See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/364351755

Analysis of pneumonia detection systems using deep learning-based

approach

Conference Paper · October 2022

DOI: 10.1109/ICSES55317.2022.9914321

CITATIONS READS

5 1,203

2 authors, including:

Beena Godbin
VIT University
9 PUBLICATIONS 26 CITATIONS

SEE PROFILE

All content following this page was uploaded by Beena Godbin on 18 October 2022.

The user has requested enhancement of the downloaded file.

2022 International Conference on Innovative Computing, Intelligent Communication and Smart Electrical Systems (ICSES) | 978-1-6654-7413-9/22/$31.00 ©2022 IEEE | DOI: 10.1109/ICSES55317.2022.9914321

Analysis of pneumonia detection systems using

deep learning-based approach
A.Beena Godbin and S.Graceline Jasmine

Vellore Institute of Technology Chennai

E-mail : [email protected] [email protected]

Abstract- Deep learning technology has been utilized clinical imaging. As a function extraction and class
to analyze images in numerous medical sectors in method, DNN using AI based convolutional neural
recent years. They are very helpful in extracting networks (CNNs) helps to diagnose pneumonia in
complicated information from input images. In the Chest X-rays and CT images.
classification and detection of lung illness, deep
learning plays a critical role.In this article, deep
learning methods are described for detecting Deep learning refers to a gaggle of computer science
pneumonia in medical images.The goal of this paper is techniques that use an oversized variety of linked
to explain about the deep learning approaches used in units to complete difficult jobs. Algorithms based
pneumonia detection systems, as well as to discuss on deep learning will learn representations from a
recent advancements in the field, as well as to vast amount of data automatically.Deep learning is a
highlight problems and potential future guidelines. A natural fit for radiology since it mostly involves
detailed description of deep learning technologies used extracting important information from images, and
to treat pneumonia is presented. We will conclude researches in this area is rapidly developing.As air
with a discussion of how deep learning methods can be pollution rises and the number of smokers increases,
applicable to medical imaging, along with future
respiratory illness is posing a grave danger to
challenges.
people's lives and health.Many early clinical
Keywords—CNN, Medical images, Pneumonia, Deep indications of respiratory problems, on the other
learning, Survey. hand, are not visible, and some patients may not
even be in pain at this point. As a result, for
I.INTRODUCTION respiratory organ disorders and medical care, public
knowledge and an early diagnosis and treatment is
Infections of one or both lungs leading to essential.The present detection approaches
inflammation of their air sacs are known as of pneumonia utilizing Artificial Intelligence
pneumonia.Whenever the air sac becomes swollen models are mentioned in this paper.
or stuffed with pus,coughing with sputum or pus,
fever, chills, and dyspepsia (purulent substance) can II. LITERATURE REVIEW
occur.There are several types of organisms that can
cause pneumonia, including bacteria,fungi and A Proposal on 3 stage learning on deep features and
virus.Viral pneumonia and bacterial pneumonia a framework of fusion was given by Zhou et al. [1].
showvery similar signs and symptoms. Symptoms of For registering images an approach based on
viral pneumonia, on the alternative hand, can convolutional neural networks has been proposed by
beextrasevere than the ones of bacterial pneumonia. Fan [2].
Pneumonia is recognized in kidsbeneath neath the
age of 5who have a cough and/or issuerespiration, Xu et al. [3] researched and developed a CNN
without or withfever, and who have short respiration model, which uses a new hierarchical loss function.
or after inhalation, their chest moves in or retracts, Yates et al. [4] reinstructed the Inception model
resulting in a reduction chest wall (in a wholesome using the last layer of the DNN model and are
person, the chest expands for the duration of classified looking into binary mode.Researchers
inhalation). AI based deep learning has been used to have recognized lung nodule malignancies by
enhance the overall accuracy of computer-assisted comparing ResNet50 and RBF classifications
analysis (CAD), mainly within side the area of developed by Nobrega et al. [5]. Ke et al. [6] used

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on October 18,2022 at 08:25:46 UTC from IEEE Xplore. Restrictions apply.
image descriptors from X-ray images (the spatial Rajpurkar et al. [22] developed a CheXNeXt
distribution of HSB), and an ANN containing self- architecture for detecting many diseases, including
learning algorithms. Behzadikhormouji et al. [7] use pneumonia. The X-ray images with the spatial
pre-trained DCNN with dataset of ImageNetare distribution of HSB were proposed by keet[23]. This
together with ChestNet which is a architecture based is because the brighter image improves the accuracy
on problems.For analyzing brain signals Zhang et al. of the approach. The element that measures the
[8] investigated signal types from the brain relating entire Pareto was generated by the generic algorithm
it with deep learning concepts.In general, the most NSGANetV1 [24], and then several generic
common justification for using the VGG16 model[9] functions were used simultaneously to gradually
is that it produces strong results to process recreate and convert all of the designmodels.To
pneumonia classification. analyze prostrate disease a computerized pipeline
that is CNN based was created by Yoo et al.
Yao L et al take advantage of the dependencies Specifically, in the present study, a data set of 527
between anomalies [10] using the combination of patients was utilized with a region below the ROC
dense net and LSTM(long-term memory rate of 0.87 [30].A few CNN structures have been
network).Khatri et al. propose using Earth Mover's created by Kieu et al [31] to identify typical and
Distance to distinguish infected from uninfected unusual chest pictures. In this method, the CNN
lungs.Rahab [12] and Okeke[13] et al. classified input was parted into 3 sections: right, left, and the
pneumonia using the CNN model, but Rahman et al. whole chest picture. These components were passed
[14] used a technique that supported deep learning. to the CNN, lastly, the final results of the CNN
Xiol [15] planned a more dimensional scale and a architectures were joined with that is known as a
3D heterogeneous deep convolutional neural combination rule. They showed the relevance of the
network(DCNN) (MSHCNN). Nada M. Elshennawy gave numerous CNN techniques a score of 96%.
[16] proposed four different models -a Long
ShortTerm Memory (LSTM), a Convolutional A deep CNN method called Thorax-Netconsists of
Neural Network (CNN), and 2 pre-trained models classification and attention section to detect 14
(MobileNetV2 and ResNet152V2), by changing the thorax diseases proposed by Wang et al. [32]. The
current models. LSTM-CNN recorded results with classifier's feature extraction method saves users
an accuracy of over 91%.Medical image from the difficulty of manually extracting and
segmentation uses the Unet model to segment classifying features.Feature extraction map from the
regions of interest from X-rayimages, achieving the classification branch was analyzed by using the
results of 97.7% and 97.1% w.r.t datasets from attention methods and chronic abnormalities. Chest
Montgomery and JavaScript Runtime (JSRT). To Net attempts to accomplish an outcome by focusing
improve the architecture of complete CNN SENet on the result from these two branches. For the
design is used. Thoraxnet model, the region under the normal ROC
per class has achieved 0.7876.
Zeshan Hussain[17] executed an accuracy of 88% as
compared to 68% with the aid of using optimizing In [33], Talo proposed the ResNet 152 model and a
the statistics augmentation methods. A linear transfer learning method to detect pneumonia. This
mixture of images from the enterstatistics set is the method achieved a 97.4% success rate from datasets
combination in mixup [18] technique, to used without pre-treatment and feature extraction.
featurevariants to the image datasets. The amount of O'Quinn et al. utilized 70% of the information for
foreground label was mixed up proposed by training the data and the remaining for data testing
Olaf[19]. SegNet [20], U-Net [3], and ardiacNet [21] trying to distinguish the presence of the illness from
are some are the important models to analyze 5659 pictures in DICOM design. Alexnet got 76%
medical images which are designed specifically for acknowledgment accomplishment with a CNN
medical purposes as the data needs to be fine-tuned a technique utilizing ensemble methods [33].
lot since the error if any may be very difficult. Lots
of work is already performed in medical data In some other review, Varshini et al[34]proposedthe
analysis and segmentation, and still many differentiation of anticipation for the pneumonia
investigations are in progress. among selective methods principally founded
absolutely on CNN, whichever they utilized by
using transfer learning method, extraction of

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on October 18,2022 at 08:25:46 UTC from IEEE Xplore. Restrictions apply.
features, and numerous classifier methods had been
utilized. They utilized their factual results to uncover
the existing CNN designs utilized along the edge of
managed classifier calculations might be extremely
helpful in concentrating on chest x-ray pictures,
especially for distinguishing Pneumonia.They
claimed that employing the classifier Support Vector
Machines (SVM) and feature extraction approaches,
the DenseNet-169 model produced the best results.
Fig1.CNN Model
Stephen et al. trained CNN chest data set which was
A CNN architecture has two main components
created from scratch in contrast to studies that
 Feature Extraction - the process of identifying
employed transfer learning to diagnose pneumonia.
and separating the various features of the image
They extended the informational collection utilizing
to analyze it.
information expansion techniques, for example,
width and height changes, flat flip, and zoom. This  Fully connected layer that utilizes convolution
CNN model has four convolutional layers with process results and predicts the image class
various quantities of 3x3 channels and the maximum depends on the features separated fromearlier
pooling layer following each convolutional layer, stages.
and also the activation function relu and the final 2
layers are completely associated.This image shown B. Deep Learning Methods
is either not entirely settled to be ordinarily utilizing Since 2016, deep learning models are mostly utilized
the sigmoid capacity in the wake of utilizing a to diagnosepneumonia.Inception_V3, Xception,
completely associated layer, the precision of this DenseNet201, ResnetV2, Resnet50 and
model has been tried at different image aspects and MobileNet_V2 are the most investigated deep
of 200x200 color images. The best performance was learning techniques.Based on accuracies and quality
achieved at 92.73% . results, only 8 techniques are selected.

This paper has many divisions: Division II consists AlexNet

of a literature review; division III contains findings AlexNet can classify extra than one
from the literature, and it has the following sub- thousandexceptionaltrainingthe utilization of
topics Convolutional Neural Network, architectures profound layers comprising of 650,000 neurons and
used, performance metrics, and the data sets used. 60 million boundaries. The organization comprises
Section IV consists of challenges and future scope. of different layers: 3 pooling layers, 5 convolutional
Section V provides the conclusion. In the last, all layers, 2completely connected layer (FCL) with
references are mentioned. theoutput of softmax layer[11]. The input image size
of AlexNet is 227 x 227 x 3, the main convolution
Layer moves the info image to the second layerusing
III. FINDINGS FROM THE REVIEW
96 kernels with a stride of 4 pixels and a size of
11×11×3.
A. Convolution Neural Network
Deep Neural Networks that mainly analyze visual
DenseNet201
images are CNNs, a class of DNN(Deep Neural
With DenseNet, no redundant maps are trained, so it
Networks) that can identify and separate specific
requires fewer parameters than a conventional
features from the images. They can be used to
CNN.The layer of DenseNet is completely thin,
recognize images and videos, classify images,
contains only twelve filters that produce a tiny count
analyze medical images, implement computer
offeature vector. DenseNet201, DenseNet264,
vision, and process natural language.
DenseNet121, and DenseNet169 are the types of
DenseNet model. DenseNet264 was used in this
study to detect pneumonia. All layers of densenet
have permission to both the actualgiven image and
the gradient values fom the loss value function.

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on October 18,2022 at 08:25:46 UTC from IEEE Xplore. Restrictions apply.
 105.00
100.00
95.00
90.00
85.00
80.00
75.00
70.00

Accuracy Sensitivity Specificity Fl Measure

Fig 4. Accuracy of models in pneumonia detection

Inception_V3
This is a CNN model (Szegedy et al., 2014)
Fig.2. Densenet Architecture developed by Szegedy in 2014, and the structure in
which the starting model is the starting block is
SqueezeNet different from that of a normal CNN.(the same input
The CNN has been trained on the ImageNet database tensor overlaps multiple filters and the result is
and is called SqueezeNet.SqueezeNet contains 50 concatenated).Inception_V3 is another adaptation of
times fewer parameters than AlexNet and was the initiation model exhibited interestingly in 2015
trained with over 1 million images.Above model, 1 1 (Szegedy et al., 2015), a superior variant of
channels are utilized in the Squeeze layer, which inceptionV1 and inceptionV2 with many boundaries
takes care of into an expand layer with a blend of having a square of equal CLwith three unique
1x1 and 3x3 convolution channel, this network structure of channels and 3×3 max pooling(pooling)
consisting of Squeeze Layer and Expand layer. is likewise performed.This results are also linked
and send off in the next step.
VGG16 and VGG19
VGG (Visual Geometry Group) members Simonyan Resnet50
and Zisserman submitted a CNN (Convolution Resnet50 accepts images with a resolution of
Neural Net) architecture in 2014 and won the ILSVR 224×224 pixels and has 50 residual
(ImageNet) contest. The major aspect of this design networks.Naseem Ansari et al developed their
is that it concentrates on a basic kernel with 3x3 for examination on pneumonia distinguishing proof
the convolution layer and 2x2 for the maximum which by investigation makers applied on two
pooling layer, rather than more hyperparameters. different data set and using profound figuring out
how to retrain the pre-arranged ResNet-50 model.A
CNNmodels are formed with more than a million
images from the image database (Szegedy et al.,
2016) is inception_resnet_V2, a hybrid technique
that combines startup and residual connections. This
model accepts the 299x299 frame image and the
result is a list of estimated layer probabilities.

Fig 3.Architecture of VGG Xception

Chollet proposed the Xception model in 2017, a 71
VGG16 produces strong results for fine-tuning tasks layers profound convolutional neural network is a
in many circumstances [9], which is the most superior form of Inception design and includes
common justification for using this model for profundity distinguishable convolutions. In ordinary
transfer learning to process pneumonia classification order issues with 299x299 pixels of input dimension,
at first. The accuracy evolution results for the Xception substitutes the typical Inception modules
Inception-Resnet-v2 and ResNet152V2 based model with profundity insightful divisible convolutions and
are similar. accomplishes great outcomes when contrasted with

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on October 18,2022 at 08:25:46 UTC from IEEE Xplore. Restrictions apply.
Resnet, Inception and VGG16 network for various CNN calculations tried for the 3
different order plans where DenseNet201 beats
Table 1 displays the summary of the execution different models as far as various execution files.

Table.1.Accuracy of models
Results
Disease Methods AUC F1 Score
Accuracy Recall Specificity
Normal, AlexNet 0.87 0.87 0.95 0.91 0.88
Bacterial DenseNet201 0.93 0.93 0.96 0.95 0.93
and Viral ResNet18 0.88 0.89 0.95 0.91 0.90
Pneumonia SqueezeNet 0.86 0.85 0.93 0.89 0.86
AlexNet 0.91 0.95 0.83 0.89 0.92
Viral and Bacterial
DenseNet20l 0.95 0.97 0.95 0.95 0.95
Pneumonia
SqueezeNet 0.84 0.91 0.74 0.83 0.84
ResNet18 0.88 0.91 0.83 0.87 0.87

C. Machine Learning Classifiers

Decision Tree
To predict the value of a target, decision trees use
data features to understand simple rules based on the
datavariables using multivariate supervised learning
methods. This tree can be thought of as a piece-wise
constant approximation.

Naive Bayes Classifier

The Naive Bayes family of probability based
algorithms determines whether a data point falls into
one or more of a set of options (or doesnot).A text
analysis technique called Naive Bayes categorizes
consumer news articles, emails, commentsand other
types of text, and organizes/categorizes them based Fig.5.KNN classifier
on the topics or "tags" associated with them.
Support Vector Machines (SVM)
K-Nearest Neighbor In supervised learning, support vector machines
An approach to pattern recognition, known as K- (SVMs) are used for regression,outlier detection and
Nearest Neighbor (k-NN), involves storing and classification.
learning patterns from trained data units, findingin
how they belongs to other datas in n-dimensional The CNN VGG19 and the SVM classifier got the
vector.The goal of KNN is to identify the best pneumonia detecting model using CXR images
future/unseen k-closest nearer data units.In text since This method had a 96.47 percent accuracy, a
identification, KNN allocates a single word or 97.47 percent sensitivity score, a 95.56 percent F1
expression to the predefined classification from score, and a 97.56 percent accuracy.
computing its closest neighbor: not entirely settled
by a greater part of its neighbors. If k = 1, it belongs Artificial Neural Networks
to that class that is very close to one. Artificial neural networks(ANN) or “deep neural
network” methods needs very huge collection of
data for training as their sequential steps are highly
advanced. With proper training, they can perform
better than other algorithms.

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on October 18,2022 at 08:25:46 UTC from IEEE Xplore. Restrictions apply.
 101.00

99.00

97.00

95.00
Random forest Naive Bayes
93.00 A classification model's effectiveness is evaluated
kNN SVM
Accuracy
ANN
Sensitivity Specificity AUC using four main performance metrics:
 Accuracy: Predicting both classes correctly with
Fig.6.Performance of classifiers in pneumonia the test
detection  Precision: “Detects positive values correctly
from all output positive values”
D. Evaluation Criteria  Recall (Sensitivity): “the tests ability to
Classification models used has to be measured to distinguish between all original positive values
find the capability for implementing a certain from false positives”
task.Best classification methods will be very  F1 Score: “Calculation ofprecision and recall
effective in prediction and can be useful in tasks that harmonic mean ”
require prediction. Whereas, poor classification
models produce unreliable results which we cannot
depend on outcomes, and thus, are not useful for the
user.

E. Data sets
The below datasets were used in the literature
survey.

Pneumonia Data Set Kaggle

Kaggle has this data set. There are 5863 X Rays in
it. This chest X-ray image collection includes 1590
normal images and 4273 pneumonia images. It
contains three folders: test, training, and validation.
There are 2 subfolders for normal and pneumonia in
each of the three folders. Anterior-posterior chest X-
rays are well-known methods for detecting diseases
Fig.7.Various metrics used because they detect their presence most effectively.
This datasetbelongs to Guangzhou Children's and
 True Positives(TP):“Predicted as a positive ladies Medical Center. Only JPEG collections are
outcome” included in this data collection.
 True Negatives(TN): “Predicted as a negative
outcome” RSNA Pneumonia dataset
 False Positives(FP): “mis predicted as a This data set was created for a Kaggle competition.
positive outcome” This competition is organized from the North
America Radiological Society(RSNA).There are
 False Negatives(FN): “mis predicted as a
30287 X-ray images in this data collection, 9555 of
negative outcome”
which are pneumonia images and 20672 of which
are normal images of chest x-ray.These images are

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on October 18,2022 at 08:25:46 UTC from IEEE Xplore. Restrictions apply.
typically 1024*1024 pixels in size. The National learning(DL), with a focus on pneumonia. The use
Institutes of Health (NIH) has compiled this data set of CNN is high, according to assessments of the
( National Institutes of Health). This includes distribution of works. Data asymmetry, dealing of
collection of CSV files with various labels, such as large medical image sizes, existence of lesser
zero for normal and onemeans pneumonia. number of datasets,and significant error correlation
when utilizing ensemble approaches were identified
NIH CXT(Chest X-ray)Dataset as four challenges with machine learning is being
Medical data from NIH clinical centers have been used to detect the lung disorder.Ultimately,
made public to support autonomous x-ray detection. determining the extent to which deep learning
There are X-ray images of lungs and chests from methods was used in pulmonary identification is
around the world in this collection. There are critical in ensuring that future research stays on
112,120 images in this chest X-ray. In this data set, track, ultimately enhancing the effectiveness of
14 diseases are listed. It contains enormous amounts disease identification systems. In the future, deep
of data, such as patient specifics, such as age and learning-aided applications for recognizing lung
gender, in CSV format. This dataset covers various illnesses could be improved, while the number of
diseases such as Edema, Emphysema, Atelectasis, applications could also be multiplied.
Infiltration, Emphysema, Effusion, Fibrosis,
Pneumonia, Consolidation, thickening of pleural, REFERENCES:
Pneumothorax, Edema,, Nodule, Mass, Hernia and
[1] Zhu, X.; Shen, D. .Zhou, T.; Thung, K.;Effective feature
Cardiomegaly. learning and fusion of multimodality data using stage-wise deep
neural network for dementia diagnosis. Hum. Brain Mapp.
IV.CHALLENGES AND FUTURE DIRECTIONS 2018, 40, 1001–1016.
[2] Wu, H.; Xu, S.; Bie, R. CXNet-m1: Anomaly detection on
chest X-rays with image-based deep learning. IEEE Access
Despite the advances of deep learning methods in 2019, 7, 4466–4477.
[3] Harvey, H. Yates, E.J.Yates, L.C.;Machine learning “red dot”:
the medical and clinical fields, numerous limitations Open-source, cloud, deep convolutional neural networks in
and obstacles remain. Deep learning analysis usually chest radiograph binary normality classification. Clin. Radiol.
necessitates a huge amount of labeled data. This is a 2018, 73, 827–831.
[4] Yap, P.T.; Shen,Fan, J.; Cao, X.; D. BIRNet: Brain image
significant difficulty when it comes to labeling of registration using dual-supervised fully convolutional networks.
medical images. Labeling medical images Med. Image Anal. 2019
[5] J.; Wei, W.; Połap, D.; Ke, Q.; Zhang, Wo ´zniak, M.;
necessitate specialist information, such as a Ko´smider, L.; Damasevi ˇ ˇcius, R. A neuro-heuristic approach
radiologist's domain knowledge. Therefore, for recognition of lung diseases from X-ray images. Expert Syst.
annotating enough medical images can be tedious Appl. 2019, 126, 218–232. 54, 193–206.
[6] H. S. Tran, T. H. Le, P. N. Kieu,T. Le, and T. T. Nguyen,
and time-consuming. Annotating medical images is "Applying Multi-CNNs model for detecting abnormal problem
not easy, but unlabeled medical images are stored in on chest x-ray images," 2018 10th International Conference on
PACS for long periods, resulting in huge volumes.It Knowledge and Systems Engineering (KSE), Ho Chi Minh
City, 2018, pp. 300-305.
would save a lot of time and effort if unlabeled [7] da Nóbrega, R.V.M.; Rebouças Filho, P.P.; Rodrigues, M.B.; da
images could be used in deep learning algorithms. Silva, S.P.P.; Dourado Júnior, C.M.J.M.; de Albuquerque,
V.H.C. Lung nodule malignancy classification in chest
Deep learning's interpretability is also a problem. computed tomography images using transfer learning and
convolutional neural networks. Neural Comput. Appl. 2018, 1–
Deep learning models are sometimes viewed as a 18.
[8] Muhammad E. H. Chowdhury2, Tawsifur RahmanAmith
black box, with no ability to interpret their success Khandakar2, Khandaker R. Islam3, Khandaker F. Islam2, Zaid
or failure. The demand for research into these B. Mahbub4, Muhammad A. Kadir1, Saad Kashem5 Transfer
techniques is growing, paving the path for clinical Learning with Deep Convolutional Neural Network (CNN) for
Pneumonia Detection using Chest X-ray
deep learning applications in medical image [9] H.; Rostami, H.; Salehi,Behzadi-khormouji,S.; Derakhshande-
processing. From a legal standpoint, deep learning is Rishehri, T.; Masoumi, M.; Salemi, S.; Keshavarz, A.;
Gholamrezanezhad, A.; Assadi, M.; Batouli, A. Deep learning,
being used in the medical industry would necessitate reusable and problem-based architectures for detection of
transparency and interpretability. consolidation on chest X-ray images. Comput. Methods
Programs Biomed. 2020, 185, 105162.
[10] Satyarth Katiyar 2,†, Avinash G Keskar 3,†, Neeraj Dhanraj
V.CONCLUSION Bokde 4,† and Zong Woo Geem 5, Mohammad Farukh
HashmiEfficient Pneumonia Detection in Chest Xray Images
Using Deep Transfer Learning.
As a result, this paper is written to provide a survey [11] Wang, X.; Monaghan,Zhang, X.; Yao, L.; J.; McAlpine, D. A
of lung illness identification using deep Survey on Deep Learning based Brain-Computer Interface:
Recent Advances and New Frontiers. arXiv 2019,

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on October 18,2022 at 08:25:46 UTC from IEEE Xplore. Restrictions apply.
 arXiv:1905.04149. [25] Youness CHAWKI2, Ali IDRI1, Khalid EL ASNAOUI1,
[12] S.M. Jadhav, S.S. Yadav,"Deep convolutional neural network- Automated Methods for Detection and Classification
based medical image classification for disease diagnosis," Pneumonia based on X-Ray Images Using Deep Learning
Journal of Big Data, 6(1), 113, 2019, doi:10.1186/s40537-019- [26] Chowdhury, M.E.; Khandakar,Rahman, T.; A.; Islam, K.R.;
0276-2. Islam, K.F.; Mahbub, Z.B.; Kadir, M.A.; Kashem, S. Transfer
[13] E.; Dagunts, DYao, L.; Poblenz.; Covington, B.; Bernard, D.; Learning with Deep Convolutional Neural Network (CNN) for
Lyman, K. Learning to diagnose from scratch by exploiting Pneumonia Detection using Chest X-ray. Appl. Sci. 2020, 10,
dependencies among labels. arXiv 2017, arXiv:1710.10501. 3233.
[14] Vashista, H.; Mittal,.Khatri, A.A.R.J.; N.; Ranjan, P.; [27] I. Gujrathi, M.A. Haider,S.Yoo,F. Khalvati "Prostate Cancer
Janardhanan, R. Pneumonia Identification in Chest X-Ray Detection using Deep Convolutional Neural Networks", A
Images Using EMD. In Trends in Communication, Cloud, and Nature Research Journal, Scientific Reports volume 9, Article
Big Data; Springer: Singapore, 2020; pp. 87–98. number: 19518 (2019).
[15] Ma’aitah, M.K.S..Abiyev, R.H.; Deep convolutional neural [28] H. Jia, L. Lu,H. Wang and Y. Xia, "Thorax-Net: An Attention
networks for chest diseases detection.J. Healthc. Eng. 2018, Regularized Deep Neural Network for Classification of
2018, 4168538. 107.Stephen, Thoracic Diseases on Chest Radiography," in IEEE Journal of
[16] C. Roberto,B. Vijay, K. Alex, “Segnet: Deep convolutional Biomedical and Health Informatics, vol. 24, no. 2, pp. 475-485,
encoder-decoder architecture for image segmentation,” 2015, Feb. 2020.
http://arxiv.org/abs/1511.00561. [29] 1 Jeremy Irvin * 1 Kaylie Zhu 1 Pranav Rajpurkar Brandon
[17] J.; Wei, W.; Połap, D,. Ke, Q.; Zhang; Wo ´zniak, M.; Yang 1 Hershel Mehta 1 Tony Duan 1 Daisy Ding 1 Aarti
Ko´smider, L.; Damaševiˇcius, R. A neuro-heuristic approach Bagul 1 Robyn L. Ball 2 Curtis Langlotz 3 Katie Shpanskaya 3
for recognition of lung diseases from X-ray images. Expert Syst. Matthew P. Lungren 3 Andrew Y. Ng CheXNet:
Appl. 2019, 126, 218–232. [30] M. Talo, "Pneumonia Detection from CT Images using
[18] U.J.; Jeong, D.U..O.; Sain, M.; Maduh An efficient deep Convolutional Neural Networks," 2019 27th Signal Processing
learning approach to pneumonia classification in healthcare. J. and Communications Applications Conference (SIU), Sivas,
Healthc. Eng. 2019, 2019, 4180949. Turkey,2019.
[19] N.; Geng, L.; Zhang, F.;Xiao, Z.; Du Wu, J.; Liu, Y. Multi-scale [31] Nijhawan Kartik Thakral Ankush Mittal Lucky Agarwal Dimpy
heterogeneous 3D CNN for the false-positive reduction in Varshni RahulPneumonia Detection Using CNN-based Feature
pulmonary nodule detection, based on chest CT images. Appl. Extraction.
Sci. 2019, 9, 3261. [32] R. J. Haddad,W. O’Quinn, and D. L. Moore, "Pneumonia
[20] M.; Ibrahim, D.M..Elshennawy, N. Deep-Pneumonia Radiograph Diagnosis Utilizing Deep Learning Network," 2019
Framework Using Deep Learning Models Based on Chest X- IEEE 2nd International Conference on Electronic Information
Ray Images. Diagnostics 2020, 10, 649. and Communication Technology (ICEICT), Harbin, China,
[21] Francisco Gimenez2, Ph.D.,Zeshan Hussain1, Darvin Yi2, 2019, pp. 763-767.
Daniel Rubin2, MD, MS 1Stanford University, Department of [33] Agarwal, R. Nijhawan and A. MittalD. Varshni, K. Thakral, L,
Computer Science, Stanford, CA; 2Stanford University, "Pneumonia Detection Using CNN based Feature Extraction,"
Department of Radiology, Stanford, CA Differential Data 2019 IEEE International Conference on Electrical, Computer
Augmentation Techniques for Medical Imaging Classification and Communication Technologies (ICECCT), Coimbatore,
Tasks. India, 2019, pp. 1-7.
[22] Felix Bragman Sebastien Zach Eaton-Rosen,Ourselin M. Jorge [34] M. Sain,O. StephenU. J. Maduh, D. Jeong, "An Efficient Deep
Car do so proving Data Augmentation for Medical Image Learning Approach to Pneumonia Classification in Health
Segmentation. care”.
[23] K. Rashed, R. Kawal, B.M. Aliasghar, Jeremy, and B. Ulas, [35] M. Revathi, A. B. Godbin, S. N. Bushra and S. Anslam Sibi,
Cardiacnet, Segmentation of Left Atrium and Proximal "Application of ANN, SVM and KNN in the Prediction of
Pulmonary Veins from MRI using Multi-View CNN, MICCAI, Diabetes Mellitus," 2022 International Conference on Electronic
Springer, New York, NY, USA, 2017. Systems and Intelligent Computing (ICESIC), 2022, pp. 179-
[24] In Proceedings of the International Conference of AI and 184, doi: 10.1109/ICESIC53714.2022.9783577.
Information Technology (ICAIIT), Ouargla, Algeria, 4–6 March
2019; pp. 486–489.

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on October 18,2022 at 08:25:46 UTC from IEEE Xplore. Restrictions apply.
View publication stats