Predicting Covid-19 pneumonia Severity on Chest x-ray with deep learning

© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1781
Predicting Covid-19 pneumonia Severity on Chest x-ray with deep
learning
Anand Kumar Y. C.1, Dr. Dhananjaya V2
1M. tech student, Computer Science and Engineering, ICEAS, Karnataka, India
2Professor and Head, Computer Science and Engineering, ICEAS, Karnataka, India
-----------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Covid-19 is a viral disease that is quickly
spreading and infects not only humans but animals as
well. The economy of the country is affected by human
beings’ health because of this deadly disease. According to
a clinical study the patients with this disease are
frequently infected from lung infections coming in
connection with the patients with this disease. The most
efficient techniques to diagnose lung-related problems are
Chest x-ray (i.e., radiography) and chest computed
tomography. A huge amount of chest x-ray images
provides a critical effect on the showing of Covid-19 using
deep learning. In this study, Posterior – Anterior sight of
chest x-ray examinations for both covid-19 affected
patients as well as healthy patients. The examination of
accuracy is compared with InceptionV3, Xception, and
ResNet models. Out of 6432, 5467 training samples were
used for training and 965 samples for validation to
analyze the model performance collected from the Kaggle
repository. The Xception model provides the maximum
precision (i.e., 97.97%) for sensing chest X-ray images as
compared to other models. This study only emphasizes on
potential methods of categorizing covid-19 diseased
patients and does not claim any medicinal accuracy.
Keywords: Artificial Intelligence; Covid-19; Convolution
Neural Network (CNN); Machine Learning; X-rays.
1. INTRODUCTION
Machine learning is a sub-branch of artificial intelligence
(AI). The aim of machine learning in general is to know
the structure of data and fit that data into models that can
be understood and utilized by people. Machine learning is
a turf in computer science and differs from traditional
computing, computers are explicitly programmed to
calculate or solve the problem. Data inputs are trained
using the numerical study to check the output values fall
inside an exact choice by using a machine learning
algorithm. Computers use sample data to automate
decision-making processes based on data inputs.
Users at present are benefitted from any technologies
from machine learning. A social media platform helps
users to tag and share photos using facial recognition
technology. Image of text can be converted to movable
type using optical character recognition (OCR)
technology.
The world is now growling about the COVID-19 pandemic
which is amusing life destruction and declared a global
health crisis. Severe acute respiratory syndrome virus
causes the COVID-19 pandemic and attacks human
epithelial cells. Proper precautions are taken to stop this
pandemic. Numerous scientists are making major hard
work to save manhood from this disaster.
The human respiratory tract of epithelial cells is attacked
by COVID-19, and X-Rays are used to detect the health of
the patient's lungs. Deep CNN has achieved surprising
growth in image recognition. Neural networks have been
productively used in identifying pneumonia from X-Rays,
and to achieve performances superior to those of
radiologists.
1.1 Objective
 Image classifiers are built based on the X-ray images
of patients who tested positive.
 Notice “normal” (i.e., uninfected) X-ray images from a
healthy patient.
 CNN is trained automatically to notice COVID-19 in X-
ray images through the dataset collected.
 The outcomes are evaluated from the trained model.
1.2 Problem Identification
The worldwide spreading of the COVID-19 deadly disease
has resulted in a considerable loss in manhood. Medicinal
and healthcare departments are in front of critical issues
because it is discovered promptly. Therefore, diagnosis of
suspected cases and facilitating to minimize the number
of infected inhabitants is of great importance. Due to its
low cost, extensive range of applications, and fast-speed
X-Ray inspection is measured to be the most common
method and plays a crucial role in the screening and
detection of COVID-19 patients. X-rays are used in the
detection of the patient’s lungs health because it attacks a
human respiratory cell.
International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 06 | June 2022 www.irjet.net p-ISSN: 2395-0072

1.3 Existing System
As COVID-19 examinations are hard to arise, they cannot
be manufactured quickly, which is creating a fright.
Because of this fright, there are the nefarious people who
are selling the fake COVID-19 test kits. Provided limited
COVID-19 test kits one must rely on other diagnosis
measures.
Analysis of COVID-19 disease is taking more time as
testing equipment is also expensive due to the speedy
spreading of the Coronavirus across the globe.
Appropriate measures are taken for automatic diagnosis
of this pandemic, and proper medicinal consideration and
action are taken for the patients as early as possible.
1.4 Proposed System
The complete system of COVID-19 for diagnosing can be
separated into two parts:
 Image collection to create the dataset: The dataset
here consists of patients’ x-ray images meant for both
training and testing purposes. In the training set, the
images of those who have COVID-19 pneumonia and
who do not suffer from COVID-19 pneumonia are
collected.
 Training and testing of data to create the model:
Convolutional neural network is used for training our
model. It is most frequently applied to studying
visual imagery in deep learning. Apiece x-ray image is
passed via convolutional and max-pooling layers
where the image data is compressed and passed
throughout the layers of the fully-connected layers
during training. The output class is predicted by the
neural network and formerly equated with the real
output is used to fine-tune the masses of the neural
network.
2. LITERATURE SURVEY
Mustafa Ghaderzadeh, Mehrad Aria et. al., proposed
artificial intelligence technique was used to decrease the
mortality rate of COVID-19 infections because it attacks
the epithelial cells of the respiratory region of a human
being. X-rays were used to check the health of human
lungs. [2]
Umashankar Subramaniam, M. Monica Subashini, et. al.,
suggested that the proposed work of the system
promised good health, and preprocessing algorithms
were introduced for preprocessing normal, COVID-19,
and pneumonia. [1]
Ebrahim Mohammed Senan, Ali Alzahrani, Mohammed Y.
Alzahrani, et. al., proposed that the symptoms were
moderate fever and it was spread through the saliva
droplets of infected patients. The Real-Time Reverse
Transcription–Polymerase Chain Reaction was used to
detect COVID-19 symptoms along with X-rays of chest
imaging and chest computed tomography. [3]
Zeming Fan, Mudasir Jamil, Muhammad Tariq Sadiq, et. al,
suggested that COVID-19 in patients can be identified
through chest x-ray scans. The planned method contains
three stages. In the first stage, works on preprocessing of
data after obtaining from open-source. Later in the
second stage, dissimilar training parameters were
defined. [4]
Fareed Ahmad, Amjad Farooq, and Muhammad Usman
Ghani suggested that the proposed model of a pathogen
can transportable up to twenty-seven feet and uses the
deep learning models, such as Alex Net, with five
convolution layers such as three fully connected layers,
and a SoftMax layer, comprised of approximately sixty
million parameters. [5]
Sharmila V J and JemiFlorinabel D suggested the use of
Long Short-Term Memory (LSTM) to forecast quickly
COVID-19 cases and aid an exact COVID-19 classification.
This study also proposed the use of the DCGAN-based
CNN model. [6]
Surya Krishnamurthy, Kathiravan Srinivasan, Saeed Mian
Qaisar, et. al., suggested that pneumonitis is an acute
illness in the lungs categorized as a result of soreness in
the alveoli, filled with pus, and unable to breathe.
Pneumonitis contaminations can be instigated by viral,
bacterial, and fungal agents where bacterial is the most
common and viral infection the most dangerous. [7]
Dongsheng Ji, Zhujun Zhang, et. al., suggested that X-ray
images of the chest were used to identify the infected
patients with COVID-19 and to train the DenseNet121
and ResNet152 series models. The experiment also
extracted output from convolutional layers of Xception,
ResNet152, DenseNet201, etc. [8]
Yousef Alhwaiti, Muhammad Hameed Siddiqi, Madallah
Alruwaili et. al., proposed that deep learning plays a key
role in the detection of COVID-19 and it is accurate. With
dissimilar radiology techniques COVID-19 showed its
robustness on X-ray, CT, and MRI datasets. [9]
Rahib H. Abiyev and Abdullahi Ismail proposed that
COVID-19 is a respiratory infection. According To WHO
Non-COVID-19, viral pneumonia also causes the death of
people, and chest X-rays are the best-obtained ways of
diagnosing pneumonia. [10]

3. SYSTEM ARCHITECTURE
Fig-1: System Architecture of Proposed Model
Pixels are the building block of images. Normally it is
considered as “color” or intensity of light. 0 to 255 is the
value in a grayscale image where 0 corresponds to black
and 255 is white respectively.
Open Computer vision stores the value of pixels in the
order of blue, green, and red, and blue, green, and red
order were popular among the camera manufacturers.
The process of maximizing or minimizing the size of
image deals with scaling and aspect ratios.
The convolution neural network algorithm classifies the
images collected into one of the following classes normal,
covid-19, and pneumonia respectively. The system
comprises an ensemble along with the preparation of
images and the operations of the neural network
combines with the convolution neural network features.
Fig-2: The structure of a convolutional network
Fig-3: Examples of the image feature detectors that a
CNN might “learn” during its training
A series of processing layers in CNN as shown in the Fig-
2 and each layer is a family of layers of convolution filters
each detector look like the Gabor-like and color blob
filters as shown in Fig-3. Each layer consists of hundreds
of thousands of filters, feeding the output result into the
next layer in the network provided.
4. MODULES
1. Gather Your Dataset
Building a component for deep learning involves
gathering the initial data set and labels are associated
with each image, and these labels are a finite set of
categories. Machine learning faces the common problem
of class imbalance.
2. Pre-Processing RGB and BGR ordering
Open Computer vision stores the color channels in
reverse order of blue, green, and red because the camera
manufacturers follow the blue, green, and red color
ordering.
3. Scaling and Aspect Ratio
The procedure of maximizing otherwise minimizing the
extent of the image deals with scaling and aspect ratios.
Maximizing and minimizing the size in the terms of width
and height.
4. Split Your Dataset
The initial datasets are split into two types:
 A training set
 A testing set
The classifier usages a training set to learn and predict
which group the image belongs to. If it is wrong then it
will correct itself.

Fig-4: Examples of common training and testing data
splits.
5. Train Your Network
Provided that a set of images for training, our network
aims to identify which category the image belongs to.
When it recognizes wrongly, it improves itself from
wrong detection of category.
6. Evaluate
The network is provided with each image in the testing
set and asked to predict correctly to which category the
image belongs and these predictions are compared with
ground truth labels and represent to which category the
image belongs.
5. DATA FLOW
Fig-5.1: Data Flow Diagram Level 0
In the above Fig-5.1, the file image is to be loaded to the
application where the loaded image is sent to the
classification unit to predict the result with the assistance
of a convolutional neural network model.
Fig-5.2: Data Flow Diagram Level 1
In the above Fig-5.2, the file image is to be loaded to the
application where the loaded image is sent to the
classification unit to predict the results to which class
label the classified image belongs, for instance usual,
COVID-19, viral pneumonia, and lung opaqueness.
6. USE CASE DIAGRAM
Fig- 6 Use Case Diagram
Fig-6 depicts the use case figure which represents the
active performance of the system and the system
functionality by using actors and use cases. In Unified
modeling language, modeling with use cases is a core
concept.

7. SEQUENCE DIAGRAM
The sequence diagram consists of 5 blocks namely user,
processor, memory, model, and labels respectively as
shown in the Figure. The user provides the input images
via already saved information that has been captured,
processing, and preprocessing of image data such as
resizing, reshaping, and storing the reminiscence unit.
As preprocessing and loading the image in the memory
unit has been done CNN model is trained, the feature of
the image is extracted for output classification. Next,
classification of the label is provided for instance usual,
COVID-19, viral pneumonia, and lung opaqueness.
Fig-7: Sequence diagram for the proposed system
8. CONCLUSION
Deep learning stood as a serious in the response to the
COVID-19 outburst. Analytic and diagnostic and classify
chest X-ray imagery into four categories Normal, COVID-
19, Viral Pneumonia, and opacity of lungs. Deep learning
provides an orientation method for the COVID-19
epidemic situation to medicinal and health institutions,
and supervision departments.
REFERENCES
[1] Umashankar Subramaniam, M. Monica Subashini,
Dhafer Almakhles, Alagar Karthick S. Manoharan. An
Expert System for COVID-19InfectionTracking in Lungs
Using Image Processing and Deep Learning Techniques.
[2021]
[2] Mustafa Ghaderzadeh, Mehrad Aria and
FarkhondehAsadi X-Ray Equipped with Artificial
Intelligence: Changing the COVID-19 Diagnostic Paradigm
during the Pandemic. [2021]
[3] Ebrahim Mohammed Senan, Ali Alzahrani,
Mohammed Y. Alzahrani, Nizar Alsharif, and Theyazn H.
H. Aldhyani Automated Diagnosis of Chest X-Ray for Early
Detection of COVID-19 Disease. [2021]
[4] Exploiting Multiple Optimizers with Transfer Learning
Techniques for the Identification of COVID-19 Patients.
Zeming Fan, Mudasir Jamil, Muhammad Tariq Sadiq,
Xiwei Huang, and Xiaojun Y. [2021]
[5] Deep Ensemble Model for Classification of Novel
Coronavirus in Chest X-Ray Images. Fareed Ahmad,
Amjad Farooq, and Muhammad Usman Ghani. [2021]
[6] Deep Learning Algorithm for COVID-19 Classification
Using Chest X-Ray Images. Sharmila V J and
JemiFlorinabel D. [2021]
[7] Evaluating Deep Neural Network Architectures with
Transfer Learning for Pneumonitis Diagnosis. Surya
Krishnamurthy, Kathiravan Srinivasan, Saeed Mian
Qaisar, P. M. Durai Raj Vincent and Chuan-Yu Chang.
[2021]
[8] Research on Classification of COVID-19 Chest X-Ray
Image Modal Feature Fusion Based on Deep Learning
Dongsheng Ji, Zhujun Zhang, Yanzhong Zhao, and
Qianchuan Zhao. [2021]
[9] Diagnosis of COVID-19 Using a Deep Learning Model
in Various Radiology Domains. Yousef Alhwaiti,
Muhammad Hameed Siddiqi, MadallahAlruwaili, Ibrahim
Alrashdi, 1SaadAlanazi, and Muhammad Hasan Jamal.
[2021]
[10] COVID-19 and Pneumonia Diagnosis in X-Ray Images
Using Convolutional Neural Networks Rahib H. Abiyev
and Abdullahi Ismail. [2021]

Predicting Covid-19 pneumonia Severity on Chest x-ray with deep learning

More Related Content

Similar to Predicting Covid-19 pneumonia Severity on Chest x-ray with deep learning (20)

More from IRJET Journal (20)

Recently uploaded (20)

Predicting Covid-19 pneumonia Severity on Chest x-ray with deep learning