Hindawi

Journal of Healthcare Engineering

Volume 2022, Article ID 8169203, 11 pages
https://doi.org/10.1155/2022/8169203

Review Article
Deep Learning in Healthcare System for Quality of Service

Dibyahash Bordoloi ,1 Vijay Singh ,1 Sumaya Sanober ,2 Seyed Mohamed Buhari ,3

Javed Ahmed Ujjan ,4,5 and Rajasekhar Boddu 6
1
Department of Computer Science and Engineering, Graphic Era University, Dehradun, Uttarakhand, India
2
Prince Sattam Bin Abdul Aziz University, Wadi Al Dawasir 1191, Saudi Arabia
3
Department of Information Technology, King Abdulaziz University, PO Box 80221, Jeddah 21589, Saudi Arabia
4
Department of Zoology, Shah Abdul Latif University Khairpur, Khairpur, Sindh, Pakistan
5
College of Animal Sciences and Technology, Northwest A and F University, Xianyang, China
6
Department of Software Engineering, College of Computing and Informatics, Haramaya University, Dire Dawa, Ethiopia

Correspondence should be addressed to Sumaya Sanober; [email protected] and Rajasekhar Boddu; rajsekhar.boddu@
haramaya.edu.et

Received 29 December 2021; Accepted 29 January 2022; Published 8 March 2022

Academic Editor: Balakrishnan Nagaraj

Copyright © 2022 Dibyahash Bordoloi et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Deep learning (DL) and machine learning (ML) have a pivotal role in logistic supply chain management and smart manufacturing
with proven records. The ability to handle large complex data with minimal human intervention made DL and ML a success in the
healthcare systems. In the present healthcare system, the implementation of ML and DL is extensive to achieve a higher quality of
service and quality of health to patients, doctors, and healthcare professionals. ML and DL were found to be effective in disease
diagnosis, acute disease detection, image analysis, drug discovery, drug delivery, and smart health monitoring. This work presents
a state-of-the-art review on the recent advancements in ML and DL and their implementation in the healthcare systems for
achieving multi-objective goals. A total of 10 papers have been thoroughly reviewed that presented novel works of ML and DL
integration in the healthcare system for achieving various targets. This will help to create reference data that can be useful for
future implementation of ML and DL in other sectors of healthcare system.

1. Introduction physicians to give better care. Patient engagement and

satisfaction have increased as interaction with doctors has
Industry 5.0 and 5G wireless communication has led to the become simpler and more efficient. Furthermore, remote
development of cost-effective sensors, thereby leading to the monitoring of a patient’s health aids in the reduction in
emergence of Internet of things (IoT). It has an indis- hospital visits and the avoidance of readmissions. IoT has a
pensable role in today’s healthcare system. Internet of big influence on lowering healthcare expenses and in-
medical things (IoMT) is a crucial component of the modern creasing treatment results. The Internet of things (IoT) is
healthcare system [1]. IoT is an ever-expanding, limitless clearly transforming the healthcare industry by redefining
ecosystem that integrates software, hardware, or any other the role of gadgets and human touch in medical solution
device equipment that collects or exchanges data. IoT in the delivery [2, 3]. IoMT of wide range is useful in patient
healthcare service network is known as IoMT. The need for conditioning, monitoring, screening, personalized medicine,
telemedicine, remote patient monitoring, automated diag- integrating healthcare professionals, doctors, and hospital
nosis, detection, and treatment of acute diseases has become networks to patients for remote telemedicines, and patient
eminent, especially after the COVID-19 pandemic. Moni- monitoring. As the IoMT and IoT increase, the data gen-
toring systems in the healthcare business are now possible erated by these devices also increase. This substantial
thanks to Internet of things (IoT)-enabled devices, which amount of data is generally stored and regulated on the
can keep patients safe and healthy while also allowing cloud servers. Health data are useful in many ways when
2 Journal of Healthcare Engineering

processed intelligently [4, 5]. The intelligent processing of 1.1. Advantages of DL and ML in Medical Healthcare Data.
health data may help to prevent pandemic outbreaks in a The deep learning and machine learning are evolutionary
city, states, or countries, helping to identify acute diseases changes in various fields such as industry, companies,
such as Alzheimer’s disease, diabetes, cardiovascular dis- schools, colleges, and healthcare systems, and we can say that
eases, and lung cancer [6]. Blockchain is more reliable and more changes are seen in the medical line by providing many
robust as once the data are created, it can only be read and types of online and offline facilities. Deep learning plays an
not edited or updated. Blockchain also facilitates time important role in the detection of cancer cells automatically.
stamping along with the use of asymmetric cryptography for The ML can be solving multiple tasks but it needs human
secure transactions in the trusted network. In addition to beings, while the DL performs alone or automatically using
that, it also provides data for the development of DL/ML machine learning. Deep learning solves the whole problem
models. The transactions do not require additional protocols unlike ML automatically. Deep learning is more beneficial
that make blockchain an adaptive nature that can be inte- for elders, coma patients, and cardiac disease diagnosis,
grated with different types of data, platforms, and devices. especially in case of children [12]. The application of AI and
The health data obtained from the electronic health record, other approaches is shown in Figure 2.
genomic sequence, and digital health wearable are so great Deep learning is beneficial for the recognition of any
and cannot be reviewed easily without the intervention of problem, and the data-driven performance is also better in
artificial intelligence, deep learning, or artificial neuro- any type of problem. The learning algorithms are simple, i.e.,
networking [7–9]. ML, DL, and AI have a proven record of easily learned or performed and easily able for application.
demonstrating the capacity to translate large unmanaged The machine learning is nonlinear, and traditional deep
data into actionable knowledge in various fields of agri- learning is able to make fast and powerful machine learning
culture, supply chain management, quality control, sales for language learning and is responsible for accurate pa-
forecast, security, fraud detection, and other major fields. tient’s data. The representation of patient’s information or
Deep learning is the part or branch of machine learning. data in various types, i.e., images and text, can be learned like
Well with the growing number of machine learning tech- the real-time condition of the patients [13]. Recent deep
niques in health care, we can see a future in which data, learning models for healthcare applications, on the other
analysis, and invention collaborate to help countless indi- hand, are sophisticated and need a great amount of com-
viduals without their knowledge. Soon, ML-based applica- puting resources for training and prediction. It also takes a
tions packed with real-time patient data from many long time to train and evaluate data using these sophisticated
healthcare systems across several nations will be ubiquitous, neural networks. The more advanced the network and the
increasing the efficacy of hitherto inaccessible treatment longer the forecast period, the higher the accuracy required.
options [10, 11]. Machine algorithm learning is discovered This has been a significant issue in healthcare and other IoT
by John McCarthy, and the division of machine learning is applications where real-time findings are crucial [14, 15].
discovered by Arthur Samuel. ML and DL are the ap- The deep learning is gifted to deliver the potential for en-
proaches of AI, where AI can be defined as the artificial hancements in worth control engineering system. Machine
creation of a human brain line system that can learn natural learning is responsible to provide beneficial tools for un-
language and plan, perceive, and process it for making derstanding the domain. The ML gives the chances for the
decisions with minimal intervention of human. Many in- examination of the dynamic system and able to change the
dividuals, organizations, small businesses, and government surroundings and able to recognize the data in different
agencies use ML and DL for complex and diverse data forms of data. With the help of deep learning, the detection
processing and achieving multi-objective goals such as of genetic diseases such as Turner’s syndrome, hemophilia,
maximizing profit, minimizing production cost, forecasting and sickle cell anemia can be easily detected by the study of
sales, and improving operational and supply chain man- genes and the doctors are able to find the treatments for the
agement. ML and DL were also found to have an indis- future and future medication. The deep learning and ma-
pensable role in healthcare data management, and the chine learning g are able to find the normal and abnormal
classification of AL, ML, and DL is shown in Figure 1. data of the patients, by that the doctors can detect the
With the advancement in the computational system condition of patients.
emerge several computing technologies such as artificial Machine learning is also known as artificial intelligence,
intelligence, IoTs, and big data in various fields, especially in which is used for recognizing or diagnosing the real-time
healthcare system IoHTs that give the fruitful result to condition of the patients. The deep learning has the ability to
recognizing the daily life activity of diverse wearable devices. regenerate the real-time condition of the patients on the
The target is achieved by deep learning of HAR in IoHTs. We basis of raw data or from previous data of the patients. Due
know the types of deep learning such as supervised, unsu- to this ability, it can be used in various fields such as
pervised, and semisupervised. With the help of semi- healthcare system, for security of cybercrime industry.
supervised deep learning, the chances of accurate data However, it plays a vital role in healthcare system, especially
related to human activities analyse the considered data fairly for children and old age elders, and recently, elders are
to train the classifier data. For the effective accurate results of preferred to be alone enduring. By considering the facts, the
labelled sample, the intelligent auto-labelled scheme based deep learning is obligatory for the advancement.
on deep learning can improve the learning efficiency in In modern time, the fatal rate increases day by day with
IoMTs. cardiovascular disease (CD). The children and elders do not
Journal of Healthcare Engineering 3

• Inductive
• Instance based
Machine • Genetic
learning
• Neural network
• Bayesian approaches

• Supervised
■ Artificial Neural
■ Convolution neural

Deep learning ■ Recurrent neural

• Unsupervised
■ Self organizing map
■ Boltzman machine
■ Autoencoder

Figure 1: Classification and subclasses of AI, ML, and DL.

Disease
prediction

Smart
Drug
health
discovery
monitoring

AL ML DL
applications in
Epidemic Health care
Image
outbreak
analysis
detection

Medical Chronic
research disease
identification

Figure 2: Applications of ML and DL in the healthcare system.

prefer the wearable devices due to irritation, and in those devices and IoT and artificial intelligence. The SHM plays a
cases, deep learning is applied for gathering the data while vital role in the healthcare system in this modern era. Ar-
the wearable devices are having the ability to find or rec- tificial intelligence is applied for the uses of machine learning
ognize the real-time condition of patients. In this modern algorithms and software in healthcare systems. The results of
era, the acute diseases such as diabetes, Alzheimer’s diseases, artificial intelligence are based on the data used in machine
genetic diseases, cancer, and respiratory malady can be learning. In essence, this is discovered by the game. The
minimized by applying smartness in healthcare system that working of deep learning models is based on the convolu-
is possible using computational learning system, SHM, IoT, tional neural network (CNN) learning, because it is a type of
and knowledgeable doctors who have the capability to detect artificial neural network (ANN) that is applied for the vi-
the real-time condition of the patients and have the proper sualization of images such as ultrasound, MRI, CT scan, and
information or prediction of future condition. X-ray.
According to the medical survey, there are about 30% of The computational learning is applied in healthcare
the population suffering from cardiovascular diseases. We system when the working of smart health monitoring de-
can say that the computational learning or robotic learning is vices and IoTs is unable to detect the patient’s data. These
used for the smart health monitoring because smart health algorithms are able to recognize the behavioral data and
monitoring is the amalgamation of health monitoring form their own models; machine learning is also known as
4 Journal of Healthcare Engineering

artificial intelligence. It is the method in which the CNN and that help in the real-time monitoring of people’s health.
ANN are able to form an artificial neural system of an animal Traditional healthcare systems were unable to provide fast,
(AI), and it works in two forms: 1. algorithms are literal and cost-effective, and trustworthy health monitoring services
2. deep learning may be applied in the form of black boxes. from remote locations, but the SHM has made it possible.
Deep learning is working on the basis of linear methods Thus, the author presents a state-of-the-art review of some of
unlike nonlinear traditional method in which the individuals the major implementations of AI, ML, and DL in healthcare
undergo unwanted visits of the hospitals, but now they can data for data management, early detection, medication, data
be communicated with the doctors computationally. security, and detection. AI and machine learning are rev-
There are various types of devices that are used for olutionizing the healthcare market, as they are in virtually
different types of disease monitoring such as accelerometer everything else industry. From the possibility of deep
for the detection of breathing rate, cardiovascular defibril- learning and neural networks in pharmaceutical research to
lator, and pacemaker that are used for the diagnosis of the increasingly sophisticated use of intelligence diagnostic
breathing rate and blood pressure along with body tem- tests and surgical robotics, AI is changing the way we design
perature of the patients, which are the basic parts of IoTs. cars, optimize our energy usage, and maintain our finances.
Deep learning is also useful for the detection of daily life It is also bringing new opportunities—and risks—to human
activities, when the whole data are collected using micro- health management. The ability of AI to bring together huge
phone sensor arrays that are working on the basis of two- oceans of data that saturate every aspect of healthcare to
dimensional spectrum convolutional neural network improve decision-making between the sector’s key stake-
(DCNN). The daily life activities and behavior of the patients holders, from pharmaceutical innovators to healthcare
can be detected by many types of wearable and nonwearable professionals, binds it all together [17, 18].
devices and also by smart homes that are totally digitalized
by fixing the sensors such as toilet seats, smart pillow, smart 2. Review of Papers
bed, and smart floor. The behavioral activities are classified
into two categories of short-term and long-term activities; In recent times, the DL and ML have been extensively used
short-term activities can be changed from hours to days, in the smart health monitoring network and healthcare
such as repeated use of toilet that may be caused by stomach services. The DL and ML were used to analyse the health data
upset, and long-term activities can be changed from a few that are generated by various IoMT devices and sensors.
days to months, or years; in this, the repetition of all ac- These health data have been stored and handled by cloud
tivities in a day indicates the memory loss or Alzheimer’s network and cloud services. ML and DL are implemented to
disease. The deep learning is more beneficial for the de- analyse these complex health data to achieve multi-objective
tection and possible treatment of these diseases [16]. The tasks. According to Mohammad Ali et al., there are about 50
deep learning consists of two phases: 1. training and 2. billion devices that were added to the medical field by the
inferring training, which involves the labelling of large data end of 2020 and face many security challenges. According to
and identifies the feature. Inferring involves memorization him, the DL/ML is also used for gathering big data along
and makes the results on the basis of previous data. with the recognition of threats in cybercrime [19].
The primary goal of this study was to analyse this re- In the recent work of Imran Ahmed et al., he has dis-
lationship by focusing on two types of operational issues. cussed the use of data that are collected through the IP-based
Using real-world examples, here we demonstrate how RGB camera that is installed at the height of 4 meters and
machine learning can improve human abilities to com- takes the top view of the patients and the data recorded at the
prehend and manage healthcare operations, resulting in Institute of Management Science, Peshawar, and these data
more efficient health care. While medical computerization are collected for the detection of discomfort of the patients
and digitalization have made tremendous advances, by the movement of patient’s body. This camera helps in the
healthcare management systems have yet to profit greatly detection of the discomfort level of patients. The deep
from these breakthroughs due to the high complexity and learning algorithms (models) are used for identifying the
unpredictability of healthcare operations. The author has level of discomfort from the top. The Mask R-CNN 19 is
found that the systematic review of AI, ML, and DL roles in applied to the patients for recognition. These experiments
healthcare is lacking in the public domain. This motivates use various kinds of measures for the evaluation such as
the author to present the review on the recent integration of TPR, FPR, TNR, and MCR, and for the video records, the
AI, ML, and DL in various healthcare frameworks for en- FDR and TDR are used. The average performance for
hancing the quality of service and health of patients. Disease various organs of the patient body in the videos is sum-
outbreaks and emergence have become a big issue in today’s marized and presented in Table 1. It is noticeable that the
fast-paced world of technology and change. In terms of model was able to detect the discomfort with FPR of 7% and
disease prevention, control, and prevention, technology has MCR of 4%.
become a major issue for healthcare practitioners and In the healthcare system, we can say that the IoT and
healthcare business. Maintaining a healthy lifestyle has SHM play an important role in this field for enhancing the
become hard due to demanding job schedules. A smart real-life smart applications; i.e., they play a vital role in this
health monitoring system is the solution to the aforemen- time when COVID-19 is established as a pandemic by the
tioned issues. The recent industry 5.0 and 5G revolutions World Health Organization, and the detection of corona-
have resulted in the development of smart, low-cost sensors virus is difficult in early time of pandemic, but now it can be
Journal of Healthcare Engineering 5

Table 1: Performance parameters for all organs of the patient in video.

Average performance Head Right arm Left arm Right leg Left leg Torso Average
Accuracy 92 93 93 93 94 94 94
MCR 7 4 3 4 4 4 4
TPR 92 93 93 93 93 94 94
FPR 9 8 6 8 7 5 7
TNR 90 91 90 91 90 92 91

easily diagnosed using X-rays of chest. There are only small Shresth Tuli et al. discuss the health fog; it is a method of
images of X-rays, CT scan, and MRI, which are detected detection of heart diseases with the help of deep learning,
because COVID-19 is a new disease entered in this era, and and it collects the patient’s data using fog bus framework.
the less number of the techniques are used such as Inception The health fog models are based on IoT, which can manage
V3, Dark Convent, and Convent. In this pandemic era, many the data effectively and able to diagnose the seriousness of
researchers are at work on the treatment, diagnosis, and patients. The healthcare system is a vast project. The re-
methodology of COVID-19. searchers are focused on the edition of health fog in the
According to Hemant Ghayvat, the deep learning is also healthcare system for automatic diagnosis of acute heart
applied to the daily life activities with the help of various kinds diseases using deep learning and IoT devices. The fog is able
of devices. The recognition of daily life activities is more useful to manage the data that are collected from different IoT
for the elder ones who prefer to survive alone today. Almost devices. The predictions of heart patients with fog systems
about 30% of the total population are above 60 years old who are much accurate compared with the prior system that lacks
need the proper caretaker and disease-free life. The mortality the utilization of deep learning and IoT, and hence, the
rate of individuals by acute diseases of these ages can be cured results were inaccurate and insufficient for the records and
or treated by applying smartness in the healthcare system. the models that are based on deep learning are highly ac-
This is done by the use of smart health devices such as curate. They used the fog bus framework in their work for
smartphones and global poisoning system (GPS), and this is the detection of real-time conditions of heart patients. The
more suitable for recognizing outdoor activities. At the health fog system is now applicable for other health domains
present time, the spatial and other information is collected such as diabetes, cancer, hepatitis, genetic diseases, and
using the convolutional neural network (CNN). The acoustic- genetic disorders. Fog system is open-source software [14]. It
based monitoring system does not show excellent results due was found that the proposed mode training accuracy in-
to security issues. The identification of daily life activities creases and test accuracy decreases with increasing number
involves three stages, and at first stage, the data are collected of nodes. Increasing nodes increases the training data and
by the daily exercises, and spread on the bandwidth capacity is decreases the test data subset. The prediction accuracy in-
shifted by the channels that are inclined by the physical creased by 16% with the implementation of DL in the
exercises and by recorded data. With this method, the col- existing framework. The performance characteristics of the
lected data are right or accurate, and this procedure uses proposed model are presented in Figure 6, indicating higher
online methods for the recording [20]. Hemant’s model is accuracy, lower latency, and low power consumption.
found with 90% and 83% accuracy during testing and The wireless body sensor network (WBSN) [22] also has
training, whereas the loss was 0.5% and 1% for testing and a great role in healthcare system that helps in the detection of
training. The accuracy of various activities along with per- the real-time condition. The WBSN faces major challenges
formance graph is presented in Figures 3 and 4. such as feasible discussion in emergency cases, and these
The framework that is used for the smart health monitor is challenges are overcome using fractional cat-based scalp
the combination of compressed sensing (CS), machine swarm algorithms (FCSSA). It records the whole data using
learning-based models, ADMM, and the binary CNN clas- several types of parameters, and WBSN is used to transfer
sifier. Abrar Zahin works on the convolutional learning that it the data to the target node and then returns it back to the
works or records accurately and works with proficiency. In target node. The hybrid harmony search algorithm and
this study, the researcher uses the terms ADMM and CS, practical swarm optimization (hybrid HAS-PSO) are used
which are helpful in simplifying the software. The wireless for the determination that forms the results, and here, the
camera nodes are used for collecting large amounts of images. deep belief network (DBN) classifies the whole data for the
Then, these images are transferred to the main processing assessment. The FCSSA works in a better way with matrices
center, and here, any changes in health are detected [21]; the of 94.604 accuracy, 0.145 energy, and 0.058 throughput
author selected the 597 cases with 14 features such as age, sex, values. The performance parameters were compared with
chest pain type, resting blood pressure, cholesterol, number of existing models, namely low-energy adaptive clustering
smoking years, and blood sugar levels. The best accuracy of hierarchy (LEACH), particle swarm optimization (PSO),
the proposed model was found to be 84% and was higher than and artificial bee colony (ABC) for three data sets from
other regression naive Bayes and decision tree models. The Cleveland, Hungary, and Switzerland. It was found that the
model accuracy peaked at 98.5% when the proposed feature integration of DL in the proposed framework led to the
selection weighting methods were applied for the detection of highest accuracy of 94.6% along with the least energy
heart diseases as shown in Figure 5. consumption and highest throughput for all data sets when
6 Journal of Healthcare Engineering

Clapping
Crying baby
Vacuum cleaner
Washing machine
Toilet flush
Pouring water
Door open close

50 60 70 80 90 100
Accuracy (%)

Accuracy
Figure 3: Accuracy performance of the model during training and testing on data.

100 2.0
90 90
83
Performance parameter (%)

Performance parameter (%)

80
1.5
70
60
50 1 1.0
40
30 0.5 0.5
20
10
0 0.0
Training data Testing data

Accuracy
Loss
Figure 4: Loss performance of the model during training and testing on data.

110

100 98.5
Classifier accuracy (%)

92.2
90 89.3
87.3
84.4 83.4 84.9
82.4 83.5
81.1
80 80.4
77.6 77.6
75.4
73.7 73.7 74.8
71.8 72.2
70 69.8 70.3 70.7
66.5
63.2 63.2
60 60
56.3 56.3

50
SVM Logistic MLP Random Decision Naive Proposed
regression forest tree bayes model
Model

Before feature selection After feature selection

General feature weighting Specific feature weighting
Figure 5: Comparison of classifier accuracy of the proposed model with existing similar models for different number of features in the set.
Journal of Healthcare Engineering 7

100

90

Performance (%)
88.4
86.2
85 85.2

80

75.2
72
70
Training Data Testing Data
Data Type

Best
Average
Ensemble
Figure 6: Training and test accuracy with different number of nodes.

compared to the competing models. The summarized results and this is a gradual process but does not take much time
are presented in Figure 7. more than a year; hence, it is difficult to identify the pro-
Heart disease and cardiovascular diseases (CD) increase gression, and the way of conversion to AD from stable MCI
the mortality rate, and the cases of heart attacks, heart stroke, can be identified with the help of deep learning. The ML and
and brain stroke are more compared with others, which are DL play an important role in the diagnosis of AD. The
prevented by the use of deep learning modified neural patient’s data are heterogeneous in nature, which involves
network (DLMNN) because it can detect the heart diseases data from magnetic resonance imaging (MRI), positron
at initial stages and helps in future medication. The DLMNN emission tomography (PIT), X-ray, CT scan, etc. Alz-
is implemented by the three steps including (1) authenti- heimer’s disease models are based on the CNN and single
cation, (2) encryption, and (3) classification. In the first step, MRI images, but these data are not accurate and do not give
the substitution cipher and SHA 12 are collectively used, proper information about the real-time condition of the
which are helpful for the authentication of heart patients patients, and hence, the advanced deep learning is applied
from different hospitals. The wearable devices that fixed into and multimodal data are used. These models are based on
the patient’s body by the patient collect the actual data and the five time-series modalities and CNN-BiLSTM design,
transfer it to the cloud. Now, the censored data are securely which can give great results that help improving the
transmitted using the PDH-AES technique. Here, the monitoring of Alzheimer’s diseases [26].
encrypted data are now decrypted. After this, the whole data The smart healthcare system is the solution for widely
undergo classification by the DLMNN classifier. The clas- spread acute diseases towards the old age population. SHM
sified final data may be normal or abnormal and show the monitors noncommunicable lethal diseases in elders by
condition of the patients [23]. The PDH-AES is 95.8% able to observing daily life activities. There are many more chal-
secure the whole data, so the security cannot be found. lenges in front of this smartness in healthcare systems such
According to a recent study, the cases of Alzheimer’s as remote monitoring environment, communication tech-
disease (AD) are increasing day by day, and there may be up nologies of this environment, and presence of intelligence
to 60% to 70% cases that can be seen in the age of 20 to 30 processing system. There are many smart sensors and
years. Alzheimer’s disease is a neurologic disorder that af- smartphones used for human activity recognition (HAR)
fects the brain to shrink (loss of function) and brain cells to algorithms. The sensors are fixed in smart devices and
die. The most common kind of dementia is Alzheimer’s wearable devices such as smartphones, Fitbit watches, and
disease, which is characterized as a gradual loss of cognitive, hand bands that are able to recognize every activity of an
behavioral, and social abilities that limit a person’s ability to individual such as walking, running upstairs by measuring
function independently. One of the earliest signs of Alz- blood pressure, respiratory rate, and fitness tracking. There
heimer’s disease is forgetting recent events or talks [24, 25]. are some drawbacks observed in these devices, i.e., the
As Alzheimer’s disease progresses, a person’s memory will inaccuracy of data acquisition [R]; it is needed for high
deteriorate and he or she will lose the ability to do daily tasks. computational system. The accuracy and classification effi-
The cases with mild cognitive impairment (MCI) disease are ciency can be improved by the addition of group-based
less than AD, a highly heterogeneous phenotypic spectrum. context of human activity recognition (GCHAR). The
There are almost about 10 to 20% of MCI growths to AD, GCHAR is structured in two-level hierarchical groups such
8 Journal of Healthcare Engineering

100.0 0.50 0.25

0.232
0.45
97.5
0.40 0.20
0.38367

Performance parameter
0.35
95.0 94.4 94.6
94.2
0.30 0.15
92.5 0.25
90.72333 0.20 0.10
90.0
0.15
0.131
0.05267 0.05767
0.04567 0.10 0.05
87.5
0.05
0.033
85.0 0.01067 0.00 0.00
Leach PSO ABC Proposed HSA
PSO FCSSA
Model

Accuracy (%)
Energy (mJ)
Throughput

Figure 7: Comparison of performance parameters of the proposed model with the existing model for different sizes of data set.

as intergroup and inner group. The accuracy of HAR is that works as outbreaks from other algorithms. It has been
improved by the GCHAR, and it is proved by the experi- found from Figure 10 that the integration of deep learning
mental comparison with the other algorithms as shown in methods in the existing frameworks leads to improvement in
Figure 8. accuracy F1 and area under curve parameters in identifying
It is worthwhile to note that GCHAR algorithms out- the diabetic patients [29].
perform the other algorithms by a good margin as shown in
Figure 8. The higher accuracy of 94.16% is achieved by this
2.1. Disadvantages and Future Challenges. The deep learning
algorithm. The accuracy of the algorithm is based on two
and machine learning are the main reason for the advance-
aspects such as the hierarchical classifier structure able to
ment in the medical field, but there are some limitations that
reduce the misclassification and inaccuracy. Secondly, the
can be seen like it needs the huge data for recognition and for
error correction utilization of context awareness improves
making results, which are impossible sometimes. With the
the correction measurement and whole performance [27].
bulky data, it may be complex and forms complex data
The routine HAR can be improved using various types of
models, and hence, these computational techniques are more
algorithms and compared among these algorithms to rec-
expensive. Therefore, this is unable to apply for everyone, and
ognize the betterment. For the experiment, the researcher
hence, they cannot take the proper benefits of this technique.
selects the minimum eight human activities denoted by As
Deep learning: the deep learning is more prone to error
A1, A2, A3, and on and along with classifying C1, C2, C3,...., so
chances of errors that are occurred in training and testing of
on. For the demonstration of all activities, we can use the
data due to bulk of data. Sometimes data become inconsistent,
receiver operating characteristic (ROC) curves, which are
and this is due to continue updating.
shown in Figure 9. The eight solid lines show the different
No doubt the robotic learning or ML/DL plays a more
human activities (HAs) in which the average value of the
important role in every field, but there are some challenges
area under the curve (AUC) is 0.95. According to the figure,
that are faced by it. First of all, this type of learning lacks
the A1 activity obtains the best performance and the A5
interpretability, requires bulky data, is sometimes unable to
shows the worst performance. For the comparison among
handle the direct symbols and decision-making support, and
the performances of these activities, we can use some other
has difficulty in dealing with natural languages. The deep
kinds of algorithms such as DNN, SVM, and RF for the
learning methods require a large amount of data; hence, the
improvement as shown in the next figure. In this figure,
handling of the data is much difficult and consumes more
when the AUC value is 0.95, all three algorithms can only
time and storage. The study of deep learning methods can
achieve the values of 0.87, 0.74, and 0.90, respectively. The
only understand by those doctors or medical individuals
reason for these values and position of activities in the figure
who is having the proper knowledge of linguistic grammar
is that the on-body positions form exclusive data for the
and patient’s health data.
accuracy than the other positions [28].
Deep learning and machine learning play an important
role in diabetic patients for the prevention of readmission in 2.2. Security Issues in ML/DL. IoT devices used a lightweight
hospitals. According to researchers, the diabetic patients are framework, thus prone to denial of service (DoS), jamming,
widely spread all over world. The number of people with spoofing, man-in-in-the-the-middle software attacks, and
diabetes increases from 1980, i.e., 180 million to 422 million privacy leakage. Sybil and spoofing attacks are also common.
in 2014. Type 2 diabetes is more spread in underdeveloped Many ML/DL-based frameworks are available in the public
countries around 17.1% in 2004, and it increases 30% in the domain for security. ML/DL-based authentication, access
last ten years. This can be done by the combination of control, secure offloading, and malware detection models are
convolutional neural network (CNN) and data engineering available but lack speed and accuracy at the early learning
Journal of Healthcare Engineering 9

GCHAR
94.1
Decision Table
65.2
KNN
90.8

Model
Bayes Net
82.9
J48
83.5
Bagging
89.5
Random Tree
80.1
50 60 70 80 90 100
Accuracy (%)

Figure 8: Comparative evaluation of classification accuracy among GCHAR and related classification algorithms.

1.0
0.95 1.1
1
0.9 0.9 0.97 0.97 0.98
0.87 1.0
0.91 0.92
RoC curve area (Area)
0.88
RoC curve area

0.8 0.9

0.74
0.8
0.7
0.7
0.6
0.6

0.5 0.5
DNN SVM RF Proposed model Class 1 Class 2 Class 3 Class 4 Class 5 Class 6 Class 7
Model Class Segment

RoC curve area

(a) (b)

Figure 9: (a) Comparative result of the activated various algorithms and (b) class-wise performance of the proposed model.

100 95
92 91
84
79
Relative performance (%)

80 75
72

60

42
40
34

20

0
Accuracy F1 AUC
Performance parameter

Other
Ensemble
Deep Learning
Figure 10: Comparative results of performance parameters before and after implementation of deep learning algorithm in the existing
framework.
10 Journal of Healthcare Engineering

stage, high computation, and communication cost along Ingénierie des Systèmes d’Information, vol. 24, no. 3,
with the risk of unsupervised learning. Different types of IoT pp. 343–351, 2019.
have many organizations that form the security standards, [4] A. Dogra, A. Kaur, and M. Shabaz, “Data collection for
and the organizations are IEEE and ETSI, because they are classification in IOT and heart disease detection,” Ann. Rom.
able to form security norms. A big IoT system needs the Soc, vol. 25, no. 4, pp. 2954–2964, 2021.
higher and robust security such as blockchain technology [5] S. Saralch, V. Jagota, D. Pathak, and V. Singh, “Response
surface methodology-based analysis of the impact of nanoclay
[30, 31]. The ML/DL of security of IoTs is categorized into
addition on the wear resistance of polypropylene,” The Eu-
IoT system, IoT threats, and learning method for IoT se-
ropean Physical Journal - Applied Physics, vol. 86, pp. 1–13,
curity/DL security. On the basis of IoT security learning, ML Article ID 10401, 2019.
approaches include supervised approaches, unsupervised [6] Y. Wang, S. Nazir, and M. Shafiq, “An overview on analyzing
approaches, semi-supervised approaches, and RL methods deep learning and transfer learning approaches for health
[32]. The DL learning is also differentiated into supervised, monitoring,” Comput. Math. Methods Med, vol. 2021, 2021.
unsupervised, and semi-supervised. [7] K. Haricha, A. Khiat, Y. Issaoui, A. Bahnasse, and H. Ouajji,
“Towards smart manufacturing: implementation and bene-
3. Conclusion and Future Scope fits,” Procedia Computer Science, vol. 177, pp. 639–644, 2020.
[8] P. Singh, M. A. Dulebenets, J. Pasha, E. D. R. S. Gonzalez,
The present work shows the integration of ML and DL in Y. Y. Lau, and R. Kampmann, “Deployment of s and existing
healthcare system for various applications, namely disease- challenges,” IEEE Access, vol. 9, Article ID 91461, 2021.
discomfort detection, diagnosing cancer, diabetes, tumor, [9] K. Niranjan, K. S. Narayana, and M. V. A. L. N. Rao, “Role of
respiratory malady such as acute diseases, and prevention of artificial intelligence in logistics and supply chain,” in Pro-
outbreak of pandemic or endemic diseases in a particular ceedings of the 2021 Int. Conf. Comput. Commun. Informatics,
region city or state. The recent developments in the ML/DL ICCCI, vol. 2021, pp. 27–29, Coimbatore, India, Jan 2021.
models and framework have shown remarkable perfor- [10] S. Vyas and D. Bhargava, “Big data analytics and cognitive
mance in the IoMT network and services. ML/DL also computing in smart health systems,” in Smart Health Systems,
showed the proven record in clinical decision-making ca- pp. 87–100, Springer, Singapore, 2021.
[11] X. Huang, V. Jagota, E. Espinoza-Muñoz, and J. Flores-
pabilities under supervised conditions. The use of ML and
Albornoz, “Tourist hot spots prediction model based on
DL is essential to get useful information from complex and optimized neural network algorithm,” International Journal of
considerable health data generated by the healthcare systems System Assurance Engineering and Management, 2021.
in terms of electronic health record, medical images, ECG, [12] M. Azimi, A. D. Eslamlou, and G. Pekcan, “Data-driven
EEG records, and data obtained from real-time health structural health monitoring and damage detection through
monitoring wearable devices and sensors. The present work deep learning: state-other- art review,” vol. 20, no. 10, 2020.
will help to generate ideas for future implementation of ML [13] Z. Xu and J. Sun, “Model-driven deep-learning,” National
and DL in other aspects of the healthcare system. However, Science Review, vol. 5, no. 1, pp. 22–24, 2018.
the present work in not including much about the role of ML [14] S. Tuli, N. Basumatary, S. S. Gill et al., “HealthFog: an en-
and DL in the security system of healthcare frameworks is a semble deep learning based smart healthcare system for au-
vast and complex domain. The author considers the review tomatic diagnosis of heart diseases in integrated IoT and fog
of such work as a future plan. computing environments,” Future Generation Computer
Systems, vol. 104, pp. 187–200, 2020.
Data Availability [15] A. M. Shah, X. Yan, S. A. A. Shah, and G. Mamirkulova,
“Mining patient opinion to evaluate the service quality in
The data pertaining to this article are included within the healthcare: a deep-learning approach,” Journal of Ambient
article. Intelligence and Humanized Computing, vol. 11, no. 7,
pp. 2925–2942, 2020.
[16] H. Ghayvat, S. Pandya, and A. Patel, “Deep learning model for
Conflicts of Interest acoustics signal based preventive healthcare monitoring and
The authors declare that there are no conflicts of interest activity of daily living,” in Proceedings of the 2nd Int. Conf.
regarding the publication of this study. Data, Eng. Appl. IDEA, vol. 2020, Bhopal, India, Feburary
2020.
[17] A. H. Jabbar, S. O. Mezan, M. Salih, and M. A. Agam,
References “Assessment of anticholinesterase effect of poly-
[1] F. Alsubaei, A. Abuhussein, V. Shandilya, and S. Shiva, vinylpyrrolidone/silver nanocomposite biosynthesized by
“IoMT-SAF: Internet of medical things security assessment Pandanus atrocarpus extract,” Materials Today Proceedings,
framework,” Internet of Things, vol. 8, Article ID 100123, 2019. vol. 42, pp. 2578–2583, 2021.
[2] S. F. Suhel, V. K. Shukla, S. Vyas, and V. P. Mishra, “Con- [18] S. Saxena, S. Vyas, B. S. Kumar, and S. Gupta, “Survey on
versation to automation in banking through chatbot using online electronic payments security,” in Proceedings of the
artificial machine intelligence language,” in Proceedings of the 2019 Amity International Conference on Artificial Intelligence
8th International Conference on Reliability, Infocom Tech- (AICAI), pp. 756–751, IEEE, Dubai, United Arab Emirates,
nologies and Optimization (Trends and Future Directions)(I- 2019 February.
CRITO), pp. 611–618, IEEE, Noida, India, June 2020. [19] M. A. Al-Garadi, A. Mohamed, A. K. Al-Ali, X. Du, I. Ali, and
[3] G. Verma and R. Chakraborty, “A hybrid privacy preserving M. Guizani, “A survey of machine and deep learning methods
scheme using finger print detection in cloud environment,” for Internet of things (IoT) security,” IEEE Communications
Journal of Healthcare Engineering 11

Surveys & Tutorials, vol. 22, no. 3, pp. 1646–1685, Article ID

2988293, 2020.
[20] I. Ahmed, G. Jeon, and F. Piccialli, “A deep-learning-based
smart healthcare system for patient’s discomfort detection at
the edge of Internet of things,” IEEE Internet of Things Journal,
vol. 8, no. 13, Article ID 10326, 2021.
[21] F. Ali and K. S. Kwak, “A smart healthcare monitoring system
for heart disease prediction based on ensemble deep learning
and feature fusion,” Information Fusion, vol. 63, pp. 208–222,
2019.
[22] A. Alameen and A. Gupta, “Optimization driven deep
learning approach for health monitoring and risk assessment
in wireless body sensor networks,” International Journal of
Business Data Communications and Networking, vol. 16, no. 1,
pp. 70–93, 2020.
[23] S. S. Sarmah, “An efficient IoT-based patient monitoring and
heart disease prediction system using deep learning modified
neural network,” IEEE Access, vol. 8, pp. 135784–135797,
2020.
[24] G. Rastogi and R. Sushil, “Cloud computing implementation:
key issues and solutions,” in Proceedings of the 2015 2nd
International Conference on Computing for Sustainable Global
Development (INDIACom), pp. 320–324, IEEE, New Delhi,
2015, March.
[25] M. K. Loganathan, S. S. Neog, and S. Rai, “Process safety and
performance improvement in oil refineries through active
redundancy and risk assessment method-A case study,” in
Proceedings of the 2018 IEEE International Conference on
Industrial Engineering and Engineering Management (IEEM),
pp. 98–102, IEEE, Bangkok, Thailand, 2018, December.
[26] S. El-Sappagh, T. Abuhmed, S. M. Riazul Islam, and
K. S. Kwak, “Multimodal multitask deep learning model for
Alzheimer’s disease progression detection based on time
series data,” Neurocomputing, vol. 412, pp. 197–215, 2020.
[27] L. Cao, Y. Wang, B. Zhang, Q. Jin, and A. V Vasilakos,
“GCHAR: an efficient group-based context – aware human
activity recognition on smartphone,” Journal of Parallel and
Distributed Computing, vol. 118, pp. 67–80, 2018.
[28] X. Zhou, W. Liang, K. I.-K. Wang, H. Wang, L. T. Yang, and
Q. Jin, “Deep-learning-enhanced human activity recognition
for Internet of healthcare things,” IEEE Internet of Things
Journal, vol. 7, no. 7, pp. 6429–6438, 2020.
[29] A. Hammoudeh, G. Al-Naymat, I. Ghannam, and N. Obied,
“Predicting hospital readmission among diabetics using deep
learning,” Procedia Computer Science, vol. 141, pp. 484–489,
2018.
[30] M. N. Kumar, V. Jagota, and M. Shabaz, “Retrospection of the
optimization model for designing the power train of a formula
student race car,” Scientific Programming, vol. 2021, Article ID
9465702, 9 pages, 2021.
[31] J. Bhola, S. Soni, and G. K. Cheema, “Genetic algorithm based
optimized leach protocol for energy efficient wireless sensor
networks,” Journal of Ambient Intelligence and Humanized
Computing, vol. 11, no. 3, pp. 1281–1288, 2020.
[32] J. Bhola, M. Shabaz, G. Dhiman, S. Vimal, P. Subbulakshmi,
and S. K. Soni, “Performance evaluation of multilayer clus-
tering network using distributed energy efficient clustering
with enhanced threshold protocol,” Wirel. Pers. Commun.,
2021.