang, S. Pirbhulal, W. Wu, and V. H. C. D. Albuquerque, "Active Balancing
Mechanism for Imbalanced Medical Data in Deep Learning–Based Classification Models," vol.
16, p. Article 39, 2020.
[62] A. A. Khan, Z. A. Shaikh, L. Baitenova, L. Mutaliyeva, N. Moiseev, A. Mikhaylov, et al., "QoS-
Ledger: Smart Contracts and Metaheuristic for Secure Quality-of-Service and Cost-Efficient
Scheduling of Medical-Data Processing," vol. 10, p. 3083, 2021.
[63] X. Wang, L. T. Yang, Y. Wang, L. Ren, and M. J. Deen, "ADTT: A Highly Efficient Distributed
Tensor-Train Decomposition Method for IIoT Big Data," IEEE Transactions on Industrial
Informatics, vol. 17, pp. 1573-1582, 2021.
[64] M. Shirasuna, "Data Analysis and System Development for Medical Professionals on Sleep Apnea
Syndrome and Orthostatic Dysregulation by Processing-Healthcare Professionals and Patients,"
vol. 22, p. 1254, 2022.
[65] W. Raghupathi and V. Raghupathi, "Big data analytics in healthcare: promise and potential,"
Health Inf Sci Syst, vol. 2, p. 3, 2014.
[66] W. Sun, Z. Cai, Y. Li, F. Liu, S. Fang, and G. Wang, "Data Processing and Text Mining Technologies
on Electronic Medical Records: A Review," Journal of Healthcare Engineering, vol. 2018, p.
4302425, 2018/04/08 2018.
[67] H. Jiang, Z. Xiao, Z. Li, J. Xu, F. Zeng, and D. Wang, "An energy-efficient framework for internet of
things underlaying heterogeneous small cell networks," IEEE Transactions on Mobile Computing,
vol. 21, pp. 31-43, 2020.
[68] B. Cao, X. Wang, W. Zhang, H. Song, and Z. Lv, "A many-objective optimization model of
industrial internet of things based on private blockchain," IEEE Network, vol. 34, pp. 78-83, 2020.
[69] B. Cheng, D. Zhu, S. Zhao, and J. Chen, "Situation-aware IoT service coordination using the
event-driven SOA paradigm," IEEE Transactions on Network and Service Management, vol. 13,
pp. 349-361, 2016.
[70] Z. Lv and H. Song, "Mobile internet of things under data physical fusion technology," IEEE
Internet of Things Journal, vol. 7, pp. 4616-4624, 2019.
[71] D. C. Klonoff, "Fog Computing and Edge Computing Architectures for Processing Data From
Diabetes Devices Connected to the Medical Internet of Things," vol. 11, pp. 647-652, 2017.
[72] D. C. Klonoff, "Fog Computing and Edge Computing Architectures for Processing Data from
Diabetes Devices Connected to the Medical Internet of Things," Journal of Diabetes Science and
Technology, vol. 11, pp. 647-652, 2017.
[73] M. H. Sarhan, M. A. Nasseri, D. Zapp, M. Maier, C. P. Lohmann, N. Navab, et al., "Machine
learning techniques for ophthalmic data processing: A review," vol. 24, pp. 3338-3350, 2020.
[74] J. Kaur and P. Kaur, "Outbreak COVID-19 in Medical Image Processing Using Deep Learning: A
State-of-the-Art Review," Arch Comput Methods Eng, pp. 1-32, Oct 19 2021.
[75] M. H. Kashani, M. Madanipour, M. Nikravan, P. Asghari, E. J. J. o. N. Mahdipour, and C.
Applications, "A systematic review of IoT in healthcare: Applications, techniques, and trends,"
vol. 192, p. 103164, 2021.
[76] A. Gatouillat, Y. Badr, B. Massot, and E. Sejdid, "Internet of Medical Things: A Review of Recent
Contributions Dealing With Cyber-Physical Systems in Medicine," IEEE Internet of Things Journal,
vol. 5, pp. 3810-3822, 2018.
[77] E. Mbunge, S. e. Jiyane, and B. Muchemwa, "Towards emotive sensory Web in virtual health
care: Trends, technologies, challenges and ethical issues," Sensors International, vol. 3, p.
100134, 2022/01/01/ 2022.
[78] D. Ben-Israel, W. B. Jacobs, S. Casha, S. Lang, W. H. A. Ryu, M. de Lotbiniere-Bassett, et al., "The
impact of machine learning on patient care: a systematic review," vol. 103, p. 101785, 2020.
[79] A. Heidari, N. Jafari Navimipour, M. Unal, and S. Toumaj, "The COVID-19 epidemic analysis and
diagnosis using deep learning: A systematic literature review and future directions," Computers
in Biology and Medicine, vol. 141, p. 105141, 2022/02/01/ 2022.
[80] H. R. Torres, P. Morais, B. Oliveira, C. Birdir, M. Rüdiger, J. C. Fonseca, et al., "A review of image
processing methods for fetal head and brain analysis in ultrasound images," Computer Methods
and Programs in Biomedicine, vol. 215, p. 106629, 2022/03/01/ 2022.
[81] S. Vahdat, "The role of IT-based technologies on the management of human resources in the
COVID-19 era," Kybernetes, 2021.
[82] H. Lee, J. Kang, and J. Yeo, "Medical specialty recommendations by an artificial intelligence
chatbot on a smartphone: Development and deployment," Journal of Medical Internet Research,
vol. 23, 2021.
[83] J. Egger, D. Wild, M. Weber, C. A. R. Bedoya, F. Karner, A. Prutsch, et al., "Studierfenster: an
Open Science Cloud-Based Medical Imaging Analysis Platform," Journal of Digital Imaging, vol.
35, pp. 340-355, 2022.
[84] R. E. Stirling, M. I. Maturana, P. J. Karoly, E. S. Nurse, K. McCutcheon, D. B. Grayden, et al.,
"Seizure Forecasting Using a Novel Sub-Scalp Ultra-Long Term EEG Monitoring System,"
Frontiers in Neurology, vol. 12, 2021.
[85] M. Aazam, S. U. Islam, S. T. Lone, and A. Abbas, "Cloud of Things (CoT): Cloud-Fog-IoT Task
Offloading for Sustainable Internet of Things," IEEE Transactions on Sustainable Computing, vol.
7, pp. 87-98, 2022.
[86] J. Liu, X. Liang, W. Ruan, and B. Zhang, "High-performance medical data processing technology
based on distributed parallel machine learning algorithm," Journal of Supercomputing, vol. 78,
pp. 5933-5956, 2022.
[87] H. Ma and X. Pang, "Research and analysis of sport medical data processing algorithms based on
deep learning and internet of things," IEEE Access, vol. 7, pp. 118839-118849, 2019.
[88] S. Niu, M. Liu, Y. Liu, J. Wang, and H. Song, "Distant Domain Transfer Learning for Medical
Imaging," IEEE Journal of Biomedical and Health Informatics, vol. 25, pp. 3784-3793, 2021.
[89] D. Połap, G. Srivastava, and K. Yu, "Agent architecture of an intelligent medical system based on
federated learning and blockchain technology," Journal of Information Security and Applications,
vol. 58, 2021.
[90] M. A. Khan, K. Muhammad, M. Sharif, T. Akram, and S. Kadry, "Intelligent fusion-assisted skin
lesion localization and classification for smart healthcare," Neural Computing and Applications,
2021/10/17 2021.
[91] P. C. Mahawaga Arachchige, P. Bertok, I. Khalil, D. Liu, S. Camtepe, and M. Atiquzzaman, "Local
Differential Privacy for Deep Learning," IEEE Internet of Things Journal, vol. 7, pp. 5827-5842,
2020.
[92] R. F. Mansour and M. M. Althobaiti, "Cognitive Computing-Based Mammographic Image
Classification on an Internet of Medical," Computers, Materials and Continua, vol. 72, pp. 3945-
3959, 2022.
[93] S. Warnat-Herresthal, H. Schultze, K. L. Shastry, S. Manamohan, S. Mukherjee, V. Garg, et al.,
"Swarm Learning for decentralized and confidential clinical machine learning," Nature, vol. 594,
pp. 265-270, 2021/06/01 2021.
[94] X. Kong, K. Wang, S. Wang, X. Wang, X. Jiang, Y. Guo, et al., "Real-Time Mask Identification for
COVID-19: An Edge-Computing-Based Deep Learning Framework," IEEE Internet of Things
Journal, vol. 8, pp. 15929-15938, 2021.
[95] L. Yang, Z. Li, S. Ma, and X. Yang, "Artificial intelligence image recognition based on 5G deep
learning edge algorithm of Digestive endoscopy on medical construction," Alexandria
Engineering Journal, vol. 61, pp. 1852-1863, 2022.
[96] M. J. Rahman, B. I. Morshed, B. Harmon, and M. Rahman, "A pilot study towards a smart-health
framework to collect and analyze biomarkers with low-cost and flexible wearables," Smart
Health, vol. 23, 2022.
[97] M. A. Khan, S. Abbas, A. Atta, A. Ditta, H. Alquhayz, M. F. Khan, et al., "Intelligent cloud based
heart disease prediction system empowered with supervised machine learning," Computers,
Materials and Continua, vol. 65, pp. 139-151, 2020.
[98] A. Manocha, G. Kumar, M. Bhatia, and A. Sharma, "Video-assisted smart health monitoring for
affliction determination based on fog analytics," Journal of Biomedical Informatics, vol. 109,
2020.
[99] S. Shukla, M. F. Hassan, M. K. Khan, L. T. Jung, and A. Awang, "An analytical model to minimize
the latency in healthcare internet-of-things in fog computing environment," PLoS ONE, vol. 14,
2019.
[100] G. J. Horng, M. X. Liu, and C. C. Hsu, "The anomaly detection mechanism using deep learning in a
limited amount of data for fog networking," Computer Communications, vol. 170, pp. 130-143,
2021.
[101] K. Mohamed Akram, S. Sihem, K. Okba, and S. Harous, "IoMT-fog-cloud based architecture for
Covid-19 detection," Biomedical Signal Processing and Control, vol. 76, 2022.
[102] P. D. Singh, R. Kaur, K. D. Singh, G. Dhiman, and M. Soni, "Fog-centric IoT based smart healthcare
support service for monitoring and controlling an epidemic of Swine Flu virus," Informatics in
Medicine Unlocked, vol. 26, 2021.
[103] B. Fan, W. Fan, C. Smith, and H. Garner, "Adverse drug event detection and extraction from
open data: A deep learning approach," Information Processing and Management, vol. 57, 2020.
[104] H. Arabi and H. Zaidi, "Deep learning-guided estimation of attenuation correction factors from
time-of-flight PET emission data," Medical Image Analysis, vol. 64, 2020.
[105] A. Y. Yuan, Y. Gao, L. Peng, L. Zhou, J. Liu, S. Zhu, et al., "Hybrid deep learning network for
vascular segmentation in photoacoustic imaging," Biomedical Optics Express, vol. 11, pp. 6445-
6457, 2020.
[106] Z. A. Al-Saffar and T. Yildirim, "A hybrid approach based on multiple Eigenvalues selection (MES)
for the automated grading of a brain tumor using MRI," Computer Methods and Programs in
Biomedicine, vol. 201, 2021.
[107] N. Momenzadeh, H. Hafezalseheh, M. R. Nayebpour, M. Fathian, and R. Noorossana, "A hybrid
machine learning approach for predicting survival of patients with prostate cancer: A SEER-
based population study," Informatics in Medicine Unlocked, vol. 27, 2021.
[108] P. A. Seba and J. V. B. Benifa, "A Hybrid Analytic Model for the Effective Prediction of Different
Stages in Chronic Kidney Ailments," Wireless Personal Communications, 2022.
[109] S. Asghari and N. J. J. M. J. o. M. P. Navimipour, "Review and comparison of meta-heuristic
algorithms for service composition in cloud computing," vol. 4, 2015.
[110] M. Chiregi, N. J. J. J. o. E. S. Navimipour, and I. Technology, "Cloud computing and trust
evaluation: A systematic literature review of the state-of-the-art mechanisms," vol. 5, pp. 608-
622, 2018.
[111] M. Chiregi and N. J. J. J. o. S. S. R. Navimipour, "A comprehensive study of the trust evaluation
mechanisms in the cloud computing," vol. 9, pp. 1-30, 2017.
[112] R. Bernsteiner, D. Kilian, and B. Ebersberger, "Mobile Cloud Computing for Enterprise Systems: A
Conceptual Framework for Research," International Journal of Interactive Mobile Technologies
(iJIM), vol. 10, p. 72, 04/28 2016.
[113] C. Doukas, T. Pliakas, and I. Maglogiannis, "Mobile healthcare information management utilizing
Cloud Computing and Android OS," Annu Int Conf IEEE Eng Med Biol Soc, vol. 2010, pp. 1037-40,
2010.
[114] S. F. Wamba, S. Akter, A. Edwards, G. Chopin, and D. J. I. J. o. P. E. Gnanzou, "How ‘big data’can
make big impact: Findings from a systematic review and a longitudinal case study," vol. 165, pp.
234-246, 2015.
[115] O. Abedinia, M. Bekravi, and N. Ghadimi, "Intelligent Controller Based Wide-Area Control in
Power System," International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems,
vol. 25, pp. 1-30, 02/01 2017.
[116] L. Mainetti, L. Patrono, and A. Vilei, "Evolution of wireless sensor networks towards the internet
of things: A survey," in SoftCOM 2011, 19th international conference on software,
telecommunications and computer networks, 2011, pp. 1-6.
[117] A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, and M. Ayyash, "Internet of Things: A
Survey on Enabling Technologies, Protocols, and Applications," IEEE Communications Surveys &
Tutorials, vol. 17, pp. 2347-2376, 2015.
[118] F. Al-Turjman and S. Alturjman, "Context-Sensitive Access in Industrial Internet of Things (IIoT)
Healthcare Applications," IEEE Transactions on Industrial Informatics, vol. 14, pp. 2736-2744,
2018.
[119] P. Asghari, A. M. Rahmani, and H. H. S. Javadi, "Internet of Things applications: A systematic
review," Computer Networks, vol. 148, pp. 241-261, 2019/01/15/ 2019.
[120] S. B. Baker, W. Xiang, and I. Atkinson, "Internet of Things for Smart Healthcare: Technologies,
Challenges, and Opportunities," IEEE Access, vol. 5, pp. 26521-26544, 2017.
[121] L. Sun, L. Sun, X. Jiang, H. Ren, H. Ren, and Y. Guo, "Edge-Cloud Computing and Artificial
Intelligence in Internet of Medical Things: Architecture, Technology and Application," IEEE
Access, vol. 8, pp. 101079-101092, 2020.
[122] X. Liu, T. Qiu, X. Zhou, T. Wang, L. Yang, and V. J. I. T. o. I. I. Chang, "Latency-aware path planning
for disconnected sensor networks with mobile sinks," vol. 16, pp. 350-361, 2019.
[123] F. Bonomi, R. Milito, J. Zhu, and S. Addepalli, "Fog computing and its role in the internet of
things," in Proceedings of the first edition of the MCC workshop on Mobile cloud computing,
2012, pp. 13-16.
[124] X. Li and Y. Sun, "Stock intelligent investment strategy based on support vector machine
parameter optimization algorithm," Neural Computing and Applications, vol. 32, pp. 1765-1775,
2020.
[125] L. E. Juarez-Orozco, O. Martinez-Manzanera, S. V. Nesterov, S. Kajander, and J. J. E. J. o. H. I.
Knuuti, "The machine learning horizon in cardiac hybrid imaging," vol. 2, pp. 1-15, 2018.
[126] X. Xie, X. Jin, G. Wei, and C.-T. Chang, "Monitoring and Early Warning of SMEs’ Shutdown Risk
under the Impact of Global Pandemic Shock," Systems, vol. 11, p. 260, 2023.
[127] J. Park, S. Samarakoon, M. Bennis, and M. Debbah, "Wireless Network Intelligence at the Edge,"
Proceedings of the IEEE, vol. 107, pp. 2204-2239, 2019.
[128] C. Zheng, Y. An, Z. Wang, X. Qin, B. Eynard, M. Bricogne, et al., "Knowledge-based engineering
approach for defining robotic manufacturing system architectures," International Journal of
Production Research, pp. 1-19, 2022.
Declaration of interests

☒ The authors declare that they have no known competing financial interests or personal relationships
that could have appeared to influence the work reported in this paper.

☐ The authors declare the following financial interests/personal relationships which may be considered
as potential competing interests: