Automatic COVID-19 lung images classification system based on convolution neural network

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 12, No. 5, October 2022, pp. 5573~5579
ISSN: 2088-8708, DOI: 10.11591/ijece.v12i5.pp5573-5579  5573
Journal homepage: http://ijece.iaescore.com
Automatic COVID-19 lung images classification system based
on convolution neural network
Lamia Nabil Mahdy1
, Ahmed Ibrahim Bahgat El Seddawy2
, Kadry Ali Ezzat1
1
Biomedical Engineering Department, Higher Technological Institute, 10th of Ramadan City, Egypt
2
Computer Science and Information System, Arab Academy for Science and Technology and Maritime Transport, Giza, Egypt
Article Info ABSTRACT
Article history:
Received Aug 27, 2021
Revised May 24, 2022
Accepted Jun 20, 2022
Coronavirus disease (COVID-19) still has disastrous effects on human life
around the world. For fight that disease. Examination on the patients who
have been sucked in quick and cheap way is necessary. Radiography is most
effective step closer to this target. Chest X-ray is readily obtainable and
cheap option. Also, because COVID-19 is a virus, distinguish COVID-19
from common viral pneumonia from common viral pneumonia is difficult.
In this study, X-ray images of 500, 500, 500, and 500 patients for healthy
controls, typical viral pneumonia, bacterial pneumonia and COVID-19, were
collected respectively. To our knowledge, this was the first quaternary
classification study that also included classical viral pneumonia. To
efficiently capture nuances in X-ray images, a new model was created by
applying convolution neural network for accurate image classification. Our
model outperformed to achieve an overall accuracy, sensitivity, specificity,
F1-score, and area under curve (AUC) of 0.98, 0.97, 0.98, 0.97, and 0.99
respectively.
Keywords:
Classification
Computed tomography
Convolution neural network
COVID-19
X-Ray
This is an open access article under the CC BY-SA license.
Corresponding Author:
Kadry Ali Ezzat
Department of Biomedical Engineering, Higher Technological Institute
Next to Small Industries Complex, Industrial Area, 10th
of Ramadan City, Ash Sharqia Governorate
Email: kadry_ezat@hotmail.com
1. INTRODUCTION
Chest computed tomography (CT) performs essential part in patients evaluation with coronavirus
disease (COVID-19). So far, deep learning (DL)-based chest CT analysis has mainly focused on diagnosing
or classifying COVID-19 in order to distinguish between positive COVID-19 patients and patients with
normal pneumonia. The widespread implementation of reverse transcription polymerase chain reaction
(RTPCR) testing reduces the need to use screening radiographs. However, radiographic imaging is still
important to detect lung intrusion in COVID-19 patients. In particular, it has been proposed to use CT scans
of lesions as surrogate imaging of disease burden and to identify critically ill patients requiring
hospitalization.
As assured by world health organization [1], coronavirus recorded the first formal cases in Wuhan,
China. To date, millions of people have been killed, and there are millions of confirmed cases worldwide.
Epidemics in the form of pandemics have a devastating impact on health of people all over world [2]. This
infectious infection, expands quicker than other common flus. RTPCR used to distinguish COVID-19 [3].
This examination seems to be accurate. Even if the patient's sample contains only a small amount of the
virus. However, polymerase chain reaction (PCR) examinations are complex, take long time and expensive.
Therefore, not all health care institutions are able to do this. With these limitations in mind, a standalone
approach to disease detection is to analyze chest radiographs to detect the presence or absence.

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 12, No. 5, October 2022: 5573-5579
5574
The main problem with radiographic-based diagnosis of COVID-19 patients is the lack of trained
and well-available physicians. Furthermore, the radiation symptoms associated with COVID-19 are
unknown, as many specialized have no expertise with COVID-19 infected patients. Recently, many methods
have emerged that use artificial intelligence to distribute pneumonia and COVID-19 from CT or X-rays.
There are basically two types: COVID-19 to non-COVID-19 and COVID-19 to other pneumonia. Xu et al.
[4] introduced a distinguish system for classifying COVID-19, influenza, and healthy individuals with 86.7%
accuracy. Li et al. [5] ResNet50 was used to distinguish COVID-19 from non-pneumonia with 90%
sensitivity. Chen et al. [6], Ouyang. [7], Apostolopoulos et al. [8]. However, because COVID-19 is also a
variant of the virus, all this work ignores the typical viral pneumonia that infects the most difficult and
typical virus.
In this study, X-ray images of 500, 500, 500, and 500 patients for healthy controls, typical viral
pneumonia, bacterial pneumonia and COVID-19, were collected respectively. According to our knowledge,
this was the first quaternary classification study that also included classical viral pneumonia. To efficiently
capture nuances in X-ray images, a new model was created by a recently developed convolution neural
network for accurate image analysis. Our model outperformed the commonly used reference model,
achieving an overall accuracy, sensitivity, specificity, F1-score, and AUC of 0.98, 0.97, 0.98, 0.97, and 0.99
respectively.
2. METHOD
2.1. Data set and model inputs
In current work, chest X-ray obtained from various open sources [9]–[15]. This data set contains
4,900 X-ray images, categorized into three sectors, namely natural, COVID-19 and pneumonia. The
pneumonia category consists of bacterial pneumonia images and viruses pneumonia images. In Figure 1, the
chest X-ray imaging dataset is rearranged into four types: normal, bacterial pneumonia, viral pneumonia, and
COVID-19 as shown in Figures 1(a) to 1(d) respectively.
(a) (b) (c) (d)
Figure 1. Chest X-ray samples from the test datasets (a) normal, (b) bacteria, (c) virus and (d) COVID-19
2.2. Pre‑processing operations
Since this data set consists of large and different lung images; and in order to handle reasonable
computing time during the convolutional neural network (CNN) training experiment, all images are adjusted
to a single dimension and rescaled to smaller images. 310×310 to adapt to the standard input of a custom
structure. To standardize the input of CNN, the original image is preliminarily divided before the resizing
step. Regarding the test, the trained model was tested on distinct dataset taken from IEEE DataPort [16].
2.3. Data augmentation
Data augmentation is method may drastically growth number of data times in dataset for training a
model [17]. For the four categories image datasets (healthy controls, typical viral pneumonia, bacterial
pneumonia and COVID-19, this method utilizes basal image procedures like flip, rotate, crop, and pad for
enlargement. Next, the dataset will be expanded to include these converted images resulting from subsisting
image set. This enlarges dataset size for training neural networks [18]–[24]. A data expansion method was
used in this study to fix availability of few datasets problem that impacts proposed CNN performance. This
technique has expanded the dataset. Moreover, extra learning countenances are available in knowledge
model. In this study, image transformation procedures, mirroring, and rotation, were utilized for data
expansion. In initial stage of data expansion, 100 X-ray images were mirrored to capture an additional 100
images. After using this operation, the resulting dataset was expanded to 180 images. In second stage,
authentic 100 images are rotated another 90 to obtain another 100 images, after that, rotated 180 degrees to

Int J Elec & Comp Eng ISSN: 2088-8708 
Automatic COVID-19 lung images classification system base … (Lamia Nabil Mahdy)
5575
obtain another 100 images, and finally the original 100 images rotated the image of 270 degrees further and
100 more images were obtained. These operations created 500 datasets each for healthy controls, typical viral
pneumonia, bacterial pneumonia, and COVID-19 X-rays. Table 1 introduces image transformation
procedures applied on all image types and the related images generated number by operations. Figure 2
introduces impact of the extension technique performed to authentic image sample of dataset utilized in this
work. Figure 2(a) is the ordinary image, Figure 2(b) is the rotated image with 90, Figure 2(c) is the rotated
image with 180, Figure 2(d) is the rotated image with 270 and Figure 2(e) is the flipped image.
Table 1. Total image number after augmentation
Image augmentation
operation
Number of healthy
controls images
Number of typical viral
pneumonia images
Number of bacterial
pneumonia images
Number of
COVID-19 images
Ordinary 100 100 100 100
Rotation of ordinary images
by 90
100 100 100 100
by 180
100 100 100 100
by 270
100 100 100 100
Flipping of ordinary images 100 100 100 100
Total 500 500 500 500
(a) (b) (c) (d) (e)
Figure 2. X-ray image augmentation methods on an (a) ordinary, (b) 90 rotation, (c) 180 rotation,
(d) 270 rotation, and (e) flipping
2.4. Convolutional neural networks (CNNs)
CNNs are invigorated through the human brain’s visual framework. So, the background of CNNs is
to allow computers to see the world people see. Therefore, CNNs can be utilized in image analysis and
classification [25]. CNN is type of deep neural network with convolution, max pool, and non-linear
activation level. The convolutional layer, which can be viewed as the CNN kernel layer, applies process
known as "convolution," which CNN is named after. The base of convolutional layer is performed to input
layer. Convolutional layers outputs are decreased to feature maps. In this work, rectified linear unit (ReLU)
was utilized for activation task with a convolutional layer that helps to expand nonlinearity of input image
because image is nonlinear. Therefore, CNNs with ReLU are easier and faster in the current situation. Since
ReLU characterized as in (1).
𝑍 = 𝑀𝑎𝑥(0, 𝑖) (1)
Here the function assumes that the output z is zero for all negative values, and positive values keep i constant.
2.4.1. The proposed system
As appeared in Figure 3, the study starts with the series of essential dataset consists of four image
categories: healthy controls, typical viral pneumonia, bacterial pneumonia, and COVID-19. The X-ray image
was preprocessed concurring to the preprocessing above steps. Dataset expanded utilizing standard expansion
methods to expand its size. Resulting dataset was utilized to train model in next stage. Then, model's disease
detection performance was tested. The proposed CNN model tests were done utilizing primary dataset test
and independent validation dataset. Table 2 details the dataset, including training sets, test sets, and number
of X-ray images for the four prediction classes. Finally, aggregation image-level predictions for image of
each case were done to provide the reasonable result.

 ISSN: 2088-8708
5576
Figure 3. System workflow
Table 2. Images number for training and testing
Dataset Healthy Controls Typical Viral Pneumonia Bacterial Pneumonia COVID-19 Total
Images number 500 500 500 500 2,000
Training 400 400 400 400 1,600
Testing 100 100 100 100 400
3. RESULTS AND DISCUSSION
The proposed classification system for COVID-19 diagnostics was implemented on a processor with
the programming language Python 3.8 and Intel R Core i58300H CPU @ 2.30GHz x 8 and 8 GB RAM,
Windows 10, RADEON AMD 1050, and 4 GB graphics. Google Collab GPU, Python 3.7 and Tensor Flow
2.2.0. The Tensor Flow 2.2.0 deep learning library is used to implement CNN, and training and testing steps
are performed on the Google Collab platform.
In this model, CNN used to effectively classify COVID-19. Proposed technique is trained on four
public datasets and outperforms itself in all classes. CNN is used for the final classification with accuracy,
sensitivity, specificity, F1 score and AUC giving the best results of 0.98, 0.97, 0.98, 0.97, and 0.99
respectively as in Table 3. So that, this approach using X-ray imaging and computational diagnostics can be
used as a large, faster, and cheaper screening method. It also reduces the required testing time. Training with
larger datasets and field testing with larger cohorts can be very helpful in making clinically effective
predictions of COVID-19.
Table 3. Model overall performance primarily based totally on unbiased validation data
Evaluation parameters Degree
Accuracy 0.98
Sensitivity 0.97
Specificity 0.98
F-1 score 0.97
ROC AUC 0.99
Receiver operating characteristic area under curve (ROC AUC)
To examine the overall performance, classification accuracy, sensitivity, specificity and F1-score,
measured:
Accuracy=
TPR + TNR
TPR + FPR + TNR + FNR
(2)
Sensitivity=
TP
TP+FN
(3)
Specificity=
TN
TN+FP
(4)
F1-score=
2*TP
2*TP+FP+FN
(5)
where TP stands for true positive, TN for true negative, FP for false positive, FN for false negative and FP
for false positive TN for true negative. Figure 4 shows train and test losses of the proposed CNN model. As
can be seen in Figure 4, proposed CNN does not take much time to converge. Figure 5 shows a graph of

5577
training and test accuracy done by the proposed CNN model. Figures 6 and 7 shows the characteristic curve
of the receiver operating characteristics at AUC=0.99 and the sensitivity-specificity curves for the various
case types respectively.
Figure 4. Train and test loss plot by using CNN
Figure 5. Train and test accuracy plot by using CNN
Figure 6. ROC curve of classification Figure 7. Sensitivity and specificity curve of
classification
As evidence of introduced CNN system in classification importance for detecting COVID-19 from
X-ray images, trained model was examined on distinct dataset from IEEE DataPort. The distinct test dataset
consists of 100 COVID-19 X-rays. 100 regular images were added to dataset for testing. In addition, a feature
map of four different X-RAY images types excerption by the classification model and examined overall

 ISSN: 2088-8708
5578
learning feature then renders possibility of network. As in Figure 8, area that the lesion is existed shows a
superior level of response, indicating that the model was eligible to learn underlying properties from
radiography of three different patients’ cases. Figure 8(a) shows healthy X-Ray image, Figure 8(b) shows
COVID-19 X-Ray image, Figure 8(c) shows viral pneumonia image and Figure 8(d) shows bacterial
pneumonia image.
(a) (b) (c) (d)
Figure 8. The X-Ray images feature maps four types of people extracted by proposed classification model
(a) healthy X-Ray image, (b) COVID-19 X-Ray image, (c) viral pneumonia image, and (d) bacterial
pneumonia image
4. CONCLUSION
By applying CNN, X-ray deep learning diagnostic system for rapid COVID-19 diagnosis was
created. This model can individually detect COVID-19 X-rays from healthy individuals, X-rays from patients
with bacterial pneumonia, and X-rays from patients with typical viral pneumonia. This model able to
differentiate so various different kinds of pneumonia at once. Results have shown that introduced model
fulfill high accuracy, specificity, sensitivity, AUC ROC and F1 score that shows the reliability of the model.
REFERENCES
[1] D. Cucinotta and M. Vanelli, “WHO declares COVID-19 a pandemic,” Acta bio-medica : Atenei Parmensis, vol. 91, no. 1,
pp. 157–160, 2020, doi: 10.23750/abm.v91i1.9397.
[2] F. Rustam et al., “COVID-19 future forecasting using supervised machine learning models,” IEEE Access, vol. 8,
pp. 101489–101499, 2020, doi: 10.1109/ACCESS.2020.2997311.
[3] R. Lu et al., “Genomic characterisation and epidemiology of 2019 novel coronavirus: implications for virus origins and receptor
binding,” The Lancet, vol. 395, no. 10224, pp. 565–574, Feb. 2020, doi: 10.1016/S0140-6736(20)30251-8.
[4] X. Xu et al., “A deep learning system to screen Novel coronavirus disease 2019 pneumonia,” Engineering, vol. 6, no. 10,
pp. 1122–1129, Oct. 2020, doi: 10.1016/j.eng.2020.04.010.
[5] L. Li et al., “Using artificial intelligence to detect COVID-19 and community-acquired pneumonia based on pulmonary CT,”
Radiology, vol. 296, no. 2, pp. E65--E71, Aug. 2020, doi: 10.1148/radiol.2020200905.
[6] J. Chen et al., “Deep learning-based model for detecting 2019 novel coronavirus pneumonia on high-resolution computed
tomography,” Scientific Reports, vol. 10, no. 1, Dec. 2020, doi: 10.1038/s41598-020-76282-0.
[7] X. Ouyang et al., “Dual-sampling attention network for diagnosis of COVID-19 from community acquired pneumonia,” IEEE
Transactions on Medical Imaging, vol. 39, no. 8, pp. 2595–2605, Aug. 2020, doi: 10.1109/TMI.2020.2995508.
[8] I. D. Apostolopoulos and T. A. Mpesiana, “COVID-19: automatic detection from X-ray images utilizing transfer learning with
convolutional neural networks,” Physical and Engineering Sciences in Medicine, vol. 43, no. 2, pp. 635–640, Jun. 2020, doi:
10.1007/s13246-020-00865-4.
[9] J. Rasheed, A. A. Hameed, C. Djeddi, A. Jamil, and F. Al-Turjman, “A machine learning-based framework for diagnosis of
COVID-19 from chest X-ray images,” Interdisciplinary Sciences: Computational Life Sciences, vol. 13, no. 1, pp. 103–117, Mar.
2021, doi: 10.1007/s12539-020-00403-6.
[10] Y. Pathak, P. K. Shukla, and K. V Arya, “Deep bidirectional classification model for COVID-19 disease infected patients,”
IEEE/ACM Transactions on Computational Biology and Bioinformatics, vol. 18, no. 4, pp. 1234–1241, Jul. 2021, doi:
10.1109/TCBB.2020.3009859.
[11] J. P. Kanne, “Chest CT findings in 2019 novel coronavirus (2019-NCoV) infections from Wuhan, China: key points for the
radiologist,” Radiology, vol. 295, no. 1, pp. 16–17, Apr. 2020, doi: 10.1148/radiol.2020200241.
[12] P. Mooney, “Chest X-Ray images (Pneumonia),” Kaggle, 2019. Accessed: Sep 1, 2019. [Online]. Available:
https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia.
[13] D. Singh, V. Kumar, Vaishali, and M. Kaur, “Classification of COVID-19 patients from chest CT images using multi-objective
differential evolution-based convolutional neural networks,” European Journal of Clinical Microbiology and Infectious Diseases,
vol. 39, no. 7, pp. 1379–1389, Jul. 2020, doi: 10.1007/s10096-020-03901-z.
[14] M. Chung et al., “CT imaging features of 2019 novel Coronavirus (2019-nCoV),” Radiology, vol. 295, no. 1, pp. 202–207, Apr.
2020, doi: 10.1148/radiol.2020200230.
[15] W. Kong and P. P. Agarwal, “Chest imaging appearance of COVID-19 infection,” Radiology: Cardiothoracic Imaging, vol. 2,
no. 1, Feb. 2020, doi: 10.1148/ryct.2020200028.
[16] Z.-H. Chen, “Mask-RCNN detection of COVID-19 pneumonia symptoms by employing stacked autoencoders in deep
unsupervised learning on low-dose high resolution CT,” IEEE DataPort, 2020, doi: 10.21227/4kcm-m312.
[17] C. Shorten and T. M. Khoshgoftaar, “A survey on image data augmentation for deep learning,” Journal of Big Data, vol. 6, no. 1,

5579
Dec. 2019, doi: 10.1186/s40537-019-0197-0.
[18] D. Ho, E. Liang, and R. Liaw, “1000x faster data augmentation – the berkeley artificial intelligence research blog,” Berkeley
Artificial Intelligence Research (BAIR), 2019.
[19] A. S. Al-Waisy et al., “COVID-CheXNet: hybrid deep learning framework for identifying COVID-19 virus in chest X-rays
images,” Soft Computing, Nov. 2020, doi: 10.1007/s00500-020-05424-3.
[20] Z. Han et al., “Accurate screening of COVID-19 using attention-based deep 3D multiple instance learning,” IEEE Transactions
on Medical Imaging, vol. 39, no. 8, pp. 2584–2594, Aug. 2020, doi: 10.1109/TMI.2020.2996256.
[21] H. Kang et al., “Diagnosis of coronavirus disease 2019 (COVID-19) with structured latent multi-view representation learning,”
IEEE Transactions on Medical Imaging, vol. 39, no. 8, pp. 2606–2614, Aug. 2020, doi: 10.1109/TMI.2020.2992546.
[22] S. Albahli, “A deep neural network to distinguish COVID-19 from other chest diseases using X-ray images,” Current Medical
Imaging Formerly Current Medical Imaging Reviews), vol. 17, no. 1, pp. 109–119, Mar. 2021, doi:
10.2174/1573405616666200604163954.
[23] A. Jaiswal, N. Gianchandani, D. Singh, V. Kumar, and M. Kaur, “Classification of the COVID-19 infected patients using
DenseNet201 based deep transfer learning,” Journal of Biomolecular Structure and Dynamics, vol. 39, no. 15, pp. 5682–5689,
Oct. 2021, doi: 10.1080/07391102.2020.1788642.
[24] R. Zhang et al., “COVID19XrayNet: a two-step transfer learning model for the COVID-19 detecting problem based on a limited
number of chest X-Ray images,” Interdisciplinary Sciences: Computational Life Sciences, vol. 12, no. 4, pp. 555–565, Dec. 2020,
doi: 10.1007/s12539-020-00393-5.
[25] D. T. Mane and U. V Kulkarni, “A survey on supervised convolutional neural network and its major applications,” International
Journal of Rough Sets and Data Analysis, vol. 4, no. 3, pp. 71–82, Jul. 2017, doi: 10.4018/IJRSDA.2017070105.
BIOGRAPHIES OF AUTHORS
Lamia Nabil Mahdy received B.Sc. degree in biomedical engineering from
Helwan University, Cairo, Egypt in 2004 and the M.Sc. and Ph.D. degree in biomedical
engineering from Cairo University, Egypt, in 2009 and 2016 respectively. In 2004 she joined
the biomedical engineering department in the higher technological institute as researcher
assistant then she promoted as assistant Lecturer in 2009 and then she promoted to be Lecturer
in 2016. Her current research interests are: diagnostic imaging, robotics, hospital design,
artificial intelligence, reproductive surgery, image processing, expert systems, biomechanics,
computer programming, data transmission, data structures, biomedical instrumentation and
electronics, pattern recognition, microcontrollers, modeling and simulation, internet of things
(IoT). She is now Lecturer (Ph.D.) in Biomedical Engineering Department at Higher
Technological Institute in 10th of Ramadan city since 2016, Member in Scientific Research
Group in Egypt (SRGE), Head of Bright Minds students (BMS) Research Group in Higher
Technological Institute; she supervised on more than 13 under graduate bachelor projects. She
can be contacted at email: englamia_82@yahoo.com.
Ahmed Ibrahim Bahgat El Seddawy is Associated Professor, Vice Dean of
Educational Affairs, and Head of Business Information System Department. He received his
bachelor from Business Administration, MIS department and M.Sc. degree in college of
Computer Science and Information Systems, I.S Department 2008 from, both from AASTMT.
Cairo, Egypt. On December 2014, he received his doctoral degree from the Department of
Information Systems, College of Computer Science, and Information Systems, I.S department,
Helwan University. Working as visitor professor in Ludwigsburg university in Germany and
Proto University Italy. He has authored/co-authored several research publications in Selected
fields such as, knowledge discovery, data science, data analytics decision support systems,
knowledge base systems, project management, and leadership management using ERP. He can
be contacted at email: ahmed.bahgat@aast.edu.
Kadry Ali Ezzat received B.SC degree in biomedical engineering from the higher
technological institute, Cairo, Egypt in 2004 and the M.Sc. and Ph.D. degree in biomedical
engineering from Cairo University, Egypt, in 2011 and 2017 respectively. In 2004 he joined
the biomedical engineering department in the higher technological institute as researcher
assistant then he promoted as assistant Lecturer in 2011 and then he promoted to be Lecturer
in 2017. His current research interests are: diagnostic imaging, robotics, satellite
communications, artificial intelligence, image processing, expert systems, biomechanics, data
transmission, data structures, biomedical instrumentation and electronics, pattern recognition,
microcontrollers, modeling and simulation. He is now Lecturer in Biomedical Engineering
Department at Higher Technological Institute in 10th
of Ramadan city since 2017, Member in
Scientific Research Group in Egypt (SRGE), Consultant member of the Egyptian Remote
Sensing and Space Science Authority (Satellite Alliance Project), Instructor in High
Technology Center, Faculty of Engineering, Cairo University. He is reviewer for papers in
many international conferences: Cairo International Biomedical Engineering Conference
(CIBEC), International Undergraduate Research Conference in Military Technical College
(IUGRC). He is reviewer in Australasian Physical and Engineering Sciences in Medicine
(APES) journal. He can be contacted at email: kadry_ezat@hotmail.com.

Automatic COVID-19 lung images classification system based on convolution neural network

More Related Content

Similar to Automatic COVID-19 lung images classification system based on convolution neural network (20)

More from IJECEIAES (20)

Recently uploaded (20)

Automatic COVID-19 lung images classification system based on convolution neural network