As per UGC guidelines an electronic bar code is provided to seure your paper

International Journal for Modern Trends in Science and Technology, 7(01): 77-80, 2021
Copyright © 2021 International Journal for Modern Trends in Science and Technology
ISSN: 2455-3778 online
DOI: https://doi.org/10.46501/IJMTST070117
Available online at: http://www.ijmtst.com/vol7issue01.html

Pneumonia Detection Using Convolutional

Neural Networks
Puneet Gupta

Information Technology, Maharaja Agrasen Institute of Technology Delhi, India

To Cite this Article

Puneet Gupta, “Pneumonia Detection Using Convolutional Neural Networks”, International Journal for Modern Trends in
Science and Technology, Vol. 07, Issue 01, January 2021, pp.- 77-80.

Article Info
Received on 22-November-2020, Revised on 18-December-2020, Accepted on 28-December-2020, Published on 04-January-2021.

ABSTRACT
Abstract— Pneumonia is a life-threatening infectious disease affecting one or both lungs in humans
commonly caused by bacteria called Streptococcus pneumoniae. One in three deaths in India is caused due
to pneumonia as reported by World Health Organization (WHO). Chest X-Rays which are used to diagnose
pneumonia, need expert radiotherapists for evaluation. Thus, developing an automatic system for detecting
pneumonia would be beneficial for treating the disease without any delay particularly in remote areas. Due
to the success of deep learning algorithms in analyzing medical images, Convolutional Neural Networks
(CNNs) have gained much attention for disease classification. In addition, features learned by pre-trained
CNN models on large-scale datasets are much useful in image classification tasks. In this work, we appraise
the functionality of pre-trained CNN models utilized as feature-extractors followed by different classifiers for
the classification of abnormal and normal chest X-Rays. We analytically determine the optimal CNN model
for the purpose. Statistical results obtained demonstrates that pretrained CNN models employed along with
supervised classifier algorithms can be very beneficial in analyzing chest X-ray images, specifically to detect
Pneumonia. In this project Transfer learning and a CNN Model is used to detect whether the person has
pneumonia or not using chest x-ray.

KEYWORDS: Pneumonia, Convolutional Neural Networks, Detection, Deep Leaning, Transfer Learning.

INTRODUCTION CT-scan, or MRI is the common method to discover

PNEUMONIA is an illness that disturbs the lung pneumonia. Medical personnel check the patient’s
air sacs of an infected person. It is triggered by radiograph of their chest to determine if they are
bacteria, fungi, or a virus that infects the air sacs of infected with pneumonia or not. In addition, the
lungs that fill up with discharge fluids that leads to usual method for finding pneumonia is through
chills, fever, coughing with mucus, and breathing medical history and laboratory results of the
trouble among persons diagnosed with this patient.
disease. Children below five years of age and Radiograph of chest is penetrated through
elderly patients with weak immune system are X-rays where the soft tissues produces a dark color
vulnerable to this type of diseases. Pneumonia has and hard tissues like bones produces a bright
killed over a million children worldwide in 2018 color. Patients diagnosed with pneumonia shows
and remains a life-threatening disease now a days the chest cavity signs of fluids filling the air sacs of
if not detected or diagnose earlier. Radiography, lungs as for the radiograph picture appears

77 International Journal for Modern Trends in Science and Technology

brighter. Several abnormalities may be seen on conclusion and recommendations,
lung cavities as brighter color may represent such acknowledgments and references.
as cancer cells, blood vessels swelling, and
abnormality of heart. To validate the range and STRUCTURE OF PAPER
spot of an infected area of the lungs, chest x-rays is The paper is organized as follows: In Section 1,
the utmost method. In these method, emergence of the introduction of the paper is provided along with
the disease can be imprecise and misinterpreted the structure, important terms, objectives
with another illness. Therefore, the undertaking is andoverall description. In Section 2 we discuss
pleasing in the improvement of the processing in related work. In Section 3 we have the complete
medical situations in isolated areas for pneumonia information about Convolution Neural Network
detection. The researchers were able to train and and in section 4 we haveMethodology and the
assessed CNN model’s performance and classify process description.
chest x-rays with normal and infected with disease Section 5 tells us about the future scopeand
using different classifiers. With the recent concludes the paper with acknowledgement
development of Computer Aided Design (CAD) tools andreferences.
becomes the most important field of research in
artificial intelligence and machine learning. CAD OBJECTIVES
systems has proven in facilitating the medical field Different models of deep learning and transfer
such as breast cancer detection, classification of learning are analyzed in this work for the image
disease using mammograms, lung cancer classification application. Using Deep Learning
detection, etc. CAD system is an applicable features are extracted from the images and are
instrument in use today for diagnosis and used for classification of x-rays having pneumonia.
classification of diseases in medical imaging. In A convolutional neural network architecture is
achieving the precise diagnosis, the medical made and it is trained with the images of x-rays
personnel integrate the CAD to assist and verify to with pneumonia and normal x-rays which will be
support their decision making. Significant features used for further classification. We also used Vgg16
of the images are valuable in employing machine and Vgg19 transfer learning technique to compare
learning techniques in this system compared to the the accuracy of the models. CNN models have been
traditional handcrafted features which has created from scratch and trained on Chest X-Ray
limitations in extracting significant features. The Images (Pneumonia) dataset on Kaggle.
progress in a more intelligent future is now
productive through generations. This technological RELATED WORK
improvement today reached new step closer in CNN is a very powerful algorithm which is widely
human intelligence. The deep learning has gained used for image classification and object detection.
the ability in simulating the function of the human The hierarchical structure and powerful feature
brain. It recommends the solution to solve real-life extraction capabilities from an image makes CNN a
problems. The deep learning by means of the very robust algorithm for various image and object
convolutional neural networks has ability in recognition tasks. There are various architectures
obtaining significant characteristics in image of CNNs available which have been key in building
classification tasks and provides medical algorithms which power and shall power AI as a
promising results in image analysis . CNN whole in the foreseeable future. Some of them are:
advantagesis capable in assisting the identification LeNet, AlexNet, VGGNet, GoogLeNet, ResNet,
of some features from an image and use this ZFNet and etc
feature to generate probabilities in classifying
specific input. The contribution of this study is
developed an optimized deep learning models of CONVOLUTION NEURAL NETWORK
CNN that can detect and classify pneumonia
Process of CNN is to detect and categorize images
diseases efficiently [10]. The work consists of an
from learned features. It is very effective in a
optimized CNN models and experimental analysis
multi-layered structure when obtaining and
of each model towards the detection and
assessing necessary features of graphical images.
classification of pneumonia diseases.This research
Figure 1 illustrates the CNN process.
article consists of the following sections:
introduction, convolutional neural network,
methodology, experimental results and discussion,

78 International Journal for Modern Trends in Science and Technology

 1.3 Transfer Learning.
For an effective classification, the models are set for
proper setting to the needed task. Transfer learning
let the model save to original parameter settings of
the previously trained model to bring effective score
without using intense computing power [12]. The
Fine-Tuning method can extract new features from
pneumonia and change the original 1000 neurons
for example the VGGNet model to only four [13].
Also, FC or dense layers of 4096 down to a more
significant number of 512 and 256 to prevent any
waste of computing resources. Hyperparameters
1.1 Convolution Layer facilitate the deep learning model to produce
Convolution in CNN is performed on an input significant results just as the actual training
image using a filter or a kernel. To understand procedure takes place. Selecting the exact values
filtering and convolution you will have to scan the can initially make a huge dissimilarity by making
screen starting from top left to right and moving hyperparameter tuning a necessary process. Still,
down a bit after covering the width of the screen there are no precise hyperparameters for every
and repeating the same process until you are done algorithm [14]. Thus, a practical experimentation
scanning the whole screen. must take place.

1.4 Classification
Every single block as shown in Fig. 1 are used in
architecture followed by wholly inter-connected
layers and softmax activation. Feature extraction is
composed of an input image, convolution,
max-pooling while classification involves fully
interconnected layers and output .

METODODLOGY:
Fig. 2 Demonstration of Convolution Layer Data augmentation: is a common step used for
increasing the dataset size and the model
1.2 Max-Pooling Layer generalizability. Essentially, it is the process of
This layer cut down the values further to half of its artificially increasing the size of a dataset via
original value by choosing only max values from transformations — rotation, flipping, cropping,
the kernel matrix. The sample illustration is the stretching, lens correction
4x4 matrix pixel values of an input image degraded Feature Extraction: In this part, the network
into 2x2 filters. Its process is described in Fig. will perform a series of convolutions and pooling
operations
duringwhichthefeaturesaredetected.Ifyouhadap
ictureofazebra,thisisthepartwhere the network
would recognize its stripes, two ears, and
fourlegs.
Classification: Here, the fully connected layers
will serve as a classifier on top of these
extracted features.
Theywillassignaprobabilityfortheobjectontheim
agebeingwhatthealgorithmpredicts itis.

Fig. 3 Illustration of Max Pooling Layer

79 International Journal for Modern Trends in Science and Technology

FUTURE SCOPE AND CONCLUSION physically so that the framework in the feature
We have assessed the performance of 3 CNN extractionis employed. The pneumonia infected
models namely; diseases and normal chest x-ray image dataset are
VGG16, VGG19 and a CNN model that we built acquired. Our study enables to identify among the
from scratch three models the best model to detect pneumonia.
For VGG16 we got validation accuracy of 92%and All the models have performed well on detecting
training accuracy of about 97% and for VGG19 we pneumonia and normal chest x-rays. We have
got validation accuracy of 89%and training demonstrated how to classify positive and negative
accuracy of about 95% pneumonia data from a collection of X-ray images.
And for the model that we built we got validation We build our model from scratch, which separates
accuracy of 93%and training accuracy of about it from other methods that rely heavily on transfer
99%. learning approach. We also used transfer learning
approach to solve and we got good results with the
Pneumonia dataset used in the study. Table 2
shows the overall performance of 3 models used CNN model we built and it was more accurate than
the transfer learning models that we used.
in thisstudy.

In the future, this work will be extended to detect

and classify X-ray images consisting of Covid-19
and pneumonia. Distinguishing X-ray images that
contain Covid-19 and pneumonia has been a big
VGG16 issue in recent times, and our next approach will
tackle this problem

ACKNOWLEDGEMENT
Aside from God Almighty, the researchers would
like to express their gratitude to their families,
relatives, and dear ones whose involvement made a
significant impact on the accomplishment of this
task.
VGG19
REFERENCES
[1] https://medium.com/@dipti.rohan.pawar/improving-perf
ormance-of-convolutional-neural-network-2ecfe0207de7E.
[2] https://arxiv.org/abs/1409.1556
[3] https://machinelearningmastery.com/how-to-use-transfe
r-learning-when-developing-convolutional-neural-network
-models/
[4] https://analyticsindiamag.com/practical-comparison-of-t
ransfer-learning-models-in-multi-class-image-classificatio
n/
Custom CNN
[5] https://medium.com/abraia/first-steps-with-transfer-lear
Model ning-for-custom-image-classification-with-keras-b941601f
cad5

Table 1. Result of Training and Validation of

Five CNN models

This research explores the deep learning function

in detecting pneumonia through computer vision
using three convolutional neural network models.
Every CNN models identified are assessed

80 International Journal for Modern Trends in Science and Technology