VISVESVARAYA TECHNOLOGICAL UNIVERSITY

“Jnana Sangama”, Belagavi – 590018”

A Project Report
On

“CHEST X-RAY OF PNENMONIA DISEASE DIAGNOSIS

USING CNN”
By

MD NOORUZZAMAN 4MT17CS056
KARTHIK R 4MT17CS046
GANESH NAIK 4MT17CS040
FATHIMA ANISHA 4MT16CS37

Project Guide
Dr. SUKHWINDER SHARMA
Associate Professor
Department of CS & E
MITE, Moodabidri
February – May 2020

DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING

(Accredited by NBA)

MANGALORE INSTITUTE OF TECHNOLOGY & ENGINEERING

(An ISO 9001:2015 Certified Institution)
BADAGA MIJAR, MOODABIDRI
DK DIST-574225
MANGALORE INSTITUTE OF TECHNOLOGY & ENGINEERING
(An ISO 9001:2015 Certified Institution)
BADAGA MIJAR, MOODABIDRI, DK DIST – 574225

DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING

CERTIFICATE

This is to certify that the project work entitled “CHEST X-RAY OF PNEMONIA

DISEASE DIAGNOSIS BY USING CNN” is a bonafide work carried out by MD.

NOORUZZAMAN (4MT17CS056), KARTHIK R (4MT17CS046), GANESH NAIK

(4MT17CS040), FATHIMA ANISHA (4MT17CS037) in partial fulfillment for the award of

degree of Bachelor of Engineering in Computer Science & Engineering of the

Visvesvaraya Technological University, Belagavi during the year 2020 – 21. It is certified

that all corrections and suggestions indicated for Internal Assessment have been incorporated

in the report deposited in the departmental library. The project has been approved as it

satisfies the academic requirements in respect of project work prescribed for the Bachelor of

Engineering degree.

………….…………………… ………………………………. ………………………………..

Signature of the Guide Signature of the H.O.D Signature of the Principal
Dr. Sukhwinder Sharma Dr. Venkatramana Bhat P Dr. G. L. Easwara Prasad
Associate Professor Professor and Head Principal
Dept. of CS&E Dept. of CS&E MITE, Moodabidri
MITE, Moodabidri MITE, Moodabidri

External Viva: Name of the Examiners Signature with date

1.

2.
 ABSTRACT
As the most common examination tool in medical practice, chest radiography has important clinical
value in the diagnosis of disease. The automatic detection of chest disease such as pneumonia,
tuberculosis and lung disease based on chest x-ray images has become one of the hot topics in medical
research. We propose an automatic chest radiography image using deep learning. The main aim is to
develop a system that can detect disease from chest X-ray datasets (images) using CNN
(Convolutional neural network). We are using CNN VGG16 architecture. We will begin by importing
some key packages which we will be using throughout the project.

For an engineering student looking to graduate in Computer Science and Engineering, working with
these professional minds was definitely a great experience. This report is done with many limitations
and obstacles. Thanks to so many people who helped me to do this process. I hope that the authority is
pleased with my efforts and work done by me during the short period of time.

i
 ACKNOWLEDGEMENTS
The satisfaction and the successful completion of this project would be incomplete without
the mention of the people who made it possible, whose constant guidance encouragement
crowned our efforts with success.

This project is made under the guidance of Dr. Sukhwinder Sharma, Associate Professor,
in the Department of Computer Science and Engineering. We would like to express sincere
gratitude to our guide for all the helping hand and guidance in this project.

We would like to thank our project coordinators Dr. Sukhwinder Sharma, Associate
Professor in the Department of Computer Science and Engineering, for their cordial support,
valuable information and guidance, which helped us in completing this project through the
various stages.

We would like to express appreciation to Dr. Venkatramana Bhat P., Professor and Head of
the department, Computer Science and Engineering, for his support and guidance.

We would like to thank our Principal Dr. G.L. Easwara Prasad, for encouraging us and
giving us an opportunity to accomplish the project.

We also thank our management who helped us directly and indirectly in the completion of
this project.

Our special thanks to faculty members and others for their constant help and support.

Above all, we extend our sincere gratitude to our parents and friends for their constant
encouragement with moral support.

MD NOORUZZAMAN 4MT17CS056
KARTHIK R 4MT17CS046
GANESH NAIK 4MT17CS040
FATHIMA ANISHA 4MT17CS037

ii
 TABLE OF CONTENTS

Contents Page No
ABSTRACT i
ACKNOWLEDGEMENT ii
TABLE OF CONTENTS iii
LIST OF FIGURES v

Chapter no TITLE
1 INTRODUCTION 1

1.1 Introduction 1

1.2 Problem Statement 1

1.3 Objectives (Purpose of the project) 2

1.4 Scope of the Project 2

2 LITERATURE SURVEY 3

3 SYSTEM REQUIREMENT SPECIFICATION 6

3.1 Specific Requirements 6

3.1.1 Hardware Requirements 6

3.1.2 Software Requirements 6

3.1.3 Functional Requirements 7

3.1.4 Non-Functional Requirements 7

4 SYSTEM DESIGN 8

4.1 CNN Architecture Diagram 8

4.2 Activity Diagram 9

4.3 System Architecture Diagram 10

5 IMPLEMENTATION 11

5.1 Backend 11

5.1.1 Importing Libraries 12

iii
 5.1.2 Uploading Datasets from Google Drive 12

5.1.3 Loading Datasets 12

5.1.4 Data Augmentation 14

5.1.5 Training Using VGG16 Model 14

5.1.6 Early Stopping 17

5.1.7 Model Competition 17

5.1.8 Plotting 19

5.1.9 Model Evaluation 20

5.1.10 Solving the Model 20

5.2 Frontend 20

5.2.1 HTML5 21

5.2.2 CSS3 21

5.2.3 Flask 21

5.2.4 MongoDB Database 24

5.2.5 html2pdf.js 24

6 RESULTS AND SNAPSHOTS 25

6.1 Result 25

6.2 Snapshots 25

6.2.1 Home Page 25

6.2.2 User form 26

6.2.3 Upload Chest X-Ray Image 27

6.2.4 Prediction Output and Report 28

6.2.5 Report in PDF Format 29

7 CONCLUSION AND FUTURE SCOPE 30

REFERENCES

iv
 LIST OF FIGURES
Fig No. FIGURE NAME Page No.

Fig 4.1.1 VGG16 Model 8

Fig 4.2.1 Activity Diagram 9

Fig 4.3.1 Block Diagram of System 10

Fig 5.1.5.1 Graph of ReLu Function 16

Fig 5.1.5.2 Graph of Sigmoid Function 17

Fig 6.2.1 Home Page 25

Fig 6.2.2 User Form 26

Fig 6.2.3 Upload Chest X-Ray Image 27

Fig 6.2.4 Prediction and Report 28

Fig 6.2.5 Report in PDF format 29

v
 CHAPTER 1
INTRODUCTION
 Chest X-ray of Pneumonia disease diagnosis using CNN

CHAPTER 1

INTRODUCTION
This chapter gives an introduction to CNN (Convolutional Neural Networks).

1.1 Introduction
A Convolutional neural network (CNN) is a neural network that has one or more
convolutional layers and are used mainly for image processing, classification, segmentation
and also for other auto correlated data. CNNs are regularized versions of multilayer
perceptron. Multilayer perceptrons usually mean fully connected networks, that is, each
neuron in one layer is connected to all neurons in the next layer The "full connectivity" of
these networks makes them prone to overfitting data. Convolutional networks were inspired
by biological processes in that the connectivity pattern between neurons resembles the
organization of the animal visual cortex. Individual cortical responses to stimuli only in a
restricted region of the visual field known as the receptive field. The receptive fields of
different neurons partially overlap such that they cover the entire visual field.

1.2 Problem Statement

The risk of disease is immense for many, especially in developing nations where billions face
problems. The WHO estimates that over 4 million premature deaths occur annually from
household air pollution-related diseases including pneumonia, tuberculosis. Over 150 are
infected with pneumonia on an annual basis, especially children under 5 years old. In such
regions, the problem can be further aggravated due to the lack of medical resources. For
example, in Africa’s 57 nations, a gap of 2.3 million doctors and nurses exists. For these
populations, accurate and fast diagnosis means everything. It can guarantee timely access to
treatment and save much needed time and money for those already experiencing poverty.
Despite several diseases. There are also advantages of X-ray image, still in some cases it is
not possible to identify the correct region of interest in radiographic image for detecting
observations as the diagnostic accuracy of automated methods reach the human level.

Dept. Of CS&E, MITE, Moodabidri 1

 Chest X-ray of Pneumonia disease diagnosis using CNN

1.3 Objectives

According to UNICEF a child dies of pneumonia every 39 seconds. Pneumonia kills more
children than any other infectious disease, claiming the lives of over 800,000 children under
five every year, or around 2,200 every day. This includes over 153,000 newborns. Almost all
of these deaths are preventable. Globally, there are over 1,400 cases of pneumonia per
100,000 children, or 1 case per 71 children every year, with the greatest incidence occurring
in South Asia (2,500 cases per100,000children. Above data from here
https://data.unicef.org/topic/child-health/pneumonia/.

1.4 Scope of the Project

The scope of this project is to train a VGG16 model that is useful to predict the normal and
abnormal image of Chest X-ray of pneumonia images. Deep Learning (DL) has a big scope in
future and the biggest reason for it is that DL doesn’t require any kind of feature engineering.
Deep Learning extracts the features from the data itself instead of us giving it the features
after extracting it from the data. This way it solves our biggest problem of feature engineering.
Also, since features are learned by the model itself, it has a better probability of producing a
model which is more generalized than the feature engineered models.

● Automated detection method provides an add on support to clinical experts.

● It is highly used in the field of computer vision.
● Today not everyone has access to good health care service. So, it will reduce time and
money.
● It is also observed that as the diagnostic accuracy of automated methods reach the
human level.

Dept. Of CS&E, MITE, Moodabidri 2

 CHAPTER 2
LITERATURE SURVEY
 Chest X-ray of Pneumonia disease diagnosis using CNN

CHAPTER 2
LITERATURE SURVEY
A literature survey or a literature review in a project report is that section which shows the
various analysis and research made in the field of interest and the results already published,
taking into account the various parameters and the extent of the project.

Diagnosis of Pneumonia from Chest X-Ray Images using Deep Learning.

By Enes AYAN * and Halil Murat ÜNVER.

In this study, they compared two CNN network's performance on the diagnosis of pneumonia
disease. While training, they used transfer learning and fine-tuning. After the training phase,
they compared two network test results. The test results showed that Vgg16 network
outperforms Xception network by accuracy 0.87%, pneumonia precision 0.91% and pneumonia
f1 score 0.90%. Whereas Xception network outperforms Vgg16 network by sensitivity 0.85%,
normal precision 0.86% and pneumonia recall 0.94%. Xception network is more successful for
detecting pneumonia cases than the Vgg16 network. At the same time Vgg16 network is more
successful at detecting normal cases.

Pneumonia Detection using CNN through Chest x-ray.

By Harshvardhan GM, Mahendra Kumar Gourisaria, Sidharth Swarup Rautaray* and
Manjusha Pandey.

In this paper, they attempt to find a simpler approach for pneumonia detection based on Chest
X-ray by comparing the performances of 15 different CNN architectures trained on the same
dataset. Based on their findings, they select the most ideal model which is easy to train and has
one of the best performance metrics. The metrics of the selected architecture compared to some
of the state-of-the-art architectures trained on Chest X-ray that goes ahead to prove that striving
for the simplification of CNN architectures is crucial for intelligibility without compromising
accuracy and quality of performance.

Dept. Of CS&E, MITE, Moodabidri 3

 Chest X-ray of Pneumonia disease diagnosis using CNN

CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays With Deep Learning.

By Pranav Rajpurkar *, Jeremy Irvin *, Kaylie Zhu, Brandon Yang, Hershel Mehta, Tony
Duan, Daisy Ding, Aarti Bagul, Robyn L. Ball, Curtis Langlotz, Katie Shpanskaya, Matthew P.
Lungren and Andrew Y. Ng.

They developed an algorithm which detects pneumonia from frontal-view chest X-ray images at
a level exceeding practicing radiologist. They also show that a simple extension of an algorithm
to detect multiple diseases outperforms previous state of the art on ChestX-ray14, the largest
publicly available chest X-ray dataset. With automation at the level of experts, They have
clearly mentioned that this technology can improve healthcare delivery and increase access to
medical imaging expertise in parts of the world where access to skilled radiologists is limited.

Viral Pneumonia Screening on Chest X-rays Using Confidence Aware Anomaly Detection.
By Jianpeng Zhang, Yutong Xie, Guansong Pang, Zhibin Liao, Johan Verjans, Wenxing Li,
Zongji Sun, Jian He, Yi Li, Chunhua Shen, and Yong Xia.

In this paper, they have proposed the CAAD model for viral pneumonia screening. Their results
on two chest X-ray datasets indicate that (1) anomaly detection works well in term of viral
pneumonia screening on chest X-ray images and is superior to binary classification methods,
and (2) learning model confidence is useful to predict failures, greatly reducing the false
negatives, and (3) achieves an AUC of 83.61% and sensitivity of 71.70% on the unseen dataset,
which is comparable to the performance of medical professionals.

Pneumonia Detection Using CNN based Feature Extraction.

By Dimpy Varshni , Kartik Thakral , Lucky Agarwal , Rahul Nijhawan and Ankush Mittal.

This paper primarily aims to improve the medical adeptness in areas where the availability of
radiotherapists is still limited. It has been observed that the performance of various pretrained
CNN models along with distinct classifiers and then on the basis of statistical results selected
DenseNet-169 for the feature extraction stage and SVM for the classification stage. They have
mentioned that performing hyperparameter optimization in the classification stage ameliorated
the model performance.

Dept. Of CS&E, MITE, Moodabidri 4

 Chest X-ray of Pneumonia disease diagnosis using CNN

Deep Learning Approach for Prediction of Pneumonia.

By Kalyani Kadam, Dr. Swati Ahirrao, Harbir Kaur, Dr. Shraddha Phansalkar, Dr. Ambika
Pawar.

This paper presents that using CNN- ResNet34 architecture which is after training the model for
500 epochs, test time augmentation was done, and the model was tested for performance on the
test data. The model accomplished an accuracy of 92.9%, considerably superior than the
baseline accuracy of 63%. The precision and recall values were approximately around 0.9088
and 0.9927 respectively.

Pneumonia Detection Using Convolutional Neural Networks (CNNs)

By V. Sirish Kaushik, Anand Nayyar, Gaurav Kataria and Rachna Jain.

In this paper it has been represented that the validation accuracy, recall and F1 score of CNN
classifier model 3 with three convolutional layers are 92.31%, 98% and 94%, respectively,
which are quite high compared to other models that were trained. CNN classifier model 4
with four convolutional layers also comes very close in performance with 91.67% validation.
Accuracy, 98% recall and 94% F1 score. Both of these models have the same recall and F1
scores.

Dept. Of CS&E, MITE, Moodabidri 5

 CHAPTER 3
SOFTWARE REQUIREMENTS
SPECIFICATION
 Chest X-ray of Pneumonia disease diagnosis using CNN

CHAPTER 3

SYSTEM REQUIREMENTS SPECIFICATION

A software requirements specification (SRS) is a description of a software system to be
developed. The software requirements specification lays out functional and non-functional
requirements, and it may include a set of use cases that describe user interactions that the
software must provide. Use cases are also known as functional requirements. In addition to
use cases the SRS also contains non-functional requirements. Non-functional requirements
are requirements which impose constraints on the design or implementation. For the
hardware requirements the SRS specifies the logical characteristics of each interface between
the software product and the hardware components. It specifies the hardware requirements
like memory restrictions, cache size, the processor, RAM size etc. those are required for the
software to run. Software requirements specification is a rigorous assessment of requirements
before the more specific system design stages, and its goal is to reduce later redesign.

3.1 Specific Requirements

This section includes the detailed description about the hardware requirements, software
requirements. Functional requirements and non-functional requirements. The Hardware and
software requirements are:

3.1.1 Hardware Requirements

● Processor: Intel i3-370M processor with 2.4 GHz, 3 MB L3 Cache.

● RAM: 8 GB RAM
● Disk: 500 MB of Hard disk spaces

3.1.2 Software Requirements

● Operating system: Windows 10 and Ubuntu 18.04.

● IDE: Google Collab and VS Code.

Dept. Of CS&E, MITE, Moodabidri 6

 Chest X-ray of Pneumonia disease diagnosis using CNN

3.1.3 Functional Requirements

• The system must be able to interact with user input like mouse, keyboard and menu.
• The system must allow an image format to be taken as an input.

3.1.4 Non-functional requirements

• Reliability: Must make sure that the system is reliable in its operation and during training the
model.
• Safety requirement: The algorithm must not cause any data loss to the user.
• Security requirements: The algorithm shall not behave abruptly or corrupt the project file on
receiving the wrong input from the user.
• Performance requirement: The algorithm reacts to the input quickly as specified.

Dept. Of CS&E, MITE, Moodabidri 7

 CHAPTER 4
SYSTEM DESIGN
 Chest X-ray of Pneumonia disease diagnosis using CNN

CHAPTER 4
SYSTEM DESIGN

System overview provides a top-level view of the entire software product. It highlights the
major components without taking account the inner details of the implementation. It describes
functionality of the product and context and design of the software product.The application will
be developed in a way which allows the user to interact with the system and simplifies the tasks by
providing a smooth user interface and user experience with an easily readable and understandable
view.

4.1 CNN Architecture Diagram

We have used VGG16. It is a Convolutional Neural Networks (CNN) archi- tecture . Which was
used to win ILSVR(ImageNet) Competition in 2014. It makes an improvement over AlexNet by
replacing large kernel-sized filters with multiple 3×3 kernel-sized filterswith stride 1 and always.
using the same padding and maxpool layer of 2×2 filters of stride 2.

Fig 4.1.1 VGG16 model

Dept. Of CS&E, MITE, Moodabidri 8

 Chest X-ray of Pneumonia disease diagnosis using CNN

The input of VGG16 is 224×224×3 pixels images , then we two convolutional layers with
each 224×224×64 size then we have pooling layer which reduces height and width of image
112×112×64 .In both layers 64 filters , same padding and non-linearity activation
function(ReLu) was used .Then we have two conv128 layers with each 112×112×128 size
after that we have a pooling layer which is again reducing the height and weight of the image
of 56×56×128 . Here we have used 128 filters with the same padding and non-linearity
activation function (ReLu). Then we have three conv256 layers with each 56×56×256 size,
after that again a pooling layer reducing the image size 28×28×256. Then we have three
conv512 layers with each 28×28×512 size, after that again pooling layer reducing the size
of the image of 14×14×512.Then again we have three conv256 layers with 14×14×512 size ,
after that we have pooling layer of 7×7×512 and then we have two dense or fully-connected
layers with each 4096 nodes and at the last we have a final dense or output layer with 1000
nodes of size which classify between 1000 class .This model processes the input image and
outputs a vector of 1000 values. This vector represents the classification probability for the
corresponding class.

4.2 Activity Diagram

Activity Diagram shows the sequence of steps that make up a complex process. It shows the
flow of control. The steps that are used in this are as follows.
● First loading datasets (i.e., pneumonia x-ray image).
● Training the model using the VGG16 model.
● RMSprop optimization algorithm was used.
● Then prediction using a trained model.

Fig 4.2.1 Activity diagram

Dept. Of CS&E, MITE, Moodabidri 9

 Chest X-ray of Pneumonia disease diagnosis using CNN

4.3 System Architecture Diagram

A System architecture diagram is a diagram of a system that is used to abstract the overall
outline of the software system and the relationships, constraints, and boundaries between
components. It is an important tool as it provides an overall view of the physical deployment of
the software system and its evolution roadmap.
So, in our System architecture diagram clinical experts (i.e. Radiologists) use web applications
to predict chest X-ray images are normal or abnormal. So, to do this radiologist has to upload an
image through a web application and then predict the output using a trained model. After
getting the output in a binary form (i.e normal or abnormal). Then the system will generate the
report in the form of a PDF.

Fig 4.3.1 Block Diagram of System architecture

Dept. Of CS&E, MITE, Moodabidri 10

 CHAPTER 5
IMPLEMENTATION
 Chest X-ray of Pneumonia disease diagnosis using CNN

CHAPTER 5
IMPLEMENTATION
5.1 BACKEND
We did the implementation using Google Colaboratory editor. Google Colaboratory (also
known as Collab) is a free Jupiter notebook environment that runs in the cloud and stores its
notebooks on Google Drive. Collab was originally an internal Google project; an attempt was
made to open source all the code and work more directly upstream, leading to the
development of the "Open in Collab”. We used Python Programming language and libraries.
Libraries we used are as follows.

● Keras: Keras is an open-source software library that provides a Python interface for
artificial neural networks. Keras acts as an interface for the TensorFlow library.
● NumPy: NumPy is a library for the Python programming language, adding support for
large, multi-dimensional arrays and matrices, along with a large collection of high-level
mathematical functions to operate on these arrays.
● Matplotlib: Matplotlib is a plotting library for the Python programming language and
its numerical mathematics extension NumPy. It provides an object-oriented API for
embedding plots into applications.

There were several steps we used to implement and train the model. The steps are as follows.

1. Importing libraries
2. Uploading datasets from Google Drive
3. Loading datasets
4. Data augmentation
5. Training the using VGG16 model
6. Early Stopping
7. Model compilation
8. Plotting
9. Model evaluation
10. Saving the trained model

Dept. Of CS&E, MITE, Moodabidri 11

 Chest X-ray of Pneumonia disease diagnosis using CNN

5.1.1 Importing Libraries

The first step of implementation is to import the libraries. We used libraries like keras, NumPy
and matplotlib. Import in python is similar to #include header file in C/C++. Python modules
can get access to code from another module by importing the file/function using import.

5.1.2 Uploading datasets from Google Drive

We uploaded datasets from Google Drive because we were using Collab and our machine was
not compatible for training the model. There are some steps we followed.

1. To connect Google Drive (GDrive) with Collab, executed the following two lines of code in
Collab.

2. Got the authorization code after signing in with Google account then we had to paste that
code then we got permission to access to Google Drive.

5.1.3 Loading Datasets

We have used X-ray image datasets which are publicly available on Kaggle. Find it here
https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia. It has a total of 5856
images (1.15 GB) which has 5216 training images, 624 test images and 16 validation images.
Overall, we have 4273 pneumonia images (73% of total datasets) and 1583 normal images
(27% of total datasets). We used the below code to access Google Drive.

Dept. Of CS&E, MITE, Moodabidri 12

 Chest X-ray of Pneumonia disease diagnosis using CNN

The whole datasets distribution is given below.

Data samples

Dept. Of CS&E, MITE, Moodabidri 13

 Chest X-ray of Pneumonia disease diagnosis using CNN

5.1.4 Data Augmentation

Data augmentation is a technique to artificially create new training data from existing training
data. So we used ImageDataGenerator (provided by keras deep learning library) to augment our
datasets and in this we used zoom scale = 0.3 and image flips via vertical flips arguments. We
used below code for image data augmentation.

5.1.5 Training using VGG16 model

VGG16 architecture has been used for training this model. It has been explained above in the
system design chapter. We trained our model using Keras.

For the VGG16 model we used keras sequential model. A Sequential model is appropriate for a
plain stack of layers. We created a Sequential model by passing a list of layers to the
Sequential constructor using model.add .

Dept. Of CS&E, MITE, Moodabidri 14

 Chest X-ray of Pneumonia disease diagnosis using CNN

Conv2D is a 2D convolutional layer This layer creates a convolution kernel that is convolved
with the layer input to produce a tensor of outputs. If use bias is True, a bias vector is created
and added to the outputs. Finally, if activation is not None, it is applied to the outputs as well.
When using this layer as the first layer in a model, provide the keyword argument input shape
(tuple of integers or None, does not include the sample axis), e.g., input shape= (150, 150, 3)
for 128x128 RGB pictures in data format="channels last". In Conv2D arguments we also used
fitters, padding (e.g., padding = same), activation function (e.g., Relu) and kernal_size (e.g.,
kernel _size (3,3)).

MaxPool2D is a max pooling for 2D data Down samples the input along its spatial dimensions
(height and width) by taking the maximum value over an input window (of size defined by pool
size) for each channel of the input. The window is shifted by strides along each dimension. The
resulting output, when using the "valid" padding option, has a spatial shape (number of rows or
columns) of output shape = math. Floor ((input shape - pool size) / strides) + 1 (when input
shape >= pool size) The resulting output shape when using the "same" padding option is: output
shape = math. Floor ((input shape - 1) / strides) + 1. We used pool size = (2,2) and stride = (2
,2) in the MaxPool2D arguments.

We used a Flatten layer for flattening the input If inputs are shaped (batch,) without a feature
axis, then flattening adds an extra channel dimension and output shape is (batch, 1).

We used the Dense layer in the VGG16 model for a fully connected layer. It is just regular
Densely connected Neural Networks layers. We used two fully connected layers and one output
layer. Dense implements the operation: output = activation (dot (input, kernel) + bias) where
activation (i.e., Sigmoid) is the element-wise activation function passed as the activation
argument, kernel is a weights matrix created by the layer, and bias is a bias vector created by
the layer.

We used two activation functions, one for convolutional layers (i.e., Relu) and other for fully
connected layers (i.e., sigmoid).

Dept. Of CS&E, MITE, Moodabidri 15

 Chest X-ray of Pneumonia disease diagnosis using CNN
ReLu : The ReLU function is a non-linear activation function that has gained popularity in the
deep learning domain. ReLU stands for Rectified Linear Unit. The main advantage of using the
ReLU function over other activation functions is that it does not activate all the neurons at the
same time.This means that the neurons will only be deactivated if the output of the linear
transformation is less than 0 . It is used in the hidden conv layers of VGG16. For the negative
input values, the result is zero, that means the neuron does not get activated.

Fig 5.1.5.1 Graph of ReLu function

Sigmoid: Sigmoid is also a non-linear activation function. It transforms the value between 0 to
1. Here is the mathematical expression.

Sigmoid is non-linear this means that when I have multiple neurons having sigmoid function as
their activation function, the output is non-linear as well.

Dept. Of CS&E, MITE, Moodabidri 16

 Chest X-ray of Pneumonia disease diagnosis using CNN

Fig 5.1.5.2 Graph of sigmoid function

5.1.6 Early Stopping

We used keras callback API to stop the training when there is no improvement of the validation
set. Early Stopping is a form of regularization used to avoid overfitting and stop early without
going to all epochs.

5.1.7 Model Compilation

Keras model provides a method, compile () to compile the model. The argument and default
value of the compile () method is as follows.

We used three arguments: Optimizer which is RMSProp, loss function and metrics which is
accuracy.

Dept. Of CS&E, MITE, Moodabidri 17

 Chest X-ray of Pneumonia disease diagnosis using CNN

RMSprop: We used the RMSprop optimization algorithm. The RMSprop optimizer is similar
to the gradient-descent algorithm with momentum. It restricts the oscillation in the vertical
direction. Therefore, we can increase our learning rate and our algorithm could take larger steps
in the horizontal direction and converge faster.
RMSprop Algorithm:
Input: A vector 𝑅 𝑑 ∋ 1𝑑 ≥ 0 , parameters 𝛽 ∈ [0,1] ,step size 𝛼, initial
starting point 𝑥1 ∈ 𝑅 𝑑 and 𝑓: 𝑅 𝑑 → 𝑅
Function RMSprop
Initialize: 𝒗𝟎 = 𝟎
for 𝒕 = 𝟏, 𝟐, 𝟑. . . . ..do
𝒈𝒕 = 𝒊𝒏𝒗𝒆𝒓𝒕𝒆𝒅 − 𝒅𝒆𝒍𝒕𝒂 𝒇(𝒙𝒕 )
𝒗𝒕 = 𝜷 𝒗𝒕−𝟏 + (𝟏 − 𝜷)(𝒈𝒕 𝟐 + 𝝃𝟏𝒅 )
𝑽𝒕 = 𝒅𝒊𝒂𝒈(𝒗𝒕 )
𝟏
−
𝒙𝒕+𝟏 = 𝒙𝒕 − 𝜶𝑽𝒕 𝟐 𝒈𝒕
end for
end function

RMSprop’s update rule

𝒗𝒅𝒘 = 𝜷. 𝒗𝒅𝒘 + (𝟏 − 𝜷). 𝒅𝒘𝟐
𝒗𝒅𝒃 = 𝜷. 𝒗𝒅𝒘 + (𝟏 − 𝜷). 𝒅𝒃𝟐
𝒅𝒘
𝒘 = 𝒘 − 𝜶.
√𝒗𝒅𝒘 + 𝝐
𝒅𝒃
𝒃 = 𝒃 − 𝜶.
√𝒗𝒅𝒃 + 𝝐

RMSprop uses adaptive learning rate instead of treating learning rate as a hyperparameter. This
means that the learning rate changes over time. The value of momentum and learning rate is
denoted by 𝛽 and 𝛼respectivily. Sometime the value of 𝑣𝑑𝑤 ≈ 0 and included parameter epsilon
which is set small to prevent the gradient from exploding and vanishing.

Loss function: Our model finds whether the patient is suffering from pneumonia or not. That
means we are finding “yes” or “No” (binary classification problem). So, we used a Binary
classification loss function i.e., Binary Cross-Entropy.

So here 𝑦 is the label and 𝑝(𝑦) is the predicted probability.

Dept. Of CS&E, MITE, Moodabidri 18

 Chest X-ray of Pneumonia disease diagnosis using CNN

5.1.8 Plotting

We used Matplotlib for plotting the graph two graphs were plotted, one was for loss and
accuracy. Below code and graph shows the loss of training and validation.

Below code and graph shows accuracy of training and validation.

Dept. Of CS&E, MITE, Moodabidri 19

 Chest X-ray of Pneumonia disease diagnosis using CNN

5.1.9 Model Evaluation

Keras provides Model.evaluate() function to develop the model to check whether the model is
best fit for the given problem and corresponding data.

After evaluation we got loss = 0.1583 and accuracy = 0.9375 on validation and on the other side
we got loss = 0.12 and accuracy = 0.9528 on training.

5.1.10 Saving the Model

We saved the model using the HDF5 library. To use this, we installed the h5py package. The
h5py package is a Pythonic interface to the HDF5 binary data format.

Our model was saved in Google Drive using mounting.

5.2 FRONTEND

We developed one web application where a Radiologist or clinical expert can easily upload the
image and be able to predict whether the Pneumonia Chest x - ray image is normal or abnormal.
After getting the output we also provided that report will be generated in PDF format.
Radiologists are easily able to download the report. In the frontend VS Code editor was used to
create web applications. There were five technologies we used in the frontend.

Dept. Of CS&E, MITE, Moodabidri 20

 Chest X-ray of Pneumonia disease diagnosis using CNN
1. HTML5
2. CSS3
3. Flask
4. MongoDB database
5. html2pdf.js

5.2.1 HTML5

We used HTML5 for structuring the web application. HTML5 is a markup language used for
structuring and presenting content on the World Wide Web. It is the fifth and last major HTML
version that is a World Wide Web Consortium recommendation. The current specification is
known as the HTML Living Standard.

5.2.2 CSS3

For styling we had used CSS3 in the web application. Cascading Style Sheets is a style sheet
language used for describing the presentation of a document written in a markup language such
as HTML. CSS is a cornerstone technology of the World Wide Web, alongside HTML and
JavaScript.

5.2.3 Flask

We also used Flask to make the site more interactive where we used our saved model in HDF5
format. Flask is a micro web framework written in Python. It is classified as a microframework
because it does not require particular tools or libraries. It has no database abstraction layer, form
validation, or any other components where pre-existing third-party libraries provide common
functions.

As we can see above, we imported flask and pymongo libraries. We used Flask for machine
learning connection with model files and loading the model using load model () . Our trained
model has been saved in my_model.h5.

Dept. Of CS&E, MITE, Moodabidri 21

 Chest X-ray of Pneumonia disease diagnosis using CNN

App routing was used in code to route the specific URL .

Dept. Of CS&E, MITE, Moodabidri 22

 Chest X-ray of Pneumonia disease diagnosis using CNN
We used the first route (‘/’) for the home page when returning rendered template medical.html.
In the second route (‘/user form/’) we accessed user form where users can give user
information like first name, last name, place, age, dob, weight, temperature, gender, email and
phone number. These data will be posted after clicking the submit button.

The code which is given below you see we used a third route (‘/upload’) to upload the image
and save using file path. We used model. Predict () to predict whether the given image is
pneumonia or normal (value of prediction will be in between 0 and 1) and storing in classes
variables. Then we applied a conditional if statement. If the prediction value is greater than 0.5
then the given image has pneumonia, otherwise normal.

Dept. Of CS&E, MITE, Moodabidri 23

 Chest X-ray of Pneumonia disease diagnosis using CNN

5.2.4 MongoDB Database

We are storing our patient information in the MongoDB database. MongoDB is a source-
available cross-platform document-oriented database program. Classified as a NoSQL database
program, MongoDB uses JSON-like documents with optional schemas. We are using pymongo
libraries which is a Python distribution containing tools for working with MongoDB and is the
recommended way to work with MongoDB from Python. See the below code where we used
pymongo. Mongo Client to make a connection with the installed MongoDB database.

5.2.5 html2pdf.js

We provided a feature where radiologists can easily generate PDF of Patient reports and are
also able to download the PDF. We achieved this using a JavaScript framework called
html2pdf.js. html2pdf.js converts any webpage or element into a printable PDF entirely client-
side using html2canvas and jsPDF.

We used the src link of the html2pdf framework (i.e., html2pdf.bundle.js). In pdf () we used
DOM to generate and save the PDF.

Dept. Of CS&E, MITE, Moodabidri 24

 CHAPTER 6
RESULTS AND
SNAPSHOTS
 Chest X-ray of Pneumonia disease diagnosis using CNN

CHAPTER 6
RESULTS AND SNAPSHOTS

6.1 RESULT

6.2 SNAPSHOTS

6.2.1 Home page

Fig 6.2.1 Home page

Dept. Of CS&E, MITE, Moodabidri 25

 Chest X-ray of Pneumonia disease diagnosis using CNN

6.2.2 User form

Fig 6.2.2 User form

Dept. Of CS&E, MITE, Moodabidri 26

 Chest X-ray of Pneumonia disease diagnosis using CNN

6.2.3 Upload Chest X-ray image

Fig 6.2.3 Upload Chest X-ray image

Dept. Of CS&E, MITE, Moodabidri 27

 Chest X-ray of Pneumonia disease diagnosis using CNN

6.2.4 Prediction Output and Report

Fig 6.2.4 Prediction and report

Dept. Of CS&E, MITE, Moodabidri 28

 Chest X-ray of Pneumonia disease diagnosis using CNN

6.2.5 Report in PDF Format

Fig 6.2.5 Report in PDF format

Dept. Of CS&E, MITE, Moodabidri 29

 CHAPTER 7
CONCLUSION AND FUTURE
SCOPE
 Low Power Cryptography Solution Based on Chaotic and DNA Replication

CHAPTER 7

CONCLUSION AND FUTURE SCOPE

Pneumonia constitutes a significant cause of morbidity and mortality. The WHO estimates
that over 4 million premature deaths occur annually from household air pollution-related
diseases including pneumonia, tuberculosis. Over 150 are infected with pneumonia on an
annual basis, especially children under 5 years old. In such regions, the problem can be
further aggravated due to the lack of medical resources. For example, in Africa’s 57 nations,
a gap of 2.3 million doctors and nurses exists. According to UNICEF a child dies of
pneumonia every 39 seconds. Pneumonia kills more children than any other infectious
disease, claiming the lives of over 800,000 children under five every year, or around 2,200
every day. So, this is really a big problem where technology can help to improve medical
facilities.
We got 93% validation accuracy and 95% training accuracy with almost negligible error. In
the future we will improve our model to predict better results. With automation at the level
of experts, this technology can improve healthcare delivery and increase access to medical
imaging expertise in parts of the world where access to skilled radiologists is limited.

Dept. Of CS&E, MITE, Moodabidri 30

 REFERENCES

[1] Diagnosis of Pneumonia from Chest X-Ray Images using Deep Learning By Enes
AYAN * and Halil Murat ÜNVER ,978-1-7281-1013-4/19/$31.00 ©2019 IEEE.

[2] Pneumonia Detection using CNN through Chest x-ray By Harshvardhan GM, Mahendra
Kumar Gourisaria, Sidharth Swarup Rautaray* and Manjusha Pandey,Vol. 16, No. 1
(2021) 861 - 876.

[3] CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays With Deep

Learning By Pranav Rajpurkar *, Jeremy Irvin *, Kaylie Zhu , Brandon Yang , Hershel
Mehta ,Tony Duan , Daisy Ding , Aarti Bagul , Robyn L. Ball , Curtis Langlotz , Katie
Shpanskaya ,Matthew P. Lungren and Andrew Y. Ng , arXiv : 1711.05225v3 [cs.CV]
25 Dec 2017.

[4] Viral Pneumonia Screening on Chest X-rays Using Confidence Aware Anomaly
Detection By Jianpeng Zhang, Yutong Xie, Guansong Pang, Zhibin Liao, Johan Verjans,
Wenxing Li, Zongji Sun, Jian He, Yi Li, Chunhua Shen, and Yong Xia,
arXiv:2003.12338v4 [eess.4] 2Dec 2020.

[5] Pneumonia Detection Using CNN based Feature Extraction By Dimpy Varshni, Kartik
Kartik Thakral, Lucky Agarwal , Rahul Nijhawan and Ankush Mittal , May 24,2021 at
10:46:14 UTC from IEEE.

[6] Transfer Learning with Deep Convolutional Neural Network (CNN) for Pneumonia
Detection Using Chest X-ray. By Tawsifur Rahman 1 , Muhammad E. H Chowdhury 2,
* , Amith Khandakar 2 ,Khandaker R. Islam 3 , Khandaker F. Islam 2 , Zaid B. Mahbub,
Muhammad A. Kadir and Saad Kashem 5. Appl. Sci. 2020, 10, 3233;
doi:10.3390/app10093233.

[7] Deep Learning Approach For Prediction Of Pneumonia By Kalyani Kadam, Dr.Swati
Ahirrao, Harbir Kaur ,Dr. Shraddha Phansalkar , Dr. Ambika Pawar , ISSUE 10,
OCTOBER 2019 ISSN 2277-8616.
[8] Efficient Pneumonia Detection in Chest Xray Images Using Deep Transfer Learning By
Mohammad Farukh Hashmi 1, † , Satyarth Katiyar 2,† , Avinash G Keskar 3,† ,Neeraj
Dhanraj Bokde 4,† and Zong Woo Geem 5. Diagnostics 2020, 10, 417;
doi:10.3390/diagnostics10060417.

[9] Pneumonia Detection Using Convolutional Neural Networks (CNNs) By V. Sirish

Kaushik, Anand Nayyar, Gaurav Kataria and Rachna Jain. https://doi.org/10.1007/978-
981-15-3369-3_36.

[10] Convergence guarantees for RMSProp and ADAM in non-convex optimization and an
empirical comparison to Nesterov acceleration By Soham De 1 , Anirbit Mukherjee 2 ,
and Enayat Ullah 3, arXiv:1807.06766v3 [cs.LG] 20 Nov 2018.

[11] Training of Deep Neural Networks based on Distance Measures using RMSProp By
Thomas Kurbiel and Shahrzad Khaleghian , arXiv:1708.01911v1 [cs.LG] 6 Aug 2017.

[12] Variants of RMSProp and Adagrad with Logarithmic Regret Bounds By Mahesh
Chandra Mukkamala 1 2 Matthias Hein 1, Proceedings of the 34 th International
Conference on Machine Learning, Sydney, Australia, PMLR 70, 2017.

[13] Data presented by UNICEF ,A child dies of pneumonia every 39 seconds

https://data.unicef.org/topic/child-health/pneumonia/