Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Project Synopsis

Title: Pneumonia Detection on chest X-ray images using Hybrid


CNN Model

Domain: Deep Learning

Contents Page No
1. Introduction 2

2. Literature Survey 3

3. Problem Formulation 5

4. Objectives 6

5. Methodology 7

6. Requirements and Feasibility 8

7. System Architecture 10

8. References 11

1
1. Introduction
Effective and accurate health care has always been the need of the hour. Early detection of
various diseases such as pneumonia, tumour and cancer are very much essential.
Pneumonia, an acute respiratory infection ranked eighth in the list of the top 10 causes of
death in the United States [1]. According to WHO, it accounts for about 1.6 million deaths
a year in this age group - 18% of all deaths among children under five [2]. The disease
frequently goes overlooked and untreated until it has reached a fatal point, especially in the
case of old patients. Series of tests such as Pleural fluid test, Sputum test, blood test, CT
scan, Pulse oximetry and Chest X-ray are conducted for the diagnosis which takes some
time. Chest X-rays are primarily used for the diagnosis of this disease. However, even for a
trained radiologist, it is a challenging task to examine chest X-rays.

2
2.Literature Survey

No. Title Author Methodology Advantages Observation References

1 Pneumonia Shangjie DeepConvDilatedNet Because of the Combining the Google


Detection Yao, K-Means++ CLAHE turbidity of the different sets of
Using an Yaowu Chen, algorithm Soft-NMS pneumonia target, work done in each
Improved Xiang algorithm
the image has network, the Scholar
further been ability of the
Algorithm Tian and algorithm to
enhanced with the
Based on Rongxin detect pneumonia
CLAHE
Faster R- Jiang3 accurately in the
Algorithm to make
CNN RSNA
the target area more
prominent. Dataset is
enhanced.

2 Pneumonia Alhassan deep convolutional Many Dl algorithms This paper Applied


detection on Mabrouk,Rebeca neural network such as MobileNet Sciences
chest X-ray proposes a CNN
P,Diaz Redondo, and DenseNet have
images using Ensemble
Abdelghani incorporated the
ensemble of
Deep Dahou, concept of Learning(EL)
Convolution Mohammed Abd depthwise separable method for
Neural Elaziz and convolutions to
Mohammed address the automatically
Networks
Kayed disadvantage of identifying
traditional normal and
operations.
pneumonia
Patients in chest
X-ray images.

3 Detecting Benjamin CheXNet 121-layer One could try a binary Google


Pneumonia Antin, dense running PCA on the classification Scholar
in Chest Joshua Convolutional images to where the inputs
XRays with Kravitz, and Neural decorrelate the data are chest X-ray
Supervised Network and use these images and the
Learning Emil Martayan
results on a logistic output is one of
regression model. two classes:
pneumonia or
nonpneumonia

3
4 A Deep Xiaowei, Noisy-OR Models with a Two Science
Learning XuXiang ao, Bayesian function a locationattentionme classification
Direct
System to JiangChu nlian threedimensional chanism can classify models are used;
Screen MaPeng, DuXuku CNN segmentation COVID-19, IAVP, one is a relative
n and healthy cases traditional
Novel model withan overall residual network
Coronaviru s accuracy rate of and the other is
Disease 86.7% designed based
2019 on the first
Pneumonia network structure
by concatenatin g
the
locationattention
mechanism to
improve the
overall accuracy
rate.

5 Pneumonia Mohammad The new Hybrid This model had the Ultimately a IEEE
Detection Yaseliani, Ali CNN Model using best performance in Mobile phone Access
Proposing a Hamadani, Abtin terms of all web delivered
Hybrid Ijadi VGG16 and VGG19 performance application of
Deep Maghsoodi and Networks as features metrics, compared pneumonia
Convolution Amir Mosavi extractors to classify to the hybrid model detection was
Neural with FC layers and designed using
CXR images
Network all ML Classifiers the proposed
based on two Ensenble
Parallel classifier, which
Visual can assist
Geometry radiologist in
group pneumonia
architecture
detection with a
and Machine
Learning significant
Classifiers accuracy

4
3. Problem Formulation
Physicians often use chest X-rays to quickly and cheaply diagnose disease associated with
the area. However, it is much more difficult to make clinical diagnoses with chest X-rays
than with other imaging modalities such as CT or MRI. With computer-aided diagnosis,
physicians can make chest X-ray diagnoses more quickly and accurately. Remind,
pneumonia is an infection in one or both lungs. It can be caused by bacteria, viruses, or
fungi. Bacterial pneumonia is the most common type in adults, pneumonia causes
inflammation in the air sacs in your lungs, which are called alveoli. The alveoli fill with
fluid or pus, making it difficult to breathe. Pneumonia is often diagnosed with chest X-Rays
and kills around 50,000 people each year (Center for Disease Control and Prevention, 2016).
With computer aided diagnosis of pneumonia specifically, physicians can more accurately
and efficiently diagnose the disease.

In this application, we hope to train a model using the dataset to help physicians in making
diagnoses of pneumonia in chest X-Rays.

Computerized applications can be using deep learning techniques to increase accuracy and
efficiency in diagnosis. These include chest X-ray images, image processing techniques and
data analysis.

5
4. Objectives
The objectives of the proposed project are as follows:

1. To study and analyse the related work carried on Pneumonia detection

2. To train the hybrid CNN model with FC layers and ML classifiers with the CXR
images.

3. To calculate and compare the accuracy of each model

4. To create a web-based CAD system to assist radiologists in pneumonia detection with


a significance accuracy.

6
5. Methodology
The model would be built using Keras open-source deep learning framework with tensor
flow backend to build and train the convolution neural network model. The architecture of
this system is divided into three different stages - the pre-processing stage, the feature
extraction stage and the classification stage.

A) Dataset: The dataset used is Chest X-ray which is publicly available on the Kaggle33
platform which consists of a total of 5836 images segmented into two main parts, a training
set and a test set. Both bacterial and viral pneumonia are considered as a single category,
pneumonia infected. The entire chest X-ray imaging was conducted as part of patients’
routine medical care. Each image has to be pre-processed according to the deep neural
network used.

B) Pre-processing: Each image had to be pre-processed according to the deep neural


network used. The two important steps involved are: resizing and normalization. Different
neural networks require images of different sizes according to their defined architecture.
All the images are to be normalized according to the respective architectures.

C) Data Augmentation: The data is imbalanced i.e., the X-ray images of the nonpneumonia
class is more as compared to the pneumonia X-ray images. Data augmentation would be
applied on our X-ray image i.e., artificially expand the size of our positive class (X-ray
images with pneumonia). Many data augmentation techniques include such as Horizontal
and Vertical Shift Augmentation, Horizontal and Vertical Flip Augmentation, Random
Rotation Augmentation, Random Brightness Augmentation and many more. Model
Convolution neural networks (CNNs) belong to a class of Deep Learning models that are
highly used in the field of computer vision. These models have multiple processing layers
to learn hierarchical feature representations from the input pixel data. It helps in learning
the model for extracting the features and then constructs the object for classifying the task.
CNN models are feed-forward networks with convolution layers, pooling layers, flattening
layers and fully connected layers employing suitable activation functions. The main
advantage of CNN is that it can detect the relevant features without any human supervision.

D) User side: GUI User side offline model interface would be built, standard GUI library
for Python. The users have to upload the X-ray Image to the GUI application to check for
the disease. The GUI will be simple and easy to use

7
6. Requirements and Feasibility
6.1 Requirements
The proposed project consists of the following requirements:
6.1.1 Hardware requirements
6.1.2 Software requirements

6.1.1 Hardware Requirements


The hardware requirements for the proposed project are depicted in Table 6.1. Table
6.1: Hardware requirements

Sl. No Hardware/Equipment Specification

1. Graphics Card

NVIDIA GTX 1060 / AMD Radeon


RX 480 or greater

2. RAM 8GB or above

6.1.2 Software Requirements


The software requirements for the proposed project are depicted in Table 6.2. Table
6.2: Software requirements (Times new roman 10)

Sl. No Software Specification

1. Anaconda Anaconda 64 bit

8
2. Python Python 3 and above

3. Framework

TensorFlow, Kerasopen-source
framework

6.2 Feasibility
All the hardware and software components required for implementing the proposed project
is readily available. Hence project can be carried out within given time constraint and
budget.

9
7. System Architecture

Figure 7.1. The architecture of the proposed hybrid CNN model with ML classifiers.

Figure 7.2. The architecture of the proposed hybrid CNN model with ML classifiers.

To train the hybrid CNN model with FC layers/ML classifiers, some pre-processing
operations were performed. VGG networks require input images of 224 × 224 pixels and
hence, all the images in the training and test sets were resized to 224 × 224 pixels. All the
images in the dataset contained pixel values in the range [0, 255] and hence, a 1. /255
rescaling was employed to convert all the pixel values of CXR images to the range [0, 1].
These two pre-processing operations were the same for both classification approaches.

To train the hybrid CNN model with FC layers, some additional pre-processing operations
were conducted. Training a CNN requires a large dataset; otherwise, some parameters may
not be appropriately estimated and accordingly, the CNN generalization will be poor. To
tackle this problem, some image augmentation techniques were employed. The number of
available images in the training set would increase by performing these image augmentation
techniques on random samples of images. The used image augmentation techniques
included flipping, zooming, and shearing.

10
8. References
[1] Pneumonia Detection Using an Improved Algorithm Based on Faster R-CNN by
Shangjie Yao, Yaowu Chen, Xiang Tianand Rongxin Jiang, Japan 2018.

[2] Classification of Bacterial and Viral Childhood Pneumonia Using Deep Learning in
Chest Radiography byXianghong Gu,Liyan Pan,Huiying Liang,Ran
Yang,Springer, 2018.

[3] Detecting Pneumonia in Chest X-Rays with Supervised Learning by Benjamin Antin,
Joshua Kravitz, and Emil Martayan, Barcelona, Spain,2019.

[4] A Deep Learning System to Screen Novel Coronavirus Disease 2019 Pneumonia by
Xiaowei ,XuXiangao ,JiangChunlian MaPeng ,DuXukun ,Japan,2020.

[5] Exploration of interpretability techniques for deep covid-19 classification using chest
x-ray images by Soumick Chatterjee,Fatima Saad,Chompunuch Sarasaen, Suhita
Ghosh,Germany,2020.

[6] A. Mangal, S. Kalia, H. Rajgopal et al., “CovidAID: COVID-19 detection using chest
X-ray,” 2020, https://arxiv.org/abs/2004 09803.

11

You might also like