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

net/publication/360738322

Pneumonia Detection and Classification using CNN and VGG16

Article in International Journal of Advanced Research in Science Communication and Technology · May 2022
DOI: 10.48175/IJARSCT-3851

CITATIONS READS
12 2,472

5 authors, including:

Sunil L. Bangare
Sinhgad Technical Education Society Sinhgad Academy of Engineering
70 PUBLICATIONS 1,358 CITATIONS

SEE PROFILE

All content following this page was uploaded by Sunil L. Bangare on 29 September 2022.

The user has requested enhancement of the downloaded file.

 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 2, Issue 3, May 2022

Impact Factor: 6.252

Pneumonia Detection and Classification using CNN

and VGG16
Dr. Sunil L. Bangare1, Hrushikesh S. Rajankar2, Pavan S. Patil3, Karan V. Nakum4, Gopal S. Paraskar5
Associate Professor, Department of Information Technology1
UG Scholar, Department of Information Technology2,3,4,5
Sinhgad Academy of Engineering, Pune, Maharashtra, India.
Savitribai Phule Pune University, Pune, Maharashtra, India.

Abstract: Pneumonia, an infectious disease caused by a bacterium in the lungs' alveoli, is frequently the
result of pollution. A lung infection causes pus to build up in the affected tissue. Professionals conduct bodily
examinations and diagnose their patients using a chest X-ray, ultrasound, or lung biopsy to determine if they
have certain conditions. Misdiagnosis, incorrect treatment, and failure to recognize the disease will result in
a patient's inability to lead a normal life. Deep learning's advancement helps specialists make better decisions
when diagnosing patients with certain diseases. The research provides a flexible and efficient deep learning
technique that uses the CNN model to predict and detect a patient who is unaffected. Using a chest X-ray
photograph, the study applies a flexible and effective deep learning technique of using the CNN model in
predicting and detecting a patient unaffected and affected by the illness. To demonstrate the overall
performance of the CNN model being trained, the researchers used an amassed dataset of 20,000
photographs and a 224x224 photograph decision with 32 batch lengths. At some point throughout the total
performance training, the trained version produced a 95 percent accuracy charge. The research study may
detect and predict COVID-19, bacterial, and viral pneumonia illnesses based solely on chest X-ray
photographs, according to the results of the testing.

Keywords: Pneumonia Detection, Adaptive Deep Learning, Deep Convolutional Neural Network
Architecture

I. INTRODUCTION
Pneumonia is an inflammation of the lung parenchyma caused by pathogenic bacteria, physical and chemical factors,
immunologic injury, and various medications.
There are several common pneumonia classification methods: (1) Pneumonia is classed as infectious or non-infectious
based solely on unique pathogeneses, with infectious pneumonia being the more common. Immune-associated pneumonia,
aspiration pneumonia induced by physical and chemical factors, and radiation pneumonia, among other things, are all types
of infectious pneumonia. Immune-associated pneumonia, aspiration pneumonia produced by physical and chemical factors,
and radiation pneumonia are all examples of non-infectious pneumonia. (2) Pneumonia is divided into three types: CAP
(community-acquired pneumonia), HAP (hospital-acquired pneumonia), and VAP (ventilator-related pneumonia), with
CAP accounting for the majority of occurrences. as a result of the wide range of disorders,
HAP makes it easier to develop resistance to a variety of antibiotics, making treatment more challenging. Every year,
more than 800,000 children under the age of five die from pneumonia. over 2200 people die. More than 1400 children per
100,000 become affected by pneumonia. Pulmonary infections, such as pneumonia, were the second greatest reason for
death in 2013, according to the Global Burden of Disease Study. The pneumococcal disease has inflamed approximately 35
percent of patients in European hospitals, and 27.3 percent of patients worldwide. According to a recent report from John
Hopkins Bloomberg College of Public Health, India had the highest rate of pneumonia mortality, with nearly 2.97 lac
pneumonia combined diarrhea mortality among children under the age of five in 2015. similarly, Pneumonia was the leading
cause of death among children under the age of five years old in 2015. Furthermore, pneumonia's death rate is inversely
proportional to its age, and pneumonia's superiority rises substantially with age, especially in humans older than sixty-five.
The huge range of baby deaths via pneumonia alarms scientists international to recommend extra effective and acute
strategies to stumble on pneumonia. With the era growing, greater and extra measures are advanced, where radiology-based
Copyright to IJARSCT DOI: 10.48175/IJARSCT-3851 771
www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 2, Issue 3, May 2022

Impact Factor: 6.252
treatments are the most popular and effective. Chest X-ray imaging, computed tomography (CT), and magnetic resonance
imaging (MRI) are all diagnostic radiological techniques for pulmonary disease, with chest X-ray scanning being the most
efficient and cost-effective because it is far more accessible and portable in hospitals, and it exposes sick people to lower
doses of radioactivity. However, even for multiple skilled and experienced medical physicians, analysing pneumonia using
X-ray snapshots remains a difficult task because X-ray images contain comparable area statistics for unique ailments, such
as lung cancer. As a result, diagnosing pneumonia with conventional procedures is time-consuming and energy-intensive,
and it is impossible to diagnose whether or not a patient has pneumonia in a uniform manner. As a consequence, in this
study, we propose using a Convolutional Neural Network to autonomously diagnosis pneumonia using X-ray pictures, with
an accuracy of 96.07 percent and an AUC of 0.9911. The remaining sections of this work are organised as follows. The
literature's perspectives on medical picture processing processes are discussed in the second section. In recent times, section
three has described a rapid way of Convolutional Neural Networks (CNN) architecture. The final segment included an
outline of Machine Learning and Deep Learning's history. The material employed in this investigation, our proposed
techniques, and the training method are all depicted in Section 4. The trials and their outcomes are presented in section 5.
This observer's belief is described in section 6.

II. RELATED WORKS

[1]There are several obstacles in creating a sustainable version to identify COVID-19 from X-ray pictures. First and
foremost, there are only a limited quantity of photos available. The pandemic has only been spreading for a few months,
that there have not been enough datasets collected and published for experts' use. Second, there may be a pressing need for
scientific researchers to develop a smart gadget for such rapid identification and deep understanding of viral infections using
breast X-ray images, which can show the severity of the problem to the human body. After evaluating the already existing
technology and the challenging scenarios we are facing, it has been determined that approach transfer is feasible and realistic
for this research challenge. This method involves training deep neural networks on photos of pneumonia including infection
in one or both lungs, which could be caused by bacteria, infections, or fungus. The illness produces irritation in the alveoli,
that are small air sacs in the lungs. The respiratory system fill with liquid or pus, making breathing difficult. The process
utilizes transfer learning on well-studied deep learning methods to identify the type of virus, validates the designs on a wide
range of data sets, and then transfers the models and insights learned in training and validation to a new data set in a similar
region, that, in this case, is a new infectious fast-growing illness. The coronavirus, also known as COVID-19, is a type of
virus. The suggested method addresses a difficult challenge in deep learning: how to build a reliable model based on a
limited data set that has not been thoroughly analysed and contains unidentified characteristics.
[2]Deep CNNs indeed acquire higher performance with the use of a huge dataset as compared to a smaller one. Granting
there is an enormous range of infected COVID-19 patients globally, however, the variety of publicly to be had chest X-ray
photos online is insignificant and dispersed. For this reason, the authors of this work have defined a reasonably huge dataset
of COVID-19 infected chest X-ray images even though normal and pneumonia photographs are right away on hand publicly
and applied in this study.
[3]Using a chest X-ray image, the observations apply flexible and powerful deep learning methodologies, including the
employment of six CNN models to predict and identifying whether the patient is untouched or impacted by the condition.
To monitor the accuracy of each version getting learned, GoogLeNet, LeNet, VGG-16, AlexNet, StridedNet, and ResNet-
50 models using a dataset of 28,000 images and a 224x224 decision with 32 and sixty-four batch sizes are used. The study
also uses Adam as an optimizer, giving all of the models an adjusted 1e-four learning rate and a 500-epoch. During the
development of models, both GoogLeNet and LeNet received a 98% accuracy rate, VGGNet-16 received a 97% accuracy
rate, AlexNet and StridedNet models received a 96% average accuracy, and the ResNet-50 model received an 80% accuracy
rate. For total performance training, GoogleNet and LeNet fashions have the highest average accuracy. The six models
identified were possible to perceive and predict a pneumonia illness, which included a normal chest X-ray.
[4]In this paper author classify the three varieties of X-rays, image writer used the ensemble method at some stage in
prediction, each picture is surpassed through the type layer where they checked whether an image is COVID-19, pneumonia,
or ordinary.
[5]The X-rays dataset is separated into training and test subsets in the first phase of the statistical preprocessing level.
Character X-ray pictures are also normalised at this level. For a strong version development, data augmentation of training
Copyright to IJARSCT DOI: 10.48175/IJARSCT-3851 772
www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 2, Issue 3, May 2022

Impact Factor: 6.252
photos was completed. The second degree is the age of a version that divides x-rays into two categories (consolidation and
non-consolidation). The use of an okay-fold cross-validation strategy (where okay is the amount of folds) was proven by
the correctness of this model. Finally, the deployment of the explainable AI strategy that we chose to improve the
comprehensibility of our Machine is the final stage. The arrival of the heatmap. We use one-of-a-kind ways to assess the
quality of our Machine: 1) generating the heatmap using the simplest version, and 2) generating the heatmap with an
ensemble of styles with the same structure but educated with one-of-a-kind record folds. The second technique allows us to
compute an uncertainty level (provided by the same old deviation) for each pixel, allowing us to assess the heatmap's
robustness.
[6]The author recommends a current technique based on an ensemble of RetinaNet and masks R-CNN in this paper.
Pneumonia regions are known by their size and ratio (Figs. 3 and 4), with a mean top of 304 pixels (29.6% of the photograph
peak) and a median width of 219 pixels (21.three percent of picture width). Pneumonia appears in a very small area of a
chest x-ray, making it a difficult assignment for modern item detectors. Because FPN provides multi-scale feature maps
with higher first-rate statistics than the typical feature pyramid, we employed it as the backbone of both approaches to solve
this problem. As shown in Fig. 1, the FPN structure blends low-resolution, i.e. linguistically having in - depth, with enhanced
capacities. i.e. semantically weak functions, through a top-down pathway and lateral connections. In comparison to the
primary stacked convolutional layers, we employed residual networks as a base spine version because they reduced the
effect of the degrading hassle and allowed us to develop deeper designs.

III. BACKGROUND
Machine learning (ML) algorithms have steadily gained the interest of researchers over the last few years. This type of
method can make full use of the massive ability to create computer calculators in image processing with pre-determined
algorithm stages. Traditional machine learning methods for dividing jobs, on the other hand, necessitate the use of manual
design algorithms or the manual setting of output layers to separate images. In response to the aforementioned situation,
LeCun et al. [7]offered a CNN approach, which can automatically extract features with the use of constantly stacking
features and exit that the included photos may not be in any class. The shallow networks are very deep and concentrate on
the image's low-level features. CNN the model increasingly exposes advanced features as the number of network layers
increases. CNN learns the distinctions between different images by combining and evaluating these priority features, and it
uses a back-propagation technique to update and record learned parameters. CNN's concept is to use a specific convolution
kernel to filter a prior picture or map component to build the next layer map element, as well as merge functions like merging
functions to minimise feature map scale and mitigation to count. The created component is then given the non-line activation
function.
A mapping to improve the model's simulation capabilities The most common integration tasks include mid and high
integration. The plural of integration denotes that the element sent to the integration layer is split into many regions, with
each sub-region having a different size in terms of horizontal and vertical steps. The sole distinction between high and
medium integration is a lower region where the aggregation rate yields the average of each sub-region. ReLU (Rectified
Linear Units) and Sigmoid are two common activation events. Image elements are automatically extracted using
segmentation and a continual accumulation of convolutional processes, integration functions, indirect opening functions,
and other completely integrated layers. Then, by evaluating these derived characteristics, it is possible to extract pneumonia
from the photos processed by the model. The model's general capacity is increased by fully utilising pixel-level image
information. The most prominent neural framework has been proposed in past few decades for in-depth learning
development, such as AlexNet [8] and VGGNet [9]. However, when the number of layers in the network increase, Instead
of learning the numerous productive features, the neural network will be modified to particular parts of the training image,
which makes the model similar to the capacity declines and creates congestion. The remaining communication framework
was proposed to overcome the problem of network depth. Since then, neural networks have advanced, garnered a lot of
attention and research, and have formed the foundation for a lot of occupations. We also looked at the efficiency of residual
connections in our reduced CNN architecture with only a few layers in this study. S. L. Bangare et al. [10-17] have worked
in the brain tumor detection. N. Shelke et al [18] given LRA-DNN method. Suneet Gupta et al [19] worked for end user
system. Gururaj Awate et al. [20] worked on Alzheimers Disease. P. S. Bangare et al [21] worked on the object detection.

Copyright to IJARSCT DOI: 10.48175/IJARSCT-3851 773

www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 2, Issue 3, May 2022

Impact Factor: 6.252
Kalpana Thakare et al [22-27] have worked on various machine learning algorithms. M. L. Bangare et al. [28] worked on
the cloud platform.

IV. MATERIALS AND METHODS

4.1 Data
The proposed database, which will be used to test the model's performance, comprises a total of 5863 X-ray pictures via
Kaggle. Dr. Paul Mooney created a Kaggle contest in 2017 to classify viral and bacterial pneumonia. It differs from the
other datasets since it contains 5,863 paediatric photos. We're talking about the updated version of this dataset.[6]
The database is further divided into three folders (train, test, and val) with subfolders for each image category (Pneumonia
/ General). Figure 1 shows a few instances of common and pneumonia photos that have been scaled to a static size. Due to
the low amount of exposure in patients, chest X-ray images always show symptoms of limited brightness, and chest X-ray
images always have black, white, and grey pants. The lungs are on both sides of the thoracic cavity, and the lung area is
plainly visible on an X-ray since it is virtually black. The heart, which is situated between the lungs, appears practically as
white as X-rays can go through it entirely. Because bones are comprised of protein and are exceedingly dense, X-rays cannot
pass through them, leaving the bones virtually white. Furthermore, the bones have distinct edges.

(a) Normal Cases

(b) Pneumonia Cases

Figure 1: Examples from the dataset

4.2 Data Preprocessing

Table 1 lists the tactics employed throughout this article. Rescale is a value that we will multiply the data by before any
other processing in our investigation. Our original photos had RGB coefficients ranging from 0 to 255, but values like this
would be too high for our models to handle (given a typical learning rate), so we scale them down by a factor of 1./255.
shear range is used to apply shearing transformations at random. When there are no assumptions of horizontal asymmetry,
zoom range is used to randomly zoom inside photographs, and horizontal flip is used to randomly flip half of the images
horizontally (e.g. real-world pictures). Data pre-processing techniques used in this study
Rescale 1./255

Zoom Range 0.2

Shear Range 0.2

Horizontal_Flip True
Table 1

Copyright to IJARSCT DOI: 10.48175/IJARSCT-3851 774

www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 2, Issue 3, May 2022

Impact Factor: 6.252
4.3 Proposed Network
In this study, we designed a VGG-based CNN model to extract the features of chest X-ray images and use those features
to detect if a patient suffers from pneumonia. In Our CNN Architecture, We started with a lower filter setting of 32 and
worked our way up layer by layer. A layer of Conv2D was used to build the model, followed by a layer of MaxPooling. An
odd number, such as 3x3, is desirable for kernel size. The activation functions Tanh, ReLU, and others can be employed,
but ReLU is the most popular. input shape accepts the width and height of an image, with the last dimension serving as a
colour channel. After that, we flattened the input and added ANN layers.

f(x)=max(0,x)
S(x) = Sigmoid
f(x) = ReLU
For the last layers (ANN Layers), I used a softmax activation function and defined units as the total number of classes. For
binary classification, I used a sigmoid and set the unit to 1.

Figure 2: Details of proposed DL model.

V. IMPLEMENTATION
Convolutional neural networks are a type of neural network that shares all of the characteristics of other neural networks.
CNN, on the other hand, was created specifically to process input images. The architecture of their organisation is therefore
more specific: it is made up of two primary parts.

5.1 Conv Layers

Because it works as a feature extractor, the first block establishes the uniqueness of this sort of neural network. It
accomplishes this by adding convolution filtering processes to template matching. The first layer uses many convolution
kernels to filter the image and produce "feature maps," which are then normalised (using an activation function) and/or
shrunk.

5.2 Pool Layers

The second block is not unique to CNNs; in fact, it appears at the end of all categorization neural networks. To return a
new vector to the output, the input vector values are transformed (using various linear combinations and activation
functions). The chance that the image corresponds to class I is represented by element I of this last vector, which has as
many elements as there are classes. As a result, each element has a value between 0 and 1, and the total value is 1. The last
layer of this block (and hence of the network) calculates these probabilities using a Sigmoid function (binary classification)
or a RELU function (multi-class classification) as an activation function.

Copyright to IJARSCT DOI: 10.48175/IJARSCT-3851 775

www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 2, Issue 3, May 2022

Impact Factor: 6.252

Copyright to IJARSCT DOI: 10.48175/IJARSCT-3851 776

www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 2, Issue 3, May 2022

Impact Factor: 6.252
5.3 Fitting Model
EarlyStopping is a function that allows you to halt the epochs early based on a measure. It aids in the avoidance of
overfitting the model. We're urging you to stop based on the val loss statistic since it needs to be as low as possible.
Patience states that once a minimum val loss is attained, if the val loss grows in any of the next three iterations, the
training will end at that epoch. EarlyStopping in our training session came to a halt in the 9th epoch, with val loss = 39.8%
and val accuracy = 68.7%. when the patience level is 3
Actual test results were 91.98 percent accurate on the model.

 This provides you a percentage estimate of the individual image, which you may load straight from your hard drive
by specifying its path.
 After importing the image as we did previously, we must recreate all of the data pretreatment procedures in order
to input the test set into the model and obtain a forecast. Importing the tensorflow. keras.preprocessing. image class
is required for pre-processing.
 Import an image with dimensions of (500,500) and a grayscale colour channel.
 To predict the case, convert the image to an array, rescale it by dividing it by 255, and extend dimension by axis =
0 as shown earlier.

VI. CONCLUSION AND FUTURE WORK

This paper presents an automated diagnosis of pneumonia in X-ray scans using deep CNN. The research was conducted
utilising the X-Rayscan dataset, which contains 5863 scans. Experiments yielded a variety of scores, including accuracy,
recall, precision, and AUCranking, proving the efficiency of our network model. The proposed framework was successful
in reaching a 91 percent categorization accuracy. To improve the model's efficiency, hyper-parameter optimizations were
examined, and multiple optimization techniques, such as stochastic gradient descent, Adagrad, and Adamoptimizer, were
used. The customizedVGG16 model's effectiveness in detecting pneumonia reveals that the model surpasses other
optimizers when compared to Adam.In the future, this research will be broadened to include the detection and differentiation
of multi-class X-ray images. The efficiency could also be improved by applying more complex feature extraction
approaches based on many recently established deep learning models for biomedical picture segmentation.

REFERENCES
[1]. Vandecia Fernandes et al., “Bayesian convolutional neural network estimation for pediatric pneumonia detection
and diagnosis”, Computer Methods and Programs in Biomedicine, Elsevier, 2021

Copyright to IJARSCT DOI: 10.48175/IJARSCT-3851 777

www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 2, Issue 3, May 2022

Impact Factor: 6.252
[2]. Hongen Lu et al., “Transfer Learning from Pneumonia to COVID-19”, Asia-Pacific on Computer Science and
Data Engineering (CSDE), 2020 IEEE
[3]. Sammy V. Militante et al., “Pneumonia and COVID-19 Detection using Convolutional Neural Networks”, 2020
the third International on Vocational Education and Electrical Engineering (ICVEE), IEEE, 2021
[4]. Nanette V. Dionisio et al., “Pneumonia Detection through Adaptive Deep Learning Models of Convolutional
Neural Networks”, 2020 11th IEEE Control and System Graduate Research Colloquium (ICSGRC 2020), 8
August 2020
[5]. Md. Jahid Hasan et al., “Deep Learning-based Detection and Segmentation of COVID-19 & Pneumonia on Chest
X-ray Image”, 2021 International Information and Communication Technology for Sustainable Development
(ICICT4SD), 27-28 February 2021
[6]. https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia
[7]. LeCun, Y.; Boser, B.; Denker, J.S.; Henderson, D.; Howard, R.E.; Hubbard, W.; Jackel, L.D. Backpropagation
applied to handwritten zip code recognition. Neural Comput. 1989, 1, 541–551.
[8]. Krizhevsky, A.; Sutskever, I.; Hinton, G.E. Imagenet classification with deep convolutional neural networks. Adv.
Neural Inf. Process. Syst. 2012, 25, 1097–1105.
[9]. Simonyan, K.; Zisserman, A. Very deep convolutional networks for large-scale image recognition. arXiv 2014,
arXiv:1409.1556.
[10]. S. L. Bangare, “Classification of optimal brain tissue using dynamic region growing and fuzzy min-max neural
network in brain magnetic resonance images”, Neuroscience Informatics, Volume 2, Issue 3, September 2022,
100019, ISSN 2772-5286, https://doi.org/10.1016/j.neuri.2021.100019 .
[11]. S. L. Bangare, G. Pradeepini, S. T. Patil, “Implementation for brain tumor detection and three dimensional
visualization model development for reconstruction”, ARPN Journal of Engineering and Applied Sciences
(ARPN JEAS), Vol.13, Issue.2, ISSN 1819-6608, pp.467-473. 20/1/2018
http://www.arpnjournals.org/jeas/research_papers/rp_2018/jeas_0118_6691.pdf
[12]. S. L. Bangare, S. T. Patil et al, “Reviewing Otsu’s Method for Image Thresholding.” International Journal of
Applied Engineering Research, ISSN 0973-4562, Volume 10, Number 9 (2015) pp. 21777-21783, © Research
India Publications https://dx.doi.org/10.37622/IJAER/10.9.2015.21777-21783
[13]. S. L. Bangare, G. Pradeepini, S. T. Patil, “Regenerative pixel mode and tumor locus algorithm development for
brain tumor analysis: a new computational technique for precise medical imaging”, International Journal of
Biomedical Engineering and Technology, Inderscience, 2018, Vol.27 No.1/2.
https://www.inderscienceonline.com/doi/pdf/10.1504/IJBET.2018.093087
[14]. S. L. Bangare, A. R. Khare, P. S. Bangare, “Quality measurement of modularized object oriented software using
metrics”, ICWET '11: Proceedings of the International Conference & Workshop on Emerging Trends in
Technology, February 2011, pp. 771–774. https://doi.org/10.1145/1980022.1980190.1.
[15]. S. L. Bangare, G. Pradeepini and S. T. Patil, "Brain tumor classification using mixed method approach," 2017
International Conference on Information Communication and Embedded Systems (ICICES), 2017, pp. 1-4, doi:
10.1109/ICICES.2017.8070748.
[16]. S. L. Bangare, S. Prakash, K. Gulati, B. Veeru, G. Dhiman and S. Jaiswal, "The Architecture, Classification, and
Unsolved Research Issues of Big Data extraction as well as decomposing the Internet of Vehicles (IoV)," 2021
6th International Conference on Signal Processing, Computing and Control (ISPCC), 2021, pp. 566-571, doi:
10.1109/ISPCC53510.2021.9609451.
[17]. S. L. Bangare, G. Pradeepini, S. T. Patil et al, “Neuroendoscopy Adapter Module Development for Better Brain
Tumor Image Visualization”, International Journal of Electrical and Computer Engineering (IJECE) Vol. 7, No.
6, December 2017, pp. 3643~3654. http://ijece.iaescore.com/index.php/IJECE/article/view/8733/7392
[18]. Shelke, S. Chaudhury, S. Chakrabarti, S. L. Bangare et al. “An efficient way of text-based emotion analysis from
social media using LRA-DNN”, Neuroscience Informatics, Volume 2, Issue 3, September 2022, 100048, ISSN
2772-5286, https://doi.org/10.1016/j.neuri.2022.100048 .
[19]. Suneet Gupta, Sumit Kumar, Sunil L. Bangare, Shibili Nuhmani, Arnold C. Alguno, Issah Abubakari Samori,
“Homogeneous Decision Community Extraction Based on End-User Mental Behavior on Social Media”,
Copyright to IJARSCT DOI: 10.48175/IJARSCT-3851 778
www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 2, Issue 3, May 2022

Impact Factor: 6.252
Computational Intelligence and Neuroscience, vol. 2022, Article ID 3490860, 9 pages, 2022.
https://doi.org/10.1155/2022/3490860.
[20]. Gururaj Awate, S. L. Bangare, G. Pradeepini and S. T. Patil, “Detection of Alzheimers Disease from MRI using
Convolutional Neural Network with Tensorflow”,arXiv, https://doi.org/10.48550/arXiv.1806.10170
[21]. P. S. Bangare, S. L. Bangare, R. U. Yawle and S. T. Patil, "Detection of human feature in abandoned object with
modern security alert system using Android Application," 2017 International Conference on Emerging Trends &
Innovation in ICT (ICEI), 2017, pp. 139-144, doi: 10.1109/ETIICT.2017.7977025
[22]. Kalpana S. Thakare, Viraj Varale, “Prediction of Heart Disease using Machine Learning Algorithm”, Bioscience
Biotechnology Research Communications (Special issue) Volume 13, Issue 12, 2020 (Dec 2020 issue).
[23]. Kalpana S. Thakare, A. M. Rajurkar, “Shot Boundary Detection of MPEG Video using Biorthogonal Wavelet
Transform”, International Journal of Pure and Applied Mathematics, Volume 118, No. 7, pp. 405-413, ISSN:
1311-8080 (printed version); ISSN: 1314-3395 (on-line version), url: http://www.ijpam.eu
[24]. Kalpana S. Thakare, A. M. Rajurkar, R. R. Manthalkar, “Video Partitioning and Secured Key frame Extraction
of MPEG Video”, Proceedia Computer Science Journal, Volume 78, pp 790-798, Elsevier, 2016. Scopus DOI:
htpp://10.1016/j.procs.2016.02.058, www.sciencedirect.com/science/article/pii/S1877050916000600
[25]. Kalpana S. Thakare, A. M. Rajurkar and R. R. Manthalkar, “Content based Video Retrieval using Latent Semantic
Indexing and Color, Motion and Edge Features”, International Journal of Computer Applications 54(12):42-48,
September 2012, Published by Foundation of Computer Science, New York, USA. DOI: 10.5120/8621-2486
[26]. Kalpana S. Thakare, Archana M. Rajurkar, R. R. Manthalkar, “A Comprehensive System Based on
Spatiotemporal Features Such as motion, Quantized Color and Edge Features”, International Journal of Wireless
and Microwave Technologies (IJWMT) ISSN 1449 (Print), ISSN: 2076-9539 (Online), Vol.1, No.3, June. 2011,
DOI: 10.5815 /ijwmt
[27]. Kalpana S. Thakare, Archana M. Rajurkar, Dr. R. R. Manthalkar, “An effective CBVR system based on Motion,
Quantized color and edge density features”, International Journal of Computer Science & Information Technology
(IJCSIT), ISSN 0975 – 3826, Vol 3, No 2, April 2011 DOI: 10.5121/ijcsit.2011.3206 78.
[28]. M. L Bangare, Sarang A Joshi, “Kernel interpolation-based technique for privacy protection of pluggable data in
cloud computing”, International Journal of Cloud Computing, Volume 9, Issue 2-3, pp.355-374, Publisher
Inderscience Publishers (IEL)

Copyright to IJARSCT DOI: 10.48175/IJARSCT-3851 779

www.ijarsct.co.in

View publication stats