Content uploaded by Yogesh Bhosale
Author content
All content in this area was uploaded by Yogesh Bhosale on Jun 08, 2022
Content may be subject to copyright.
Content uploaded by Yogesh Bhosale
Author content
All content in this area was uploaded by Yogesh Bhosale on Jun 08, 2022
Content may be subject to copyright.
2022 8th International Conference on Advanced Computing and Communication Systems (ICACCS)
Deep Convolutional Neural Network Based
Covid-19 Classification From Radiology X-Ray
Images For IoT Enabled Devices
Yogesh H. Bhosale
Dept. of Computer Science and
Engineering
Birla Institute of Technology (BIT)
Mesra, Ranchi, India.
yogeshbhosale988@gmail.com
Mahendra Nakrani
Dept. of Computer Vision Engineering
WeAgile Software Solution Pvt. Ltd
Pune, India
nakrani.mahender@gmail.com
Shrinivas Zanwar
Dept. of Artificial Intelligence and Data
Science
CSMSS, Chh. Shahu College of
Engineering, Aurangabad, India
shrinivas.zanwar@gmail.com
Devendra Bhuyar
Dept. of Electronics and Computer
Engineering
CSMSS, Chh. Shahu College of
Engineering, Aurangabad, India
devbhuyar@gmail.com
Zakee Ahmed
Dept. of Artificial Intelligence and Data
Science
CSMSS, Chh. Shahu College of
Engineering, Aurangabad, India.
zakee4@gmail.com
Ulhas Shinde
Dept. of Electronics and Telecomm.
Engineering
CSMSS, Chh. Shahu College of
Engineering, Aurangabad, India
devbhuyar@gmail.com
Abstract- The Coronavirus Disease 2019 (COVID19)
epidemic, which erupted at the end of 2019, continued rapidly
throughout the nations from Wuhan, China. This highly
contagious infectious disease is rapidly spreading among the
public. Early research on COVID-19-affected patients has
revealed distinctive anomalies in chest radiography images. As
a result, it is now necessary to identify various risk factors that
can move an infected person from a mild to a serious stage of
sickness. In Deep Learning (DL), strategies as a subset of
Artificial Intelligence (AI) are used to deal with many real-life
glitches. This paper introduces a Deep Convolutional Neural
Network (DCNN) to perform multiclass classification for
COVID-19, Pneumonia, and Normal Patients from radiological
imaging of the chest. Also, the work is implemented with an IoT
framework, used for communicating user and DCNN model.
This Deep Convolutional Neural Network (DCNN) classification
mechanism achieved a perfect test accuracy of 94.95% for
COVID-19. The used datasets are acquired from Kaggle and
GitHub.
Keywords- Deep Learning, Classification, Deep Convolutional
Neural Network, Radiology Images, X-ray, Diagnosis.
I. INTRODUCTION
Nowadays, the notion of the Internet of Things (IoT)
is very much popular in artificial intelligence applications.
So, here we have used IoT-based automation in the deep
learning model. COVID-19 continues to be a major threat
to human health worldwide, with millions estimated to
be affected within a few months after the outbreak and
thousands of deaths [1]. Humans are infected with the
acute respiratory disease coronavirus-2, which causes
COVID-19 (S.A.R.S.-Co.V-2). Effective and accurate
examination of infection is one of the crucial steps in
struggling infection, allowing infected people to start
instant action to cure, also isolating and separating those
infected to prevent the virus from spreading. The most
public diagnostic techniques used to test for infected people
with COVID-19 infection are Antigen Test or Reverse
Transcription Polymerase Chain Reaction (RT-PCR) test [2]
that can detect S.A.R.S-Co.V-2 RNA from
respiratory samples obtained by several
combinations such as nasopharyngeal or oropharyngeal
rods. Although RT-PCR testing is now the yardstick for
COVID-19 infection because of its sensitivity, it is
laborious and demanding. The limited
availability of RT-PCR kits and the need to access a research-
level research center becomes a daunting challenge.
Fig. 1. Chest Radiography Image
Another strategy for RT-PCR testing may be radiography
testing for COVID-19 testing. In the radiographic
examination, experienced radiologists perform and analyze
chest radiography imaging. Radiologists then try to remove
the visual cues to diagnose SARS-CoV-2 virus contamination,
as shown in Fig. 1. Most of the visual cues from radiology
images of the COVID-19 chest showed similar features such
as circular morphology, ground-glass opacities, lung
distribution, and lung integration [3]. Although radiological
chest pictures may assist in the initial diagnosis of doubted
cases, the features of diverse major pneumonia are similar.
Consequently, it is difficult for diagnosticians to detect
COVID-19 from other types of pneumonia. This leads to the
search for a computer-assisted test program (CAD) to assist
the radiologist in translating radiological chest images of
COVID-19 classification accurately and rapidly.
The major contributions are as follows:
•A custom DCNN classification model is
recommended that will be employed to distinguish
COVID-19 individuals using X-ray images.
•To improve classifier efficiency, different
preprocessing and training methods were used.
•The samples in public repositories are small and
skewed. We used multi-control data augmentation to
address this while considering the samples for
all classes.
1398
978-1-6654-0816-5/22/$31.00 ©2022 IEEE
2022 8th International Conference on Advanced Computing and Communication Systems (ICACCS) | 978-1-6654-0816-5/22/$31.00 ©2022 IEEE | DOI: 10.1109/ICACCS54159.2022.9785113
Authorized licensed use limited to: Birla Institute of Technology. Downloaded on June 08,2022 at 05:57:00 UTC from IEEE Xplore. Restrictions apply.
2022 8th International Conference on Advanced Computing and Communication Systems (ICACCS)
•This DCNN classification mechanism can help the
radiologist and clinicians to find COVID-19.
The remaining section is arranged as follows. Section II
argues the associated work. Section III offers implemented
work. Results and conclusion are mentioned in sections IV
and V, respectively.
II. RELATED WORK
The need for a CAD program to acquire COVID-19
accurately and rapidly has led to many in-depth AI learning
solutions over the past few months. For instance, in [4], the
region of interest (ROIs) are extracted from the computer
tomography (CT) images by using a modified inception
network. The features are extracted from these ROIs for
classification. The classification was done using ensemble
learning of Adaboost and Decision Tree. In [5], transfer
learning was proposed to extract the features using pre-trained
CNNs such as Resnet, Xception, Inception, Densenet, and
Googlenet. The classification of COVID-19 from the
extracted features was done using a support vector machine
(SVM).
COVID-net, a residual architecture-based CNN, was
proposed in [6] to classify the chest radiography images.
DeTraC, a pre-trained CNN, was proposed in [7] to extract the
deep local features. In this case, a layer of phase drop is
attached and a layer of phase structure to provide final
separation. In [8], CNN's comprehensive study model
consisting of three elements, a spinal network, a split head,
and a discovery head, was proposed. Advanced features are
extracted using a spinal network. These high-level features are
applied to the split head and to the wrong detection head,
which gave the separation points and anomaly points,
respectively. The classification is done by applying the
threshold to the average classification score and anomaly
score.
Ibrahim et al. [9] proposed the recognition of pneumonia
with COVID-19 or non-COVID-19 and bacterial pneumonia
using a DNN. Two binary classes and a multiclass model were
used to train models. These infections are all identified in two
binary categories, as well as healthy CXR images. Emtiaz
Hussain et al. [10] introduced CoroDet, a CNN method for
automatic recognition using X-ray of green chest and scanning
computed tomography scans. CoroDet was created as an
appropriate diagnostic tool for the division of two classes,
three classes, and four categories.
The Support Vector Machines (SVM) separator has been
schooled distinguish features, with multiple kernels functions
like Linear, Quadratic, Cubic, and Gaussian by Aras M.
Ismael et al. [11]. The good rescue process also uses in-depth
CNN-trained advanced models mentioned earlier. This project
proposes a new CNN method for end-on training. Research
performance was measured using category accuracy.
Experimental results suggest that COVID-19 can be obtained
from X-ray images. The in-depth CNN study is proposed in
this paper to separate radiology chest images into Pneumonia,
COVID-19, and standardized images.
III. PROPOSED CLASSIFIER WITH IOT FRAMEWORK
The IoT framework for the classification of COVID19 is
as shown in Figure 2. It describes as patient/doctor will have
a handheld device to capture X-ray images uploaded on server
AWS S3 storage. The web server accepts the uploaded image
after authentication and tests it on the trained model.
The deployed model is already running on the AWS
Cloud9 EC2 environment; the training weights are stored. On
the other hand, testing will give a decision and send it back to
the user application.
Fig. 2. IoT framework
Fig. 3. Classification Model
The classification model is shown in Figure 3, which is
used to model the classification analysis for COVID 19.
Initially, the data set is collected from a variety of sources.
Preprocessing stages are necessary for modifying data;
thresholding and segmentation are done at this stage. Then,
construct a convolutional neural network to determine
whether or not it is COVID-infected.
A. Data set
The two publicly available databases were combined to
train, validate, and test the suggested DCNN. The first data
comes from the RSNA Pneumonia Detection Challenge [12]
of the Kaggle competition. This database contains pneumonia
and non-pneumonia (e.g., General) of 25000 cases of
radiology imaging. Database estimates only 165 pneumonia
images and 170 randomized radiology images were randomly
selected for testing. The process of adding an image is used to
increase the size of the database; the final details of the
database are given in Table 1.
The second database is the COVID-19 [13] database from
GitHub. This database contains 350 COVID-19 radiology
chest images, of which only 220 chest radiology images were
Data Set Preprocessing
Deep
Convolutional
Neural
Network
Classification
1399
Authorized licensed use limited to: Birla Institute of Technology. Downloaded on June 08,2022 at 05:57:00 UTC from IEEE Xplore. Restrictions apply.
2022 8th International Conference on Advanced Computing and Communication Systems (ICACCS)
selected. It is unknown at this time what will do after leaving
the post. This highlights the lack of COVID-19 radiology
images offered in the community domain and the need to
improve the quality of research data needed to manage this
epidemic in the future.
TABLE I. DATABASE DETAILS USED FOR TRAINING AND TESTING
Disease Type Image Samples
Normal 1540
COVID-19 1520
Pneumonia 1560
B. Preprocessing
Since we are using CNN for feature extraction and
classification purpose, there is not necessary to perform more
data correction through preprocessing. Till three stages are
followed here in this phase as the resize, augmentation, and
equalization to increase the number of training images and
segmentation to correct dataset.
1) Resize Data
Resizing is the process of changing the size of an image
without leaving out anything. All chest radiography image
images are resized 256x256 in preprocessing. Improving our
system's performance is needed by reducing training time
[14]. After this step, data is labeled, which is required to train
the neural network.
2) Image Augmentation
It is a method that allows you to improve the number of
training images by intelligently modifying your existing
images. There are several methods in augmentation as image
rotating, shifting, flipping, noising, blurring, etc. [15]. Image
augmentation is a regularization strategy for reducing model
overfitting by producing new training samples from existing
train set. The flipping, noising, and blurring methods are
applied here in the image augmentation process.
3) Histogram Equalization
It is a technique used to enhance brightness by changing
the intensity of the image. It does this by successfully
distributing the most common solidity values, i.e., extending
the image's width [16].
C. Deep Convolutional Neural Network (DCNN)
We suggested 2-dimensional DCNN for COVID-19
classification from X-ray based with residual network, as
shown in Fig 4.
The proposed DCNN has five Convolutional blocks
passed through the residual connections. Each convolutional
block contains three batch normalization layers that comprise
two convolution layers, as shown in Fig 5. Filter numbers 16,
32, 64, 128, and 256 are used for each convolution layer. The
residual connection has one layer of convolution with the
same size and number of filters in the convolution block to be
added to the add-on layer. The max-pooling layer is
introduced after the additional layer, reducing the range of the
features.
The Rectified Linear Unit (ReLu) function is applied on
convolution layers, as described in Eq. 1. The formation of a
deep neural convolution network model has different layers,
as exposed in Fig. 3. The first layer is the insertion layer, with
a size of (256, 256, 3). Softmax[17] function, as in Eq. 2, is
used to activate the final layer, i.e., dense layer, and provides
three possibilities for each class, namely COVID-19,
Pneumonia, and Normal.
Fig. 4. Deep Convolutional Neural Network Model
Fig. 5. N- Filter Convolution Blocks.
The dataset is balanced by considering 220 normal images
and 210 pneumonia images selected randomly for the
execution. The image augmentation technique is used to
From Previous Layer
Batch Normalization
Layer
Convolution Layer @ N
Batch Normalization
Layer
Convolution Layer @ N
Batch Normalization
Layer
To Next Layer
1400
Authorized licensed use limited to: Birla Institute of Technology. Downloaded on June 08,2022 at 05:57:00 UTC from IEEE Xplore. Restrictions apply.
2022 8th International Conference on Advanced Computing and Communication Systems (ICACCS)
increase the dataset(section III.B.2), and the final dataset
details are given in Table I.
= max(0, ) (1)
()=
∑
= 1 (2)
IV. RESULTS AND DISCUSSION
The experimentation is executed on Google Colaboratory
(colab) environment based on Python with access to RAM of
25 GB and Tesla K80 GPUs. The python library packages like
Pandas, NumPy, Matplotlib, Sci-Kit learns, etc along with AI
frameworks like TensorFlow or Keras. The planned system is
implemented using the Keras framework.
(a)
(b)
(c)
Fig. 6. Sample radiology images. (a) Normal, (b) COVID- 19, (c) Pneumonia.
The database is divided into training, validation, and
testing at a scale of 70:15:15 for this Convolution Neural
Network model. A training learning rate of 0.001 with an
adaptive algorithm was used to train a model for 25 epochs.
The loose function used was cross-entropy of the phase with
precision as a metric. Fig. 6 shows radiology images of a
patient with standard COVID-19 or Pneumonia.
The planned model performance is measured with
sensitivity, specificity, and recognition rate, as observed in
Table II. Along with that, the confusion matrix and ROC
curves are also drawn.
TABLE II. SPECIFICITY, SENSITIVITY, AND RECOGNITION RATE FOR EACH
DISEASE TYPE
Infection
Type
Specificity
(%)
Sensitivity
(%)
Recognition
Rate (%)
Normal 97.17 94.48 95.55
COVID-19 96.88 94.95 94.94
Pneumonia 98.18 95.17 96.59
Fig. 7. Confusion matrix for proposed DCNN model.
The confusion matrix(CM) of the projected deep
convolutional neural network classifier on the test dataset is
shown in Fig. 7. This CM describes the evaluation of this
DCNN classification model. It gives actual values of true
positive(TP), false negative(FN), true negative(TN), and
false-positive(FP). The columns are treated as predicted
values, rows are actual values, and diagonal elements values
are correctly predicted values. The values (0.94, 0.96, 0.95)
are observed as DCNN prediction.
The Receiver Operating Characteristic Curve(ROC) is as
publicized in Fig. 8, which explains the performance of
combined confusion matrices. The graph is plotted by
considering the false positive and true positive rates.
Fig. 8. ROC Curve for proposed DCNN model.
As we go through comparative analysis with existing
work, Fig. 9 represents the proposed convolution neural
network gives better results. The comparison is made with the
parameters like sensitivity, specificity, and accuracy.
Fig. 9. Comparative analysis with existing work.
V. CONCLUSIONS
The present research work of the DCNN for identifying
COVID-19 and pneumonia from radiology chest imaging is
tested and found useful with better accuracy. The model is
encouraged by the residual network and achieved 94.94% and
96.59% accuracy for COVID-19 and Pneumonia,
respectively, to a considerable amount of database. This
model can also be enhanced using a large database of infected
COVID-
19
Pneumo
nia Normal COVID-
19
Pneumo
nia Normal
Proposed CNN Model Existing ResNet50V2
Models [18]
Sensitivity 94.89 95.17 94.48 74.02 85.54 92.6
Specificity 96.88 98.18 97.28 97.33 92.98 86.64
Accuracy 94.04 96.59 95.55 97.26 98.39 94.85
0
20
40
60
80
100
120
Sensitivity Specificity Accuracy
1401
Authorized licensed use limited to: Birla Institute of Technology. Downloaded on June 08,2022 at 05:57:00 UTC from IEEE Xplore. Restrictions apply.
2022 8th International Conference on Advanced Computing and Communication Systems (ICACCS)
or non-infected cases. The proposed method inspires further
investigation of the developed deep convolutional neural
network model with a larger database for better accuracy in
diagnosis. The use of the IoT has reduced the time of getting
diagnosis reports at the user side with the effective use of the
internet and modern computing technologies.
REFERENCES
[1] Coronavirus disease (COVID-19) Pandemic,
https://www.who.int/emergencies/diseases/novel-coronavirus-2019,
accessed on 25-April-2020.
[2] Wang et al. Detection of SARS-CoV-2 in different types of clinical
specimens. JAMA, 2020.
[3] Huang C, Wang Y, Li X, Ren L, Zhao J, Hu Y, et al. Clinical features
of patients infected with 2019 novel coronavirus in Wuhan, China.
Lancet 2020.
[4] Shuai Wang, Bo Kang, Jinlu Ma, Xianjun Zeng, Mingming Xiao, Jia
Guo, Mengjiao Cai, Jingyi Yang, Yaodong Li, Xiangfei Meng, Bo Xu,
"A deep learning algorithm using CT images to screen for Corona
Virus Disease (COVID-19)", medRxiv 2020.02.14.20023028, 24 April
2020.
[5] Sethy, P.K, Behera, S.K, “Detection of Coronavirus Disease (COVID-
19) Based on Deep Features”, Preprints 2020, 2020030300 (doi:
10.20944/preprints202003.0300.v1).
[6] Wang L, Wong A, "COVID-Net: A Tailored Deep Convolutional
Neural Network Design for Detection of COVID-19 Cases from Chest
Radiography Images", arXiv:2003.09871v2 [eess.IV], 30 Mar 2020.
[7] Asmaa Abbas, Mohammed M. Abdelsamea, Mohamed Medhat Gaber,
"Classification of COVID-19 in chest X-ray images using DeTraC
deep convolutional neural network", arXiv:2003.13815v2 [eess.IV] 18
Apr 2020.
[8] Jianpeng Zhang, Yutong Xie, Yi Li, Chunhua Shen, and Yong Xia,
"COVID-19 Screening on Chest X-ray Images Using Deep Learning-
based Anomaly Detection", arXiv:2003.12338v1 [eess.IV] 27 Mar
2020.
[9] Ibrahim, A.U., Ozsoz, M., Serte, S. et al. Pneumonia Classification
Using Deep Learning from Chest X-ray Images During COVID-
19. Cogn Comput (2021). https://doi.org/10.1007/s12559-020-09787-
5
[10] Hussain, Emtiaz, Mahmudul Hasan, Md Anisur Rahman, Ickjai Lee,
Tasmi Tamanna, and Mohammad Zavid Parvez. "CoroDet: A deep
learning based classification for COVID-19 detection using chest X-
ray images." Chaos, Solitons & Fractals 142 (2021): 110495.
[11] Ismael, Aras M., and Abdulkadir Şengür. "Deep learning approaches
for COVID-19 detection based on chest X-ray images." Expert
Systems with Applications 164 (2021): 114054
[12] Radiological Society of North America. RSNA pneumonia detection
challenge. https://www.kaggle.com/c/rsnapneumonia-detection-
challenge/data, 2019.
[13] Cohen et al. COVID-19 image data collection.
https://github.com/ieee8023/covid-chestxray-dataset, 2020.
[14] Haralick, Robert M., and Linda G. Shapiro. "Image segmentation
techniques." Computer vision, graphics, and image processing 29, no.
1 (1985): 100-132.
[15] Perez, Luis, and Jason Wang. "The effectiveness of data augmentation
in image classification using deep learning." arXiv preprint
arXiv:1712.04621 (2017).
[16] Togacar, Mesut, Burhan Ergen, and Zafer Cömert. "COVID-19
detection using deep learning models to exploit Social Mimic
Optimization and structured chest X-ray images using fuzzy color and
stacking approaches." Computers in biology and medicine 121 (2020):
103805.
[17] Alazab, Moutaz, Albara Awajan, Abdelwadood Mesleh, Ajith
Abraham, Vansh Jatana, and Salah Alhyari. "COVID-19 prediction and
detection using deep learning." International Journal of Computer
Information Systems and Industrial Management Applications 12
(2020): 168-181.
[18] Md. Rahimzadeh, Abolfazl Attar. A modified deep convolutional
neural network for detecting COVID-19 and pneumonia from chest X-
ray images based on the ResNet50V2. Informatics in Medicine.
1402
Authorized licensed use limited to: Birla Institute of Technology. Downloaded on June 08,2022 at 05:57:00 UTC from IEEE Xplore. Restrictions apply.