Convolutional Neural Network-based Covid-19 Analysis with Internet of Things

Abstract

Abstract— The very hazardous respiratory illness known as
COVID-2 (SARS-CoV-2), which is the root cause of the even
more serious illness known as COVID-19, was caused by the
COVID-2 virus. The COVID-19 virus was identified in Wuhan
City, China, in the month of December in 2019. It began in
China and then spread to other parts of the world before it was
officially classified as a pandemic. It has had a significant impact
on day-to-day life, the welfare of people in general, and the
economy of the whole globe. It is of the utmost importance,
particularly in the beginning stages of treatment, to pinpoint the
constructive experiences that are useful at the proper time. The
identification of this virus involves a substantial number of tests,
each of which takes a certain amount of time; nevertheless, there
are currently no other automated tool kits that can be used in
their place. X-ray photos of the chest that are obtained via the
use of radiology imaging methods may provide significant
insight into the COVID-19 infection if they are analysed
carefully. An accurate diagnosis of the infection may be
obtained via the application of deep learning techniques, which
are applied to radiological images and make use of cutting-edge
technology such as artificial intelligence. Patients who reside in
distant places, where it may not be feasible for them to have
rapid access to medical facilities, may benefit from this kind of
analysis throughout the course of their therapy. One of the deep
learning strategies that are used in the creation of the model that
has been proposed is the use of convolutional neural networks.
The images of chest X-rays are analysed by these networks to
detect whether a person has a positive or negative result for the
Covid gene.
Keywords— Real-time traffic, Machine Learning, YOLOv4,
Transfer Learning, Easy-OCR

Full text

2022 Second International Conference on Advanced Technologies in Intelligent Control, Environment, Computing and Communication Engineering
ICATIECE
978-1-6654-9396-3/22/$31.00©2022 IEEE
Convolutional Neural Network-based Covid-19
Analysis with Internet of Things
T. Saravanan,
Department of CSE,
GITAM School of
Technology,
GITAM (Deemed to be
university),
Bengaluru, India.
tsaravcse@gmail.com
D. Ramalingam
Dept. of CSE,
R. M. K. Engineering
College,
Chennai, India
drm.cse@rmkec.ac.in
K. Keerthika,
Department of CSE,
Karpagam College of
Engineering,
Coimbatore, India
kikeerthika@gmail.com
T.Sathish
Department of CSE
Malla Reddy of Engineering
And Technology
Secunderabad,India
sathish46nkl@gmail.com
Abstract— The very hazardous respiratory illness known as
COVID-2 (SARS-CoV-2), which is the root cause of the even
more serious illness known as COVID-19, was caused by the
COVID-2 virus. The COVID-19 virus was identified in Wuhan
City, China, in the month of December in 2019. It began in
China and then spread to other parts of the world before it was
officially classified as a pandemic. It has had a significant
impact on day-to-day life, the welfare of people in general, and
the economy of the whole globe. It is of the utmost importance,
particularly in the beginning stages of treatment, to pinpoint
the constructive experiences that are useful at the proper time.
The identification of this virus involves a substantial number of
tests, each of which takes a certain amount of time;
nevertheless, there are currently no other automated tool kits
that can be used in their place. X-ray photos of the chest that
are obtained via the use of radiology imaging methods may
provide significant insight into the COVID-19 infection if they
are analysed carefully. An accurate diagnosis of the infection
may be obtained via the application of deep learning
techniques, which are applied to radiological images and make
use of cutting-edge technology such as artificial intelligence.
Patients who reside in distant places, where it may not be
feasible for them to have rapid access to medical facilities, may
benefit from this kind of analysis throughout the course of
their therapy. One of the deep learning strategies that are used
in the creation of the model that has been proposed is the use of
convolutional neural networks. The images of chest X-rays are
analysed by these networks to detect whether a person has a
positive or negative result for the Covid gene.
Keywords— Real-time traffic, Machine Learning, YOLOv4,
Transfer Learning, Easy-OCR
I. INTRODUCTION
Coronavirus-2 is the causative agent of the COVID
infection, which is responsible for severe acute respiratory
syndrome (SARS-CoV-2). The Chinese authorities issued
the first notification of the coronavirus in Wuhan in
December of 2019. Since then, this virus has spread over the
world, affecting up to 25 million people. The coronavirus has
caused injury to a sizeable number of people all over the
globe and is rapidly increasing in its prevalence. There were
about 28 million cases of COVID as of the month of August
in the year 2020, which resulted in the deaths of 875,000
individuals all across the globe [1]. Because there is now no
treatment that is shown to be successful against the
coronavirus, the most effective method for preventing
infection with this virus is to be tested for it and to stay
isolated if the test results come back positive for the
condition. Isolation is the sole method that can be used to
stop the spread of the disease and prevent its further spread.
Individuals who have a compromised immune [2] system,
people who are elderly, and persons who have a past medical
history that includes a pulmonary infection are at a larger
risk of becoming infected with COVID-19. This risk also
increases for people who have a history of COVID-19
infection in their family.
Fig. 1. Overall virus spread
The primary symptoms of the COVID-19 virus include a
high fever, a cold, difficulty breathing, and a variety of other
symptoms. The virus may be diagnosed by these symptoms.
We should wash our hands often, make an effort to avoid
touching our nose, eyes, mouth, and face, and keep a space
of at least two feet between ourselves and other people in
order to protect ourselves from the COVID-19 virus.
Because there is currently no antibody that can be used to
treat infections caused by COVID-19, [3] the number of
people who have been infected with the virus is continually
rising. Because the number of positive cases has been
growing at an alarmingly quick rate, the tests that detect the
presence of Coronavirus need to review the data in a
considerably shorter length of time. It has been demonstrated
that increasing testing for any symptoms found and
immediately isolating infected individuals is the most
effective way to control the rapid spread of the disease. In
light of the fact that the World Health Organization (WHO)
has declared COVID-19 [4] a pandemic and in light of the
fact that the number of people getting infected is rising at an
2022 Second International Conference on Advanced Technologies in Intelligent Control, Environment, Computing & Communication Engineering (ICATIECE) | 978-1-6654-9396-3/22/$31.00 ©2022 IEEE | DOI: 10.1109/ICATIECE56365.2022.10047327
Authorized licensed use limited to: GITAM University. Downloaded on March 13,2023 at 12:39:34 UTC from IEEE Xplore. Restrictions apply.

astounding rate, it has been demonstrated that this is the most
effective way to control the rapid spread of the disease. Both
computed tomography (CT) sweeps and chest tomography
imaging, which are more often referred to as X-rays, are two
procedures that are well-known for diagnosing COVID and
that are utilised extensively in contemporary medicine [5]. It
requires a complete report on CT and ten to twelve days to
identify whether or not a patient's respiratory organs or lungs
have been infected by coronavirus. This determination is
made after the patient has undergone CT. It is possible that
the patient would be adversely impacted by the disease as a
consequence of the damage sustained by the respiratory
organ. Because the RT-PCR tests require more effort,
clinical specialists believe that early and prudent disclosure
from clinical preliminary X-rays can assist in determining
whether or not the patient should be maintained in detention
until the examination office test results arrive. This can help
determine whether or not the patient should be kept in
detention until the results of the examination office test
arrive. This information may be helpful in deciding whether
or not the patient should be held in detention until the results
of the examination office test are received.
X-rays allow for a quick diagnosis, which puts a halt to
the fast spread of the illness from one person to another. An
isolating variable is the refinement of X-ray pictures of the
chest; if the image of the chest X-ray is normal, patients
could get back thus hold it together for exploratory office test
results. If the image of the chest X-ray is abnormal, patients
could not get back. Patients may not be able to get back on
their feet if the picture on the chest X-ray is aberrant; thus,
they should keep it together. At that time, the importance of
our work had already been attained in the particular piece of
literature that we were looking at. The use of artificial
intelligence (AI) [6] to the modified and distinctive
verification of diseases has been the subject of significant
research efforts, and as a consequence, the discipline has
evolved in both quantity and quality. Significant learning
techniques are utilised widely in clinical challenges such as
the identification of cancer, the depiction of carcinoma, and
the detection of respiratory concerns from photographs of
chest X-rays. Examples of these types of difficulties include.
It is difficult to staff all crisis centres with medical personnel
due to the limited number of accessible radiologists. This is
because there is a shortage of radiologists currently working
in the field. Consequently, making use of the right AI models
in order to manage the non-error prone outputs and verify
that the diseases are necessary to beath is a form of
pandemic. This is accomplished by using Convolutional
Neural Network (CNN) connections everywhere for the
purpose of normalisation modifications, and then utilising
that data in response to a request. A critical learning model
that takes use of deep learning is presented in this study for
the goal of bringing about change as well as offering a fast
and clear demonstration of COVID19 [7]. This research was
carried out in order to: This significant educational strategy,
which has been proposed, takes use of significant chest X-
rays in order to present undeniable proof of COVID-19.
While the model is still at the ready stage, there have been
3824 chest X-ray scans conducted on it.
II. RELATED WORK
A part of the work associated to the COVID-19 chest
infection check, the pneumonia check, and other chest-
related contaminations may be found in this area. For the
purpose of using robots to disclose the information, the
author [1] gathered a dataset of patient chest X-ray pictures
from patients suffering from a variety of respiratory illnesses,
such as COVID-19, as well as various defilements from
public shops. This was done with the intention of using
robots to reveal the information [3].
They used CNN and trade learning to search for any
anomalies in clinical chest X-ray picture datasets, and the
results they obtained were uncommon outcomes that
occurred around 96 percent of the time. In addition to this,
they reasoned that by applying deep learning to the image of
the chest X-ray, they would be able to efficiently monitor
crucial biomarkers that are without a doubt connected with
Coronavirus.
The authors Narin An et al. devised a modified
acknowledgment structure as one of the fast amazing
diagnostic choices to determine the distribution of COVID-
19 among individuals [8]. This building was completed as
one of the speedy and outstanding diagnostic choices
available. During the course of the research that they were
doing, they made use of models that had been handled by
CNN before. These models included ResNet50, InceptionV3,
and Inception-ResNetV2 in that order. They investigated the
patients using X-light emissions in order to identify between
the various COVID abnormalities that might be present in
the respiratory system. When compared to the other two
models, Inception-ResNetV2 and InceptionV3, the pre-
arranged model of ResNet50 provided the best degree of
accuracy in terms of data collecting. This was because the
pre-arranged model had the fewest number of variables. The
author of research [9] made use of the independent game
plan computation so that he could be specific and secure C-
infers for pneumonia ID in contamination in X-ray photos of
the chest. When compared to the other existing approaches,
such as WFT, DWT, and WPT, the C-infers strategy, which
is both independent and secure, generated more favourable
results for noteworthy evidence than any of the others.
Following that, the authors of the article [6] presented a
way of utilising pictures treating systems for the area of the
presence of respiratory issue fogs in X-radiates images of
chest [6]. After that, they proposed calculating a percentage
of the region of healthy respiratory area and adding in the
position on the respiratory organ in order to detect the
existence of pneumonia. This was done in order to determine
whether or not the patient had pneumonia. A Convolutional
Neural Network (CNN) was used by the author of research
[10] in order to diagnose pneumonia from a chest X-ray
image, which resulted in an order precision of 83%.
III. PROPOSED METHOD
Because of the vast amount of strategical learning it has,
CNN is one application of the deep learning technology that
is now in use [11]. The GoogleNet model was used to
facilitate deep learning, which was successfully completed
[12]. This informative collection is made up of open
samples and has a total of 3824 chest X-ray images of both
positive and normal persons. There are a total of 1912 chest
X-ray photographs for each patient type in this collection.
The model chest X-ray scans of patients who have positive
results and normal findings, respectively, are shown in Fig
1(a) and (b), which may be seen below. Deep learning
approaches make use of artificial neural networks that are
built in a hierarchical fashion. Hierarchical structures are
used to develop artificial neural networks. The generation of
Authorized licensed use limited to: GITAM University. Downloaded on March 13,2023 at 12:39:34 UTC from IEEE Xplore. Restrictions apply.

each of the layers is accomplished with the use of
perceptron hubs [13]. The key point that determines
commitments based on the data by using certain coefficients
or burdens that amplify or hose wellsprings of data,
appropriately strengthened commitments to deliver the
central point.
Fig.2(a) Normal chest x-ray images
Fig. 2(b)COVID affected chest x-ray images
The expansion of the organisation and the location of
its core inside the association are both open to
adjustment so long as they continue to meet the criteria
of the task. Convolutional neural networks and phoney
neural association are two examples of approaches that
are lowering the bar for the quality of convolution. A
data layer is a part of this, and after that comes a series of
mystery levels and convolution layers in addition to the
ones that came before them. This collection of
convolution layers will convolve the data structure that
has been supplied by using a piece or channel as part of
their process. Covering up the channel is one way that
the movement of convolution may be represented on the
information layer. The delayed impact of the convolution,
which is subsequently dealt with through the capacity of
initiating the nonlinear change [14]. As a result, in order
to minimise its size, it is processed by two further
convolution layers, which are also known as pooling
layers. This procedure is repeated anytime it is essential
to completely process linked layers as well as
normalising layers. It also occurs whenever it is
necessary to process layers in their normalised state.
Because they are sandwiched between the data layer and
the yield layer, all of the layers are regarded as being
hidden for the simple reason that the systems that are
situated on the periphery of the structure are unable to
see them.
GOOGLENET
CNN's concept for GoogleNet, which is being implemented
for the purpose of displaying pictures, is called GoogleNet.
In addition to that, the inceptionV1 version is another name
for it. In 2014, Google's research bundle was responsible for
developing the first idea for GoogleNet. This exemplifies the
value that may be gained by using the Deep Convolution
relationships. In 2014, Google announced a visual
confirmation challenge that widened the reach of the massive
GoogleNet model. This challenge was successful. It had a
cruising speed of 6.69% overall, which ranked in the top five
overall. In comparison to AlexNet in 2013, ZF-Net in 2013,
and VGG in the year 2014, its error rate is reduced, and it
looks differently in relation to each of those networks. For
the purpose of this experiment, the GoogleNet was selected
since its performance was equivalent to that of people. In our
model, there are a total of 22 layers of significant CNN, but
there are 12 times less constraints. This is due to the fact that
our model is more flexible. There is no clear parallel
between the AlexNet plan rule in GoogleNet and the ZF-Net
links that exist in GoogleNet. It makes use of a variety of
methods, including standard pooling layers in addition to 1D
convolution, which ultimately results in a more
comprehensive association plan. A handful of the levels
included within this architecture make use of convolutions
with a ratio of one to one in the scale of the convolutions. In
the first layer, we make use of convolutions on a scale of one
by one throughout the course of an extended time. It brings
the overall number of constraints, including loads and
predispositions, that each layer is subject to down to a more
manageable level. It is certainly conceivable to relax the
rules, which will lead to an increase in the significance of the
connection that has been established. After that, it swaps a
rather high number of channels for a perceptron layer that is
less invasive and makes use of convolution. This layer is
added after the previous one [15].
Fig. 3 The
Architecture of GoogleNet
Themodelshowedupunderneathtoperform5×5convolution
with 48 channels, we display in Fig. 3(a) and
3(b)thequalificationwithboth1x1convolution.
Not as appealing as using 1x1 convolution:
Authorized licensed use limited to: GITAM University. Downloaded on March 13,2023 at 12:39:34 UTC from IEEE Xplore. Restrictions apply.

Fig. 4(a) Without 1x1 Convolution network
The total number of operations in this example of
Convolution without 1x1 is (14 1448) (5 5480) =112.9 M.
With 1x1 convolutions:
Fig. 4(b)
With 1x1 Convolution network
Inthiscasein1x1convolutionstheabsolute The activity
Figs are calculated as follows: 14 × 14 x 16 multiplied by 1
x 1 x 480, in addition to 14 x 14 x 48 multiplied by 5 x 5 x
16, for a total of around 5.4 million. It has been found that
the element that brings about a change is a very huge one,
such as the 107.7M initiating module. [Citation needed]
The capability of making use of an origin layer in order to
cover a greater region while achieving a predefined goal
utilising a constrained quantity of data on the picture. This
may be done by employing a combination of the two
phrases: "cover" and "achieve." Therefore, the modules
that were discussed earlier are used in the process of
developing the capacity of conducting computations in
large convolutional associations. As was discussed in the
introduction, these relationships might also be improved by
receiving aid in lowering the dimensionality of the data by
the use of stacked 1x1 convolution. This information was
presented before.
Fig 2 illustrates how the first module of Credulous works.
This module conducts a convolution of a contribution
using three distinct estimated channels (5x5, 3x3, and 1x1)
and a maximum pooling layer, in which all of the yields
are connected. This issue of the computational cost as well
as the challenges caused by overfitting may now be
considered addressed thanks to the module that is being
provided. The Fig depicts a modified and unusually
designed starting layer in an effort to cut down on the
computational cost of more essential associations. This was
done in order to achieve this goal. 3(a) An extra 1x1
convolution layer could be inserted before the 3x3 and 5x5
convolution levels, as well as the max pool layer, in order
to limit the amount of data that can be sent over the
channel. This is done in order to control how much
information can be communicated. Including more
convolution action is a foolish concept, regardless of how it
has shown itself in the past. It not only helps in reducing
the overall amount of constraints, but it also makes the
computation less expensive. In this result of similar jobs,
convolutions with a size of 5x5, 3x3, and 1x1 paired with a
maximum pooling of 3x3 are related with greater yields.
The use of variable-size convolution channels makes it
easier to get rid of object properties at different sizes,
which eventually results in gains in performance.
The structure of GoogleNet, which includes 22 layers
of deep learning, is shown in Fig 4. This diagram offers an
overview of the system. The layout was created with the
computational efficiency of the finished product in mind,
and it is possible to operate on devices that only need a
little amount of computer power to do so. The GoogleNet
model, which is detailed further down on this page, is used
to create a prediction about COVID in the event that the
individual has previously been exposed to COVID. It starts
with an image of an X-ray of the chest and then goes
through numerous layers of convolution, each of which has
a separate channel. After that, it proceeds through pooling
layers and completely associated layers.
Fig. 5 Inception Module with dimension reduction
At a time when earlier models were only delving deeper
in order to improve both their productivity and their
accuracy while simultaneously balancing the cost of
computing, Inception net made a significant advancement in
CNN classifiers. This occurred at a point in time when
Inception net made its breakthrough. On the other hand, the
Inception network seems to have been put together with
great care. It makes use of extra methods to boost impulse
efficiency, both in terms of speed and accuracy, and as a
result, it has outperformed AlexNet, ZFNet, and VGGNet in
terms of the results it has generated. This is due to the fact
that it utilises more techniques to increase impulse
efficiency. The use of GoogleNet allowed for the completion
of binary classifications, which are also referred to as the
categorising of COVID and non-COVID chest X-ray
photographs.
Authorized licensed use limited to: GITAM University. Downloaded on March 13,2023 at 12:39:34 UTC from IEEE Xplore. Restrictions apply.

The data set that we created is comprised of the
following components: reports of chest scans received from
persons who have been tested for COVID and found to be
positive or negative; reports of chest scans obtained from
individuals who have been diagnosed with COVID; In
addition to this, the results of the X-rays of the different lung
illnesses and ailments are integrated with them.
The data from the samples are distributed as follows:
seven training datasets, three testing datasets, with the ratio
being seven to three.
Which will lower the operational expenses associated
with numerous channels in convolutional neural networks,
which are assessed according to their properties. In addition,
the material that is being done over the image will be pooled
to incorporate information that was generated from the maps
that were made by CNN channels. Pooling is a function that
has the ability to cut down on the restricting limits and time-
consuming processes that are imposed by the association
[10]. The approach that is being recommended makes use of
a few initial layers, and the subsequent yield of each
segment is connected with, and then communicated along
with, the segment that it is paired with.
After the development of the beginning module for the
reformist beginning is complete, the overall ordinary pooling
will be put into effect. Because doing so would bring the
overall number of limitations in the system down, this
pooling does not allow for over-fitting.
The portion that comes after this one is known as the
dropout section. The layer implements a randomization
mechanism to set the input units to zero, which contributes
to the reduction of over-fitting that occurs as a result.
The next layer is the thick layer, which is the very final layer
in the building of the structure.
IV. RESULT
ANALYSIS
Convincing local shops to provide their goods and
services over the internet is one of the most difficult issues.
If you go through all of the many applications that are now
available, you will only discover a select few of them that
sell their wares. Again, the question of trust arises in this
context since the vendors do not actually have any
awareness of the people to whom they are selling their
wares.
The model was constructed for COVID identification
using GoogleNet, with the dataset separated into two parts:
first, 70% of the information was used for preparation, and
then 30% of the information was used for testing. We were
able to get a preparation accuracy of 99% and an approval
exactness of 98% when we were working on the model.
During the preparatory interaction, the values of the
Epochs parameter were observed, which is beneficial for the
improved presentation of the model.
TABLE.1 HYPERPARAMETERVALUES
Epoch
s
Tes
t
siz
e
Batc
h
size
Initial
learni
n g
rate
Trainin
g
Accurac
y
Validatio
n
Accurac
y
10
32
51 49
20
32
96 93
25
32
97
30
32
98 98
40
32
98
50
32
98 96
Finally, we were able to accomplish an accuracy of 98%
in our preparation by having an execution check with an
unparalleled precision of 1.00 and a recall of 0.97. Fig
displays the confusion matrix that was calculated for the
dataset that was tested. 5 and it thus demonstrates the
performance of the categorization model. This confusion
matrix, which can be shown in Fig 5, is used in order to
evaluate performance parameters like as accuracy, precision,
and recall.
Fig. 6 Test set’s confusion matrix
In artificial intelligence (AI), assumptions to learn and
alter are a device for expression that is used widely in
algorithms that constantly acquire a dataset. These are also
known as learning assumptions and modifying assumptions.
Following each update that is made while getting ready, the
model could be polled on both the intended dataset and a
holdout endorsement dataset. In addition, graphs of the
purposeful introduction may be constructed in order to
emphasise the assumptions that must be made in order to
keep the data. This learning picking up preparation history
plot displaying accuracy and misfortune bends demonstrates
that our model is not overfitting despite the limited COVID-
19 X-rays preparing information that is being used in our
Keras / TensorFlow models. This plot shows how the bends
in the accuracy and misfortune curves differ. Fig. 6 When
preparing, it may be useful to review the expectations to
learn and adjust the models as part of the preparation
process. This may help uncover problems about learning,
such as determining whether the preparation and approval
datasets are handled in an appropriate manner. Reviewing
these expectations may help evaluate, for instance, if a
model is underfit or overfit for its purpose. In addition, there
is a need to educate people about the false positive rate,
which specifies that a person who has not been infected with
the virus should not be wrongly labelled as "Coronavirus
positive." This is something that has to be brought to more
people's attention. Placing them in isolation with other
people who are positive for COVID-19, which will result in
the individual's health deteriorating even if they were never
infected with the virus in the first place.
Authorized licensed use limited to: GITAM University. Downloaded on March 13,2023 at 12:39:34 UTC from IEEE Xplore. Restrictions apply.

Fig.
7 Graph of Training & Validation Accuracy
V. C
ONCLUSION
The writers of this research have developed a way with
the aid of GoogleNet to analyse the X-ray pictures of the
patients to determine whether the samples included the
coronavirus. This approach was used to identify the
presence of the virus. To combat the virus, we devised a
method that included giving the system specific instructions
to follow to improve its detection capabilities. With the
assistance of this model, it is possible to make an accurate
prediction of the COVID-19 forecast. During the period of
model preparation, the level of accuracy achieved by our
prepared model may reach up to 98.019%, however during
the phase of model testing, it averages approximately
97.096%. The framework of Our technique might
furthermore be picked throughout the process of producing a
big sample data set of X-ray scans, as well as even regular
individuals who are suffering from other ailments, such as
pneumonia or heart difficulties. This improves one's identity
as well as their power to affect others. It is probable that the
healthcare specialists would find the suggested solution to
be helpful when paired with robots.
R
EFERENCES
[1]. S. Bazaid, A. Aldarhami, N. K. Binsaleh, S. Sherwani, and O. W.
Althomali, “Knowledge and practice of personal protective measures
during the COVID-19 pandemic: a crosssectional study in Saudi
Arabia,” PLoS One, vol. 15, no. 12, article e0243695, 2020.
[2]. A. S. Bazaid, H. Barnawi, H. Qanash et al., “Bacterial coinfection and
antibiotic resistance profiles among hospitalised COVID-19 patients,”
Microorganisms, vol. 10, no. 3, p. 495, 2022.
[3]. M. Y. E. Yousif, M. Eljack, M. S. Haroun et al., “Clinical
characteristics and risk factors associated with severe disease
progression among COVID-19 patients in Wad Medani isolation
centers: a multicenter retrospective cross-sectional study,” Health
Science Reports, vol. 5, no. 2, article e523, 2022.
[4]. M. H. Alyami, A. Y. Naser, M. A. A. Orabi, H. Alwafi, and H. S.
Alyami,“Epidemiology of COVID-19 in the Kingdom of Saudi
Arabia: an ecological study,” Frontiers in Public Health, vol. 8, pp.
506–506, 2020.
[5]. S. Alle, A. Kanakan, S. Siddiqui et al., “COVID-19 risk stratification
and mortality prediction in hospitalized Indian patients: harnessing
clinical data for public health benefits,” PLoS One, vol. 17, no. 3,
article e0264785, 2022.
[6]. G. Ponti, M. Maccaferri, C. Ruini, A. Tomasi, and T. Ozben,
“Biomarkers associated with COVID-19 disease progression,”
Critical Reviews in Clinical Laboratory Sciences, vol. 57, no. 6, pp.
389–399, 2020.
[7]. J. Sharma, R. Rajput, M. Bhatia, P. Arora, and V. Sood, “Clinical
predictors of COVID-19 severity and mortality: a perspective,”
Microbiology, vol. 11, p. doi:10.3389/fcimb.2021.674277, 2021.
[8]. Y.-C. Chang, P.-H. Tsai, Y.-C. Chou, K.-C. Lu, F.-Y. Chang, and C.-
C. Wu, “Biomarkers linked with dynamic changes of renal function in
asymptomatic and mildly symptomatic COVID-19 patients,” Journal
of Personalized Medicine, vol. 11, no. 5, p. 432, 2021.
[9]. M. Laverack, R. L. Tallmadge, R. Venugopalan et al., “Clinical
evaluation of a multiplex real-time RT-PCR assay for detection of
SARS-CoV-2 in individual and pooled upper respiratory tract
samples,” Archives of Virology, vol. 166, no. 9, pp. 2551–2561,
2021.
[10]. E. J. Williamson, A. J. Walker, K. Bhaskaran et al., “Factors
associated with COVID-19-related death using Open safely,” Nature,
vol. 584, no. 7821, pp. 430–436, 2020.
[11]. F. Cosentino, V. Moscatt, A. Marino et al., “Clinical characteristics
and predictors of death among hospitalized patients infected with
SARS-CoV-2 in Sicily, Italy: a retrospective observational study,”
Biomedical Reports, vol. 16, no. 5, p. 34, 2022.
[12]. T. Struyf, J. J. Deeks, J. Dinnes et al., “Signs and symptoms to
determine if a patient presenting in primary care or hospital outpatient
settings has COVID-19,” Cochrane Database of Systematic Reviews,
vol. 2021, no. 3, article CD013665, 2021.
[13]. A. Carfì, R. Bernabei, F. Landi, and for the Gemelli Against COVID-
19 Post-Acute Care Study Group, “Persistent symptoms in patients
after acute COVID-19,” JAMA, vol. 324, no. 6, pp. 603–605, 2020.
[14]. M. Zakerkish, M. S. Fooladi, H. B. Shahbazian, F. Ahmadi, S.
Peyman Payami, and M. Dargahi-Malamir, “Assessment of mortality
rate, need for ICU admission and ventilation in COVID-19 patients
with diabetes mellitus,” Qatar Medical Journal, vol. 2022, no. 1,
2022.
[15]. Islam, N., Ebrahimzadeh, S., Salameh, J. P., Kazi, S., Fabiano, N.,
Treanor, L., ... & COVID, C. (2021). Thoracic imaging tests for the
diagnosis of COVID‐19. Cochrane Database of Systematic Reviews,
(3).
Authorized licensed use limited to: GITAM University. Downloaded on March 13,2023 at 12:39:34 UTC from IEEE Xplore. Restrictions apply.