PosterPDF Available

WeboDoc: A Web Based Application for Classifying Pneumonia and Malaria Infected Images

Authors:
  • Nossa Data

Abstract

In this work, we trained two convolutional neural network on Malaria and Pneumonia images and embed it into a web application that can run locally. This application can help health practitioners to quickly detect pneumonia and malaria infected patients. Malaria images are in the form of microscopic slide smears while pneumonia images are from chest Xrays. We show that our application can be infused into any hospital’s existing system and can be useful in areas with scarce health personnel and limited resources.
WeboDoc: A Web Based Application for Classifying
Pneumonia and Malaria Infected Images
Rising O. Odegua, Nelson Ogbeide, Ojeabulu O. Gift, Comfort Igboko, Caleb Emelike, Precious Cadeton
Ambrose Alli University, Computer Science Department
Fig.3 Comparison of Bagging
and Boosting Ensembles
Fig.4 Comparison of Boosting
and Stacking Ensemble Fig.5 Comparison of all models
Single Models
Bagging Boosting
Stacking
Boosting
Stacking
Bagging
Boosting
Fig.2 Comparison of Single and
Bagging Ensemble
Linear Regression
KNN
RandomForest
ExtraTrees
DATASETS
RESULTS
METHODOLOGY
We obtained two data sets comprising of pneumonia chest x-ray
images (anterior-posterior) and malaria cell images, trained a
convolutional neural network from scratch to automatically detect
these infections, and infuse the trained models into a web
application that can run locally as well as online.
We start by processing our dataset, thereby splitting it into
two folds, train and validation. and performed some resized
pneumonia images to 150 x 150 and Malaria images to
100x100
We trained two separate convolutional neural network to
detect infections.
Next, we saved, exported and embed the models into a Django
powered web application with user interface that allows for
upload and classification of images belonging to any of the
categories
Some key features of our web application are:
It can be extended to include other trained classifiers/models.
It has an easy to use interface.
Ability to run locally as well as online.
DATASETS
S
/
N
Features Features Size
Num . Cat. after
encoding
112 9 63
214 10 67
39 2
45 - -
512 18 130
626 54 221
78 11 95
82 8 69
94 8 33
CONCLUSION
In this work, we trained two convolutional neural network on
Malaria and Pneumonia images and embed it into a web
application that can run locally. This application can help
health practitioners to quickly detect pneumonia and malaria
infected patients. Malaria images are in the form of
microscopic slide smears while pneumonia images are from
chest X rays. We show that our application can be infused into
any hospital’s existing system and can be useful in areas with
scarce health personnel and limited resources.
Single Base Moels Bagging Boosting Stacking
DATASET
S
LR KN
N
R
F
ET LR_B
G
KNN_
BG
RF_
BG
ET_
BG
GB AD
AB
LGB
STACKE
D
MODEL
Loan_Dat
a0.32
1
0.32
4
0.32
8
0.32
8
0.319 0.314 0.311 0.3
2
0.303 0.35
5
0.293 0.3
House_Pric
ing 2.45
7
0.16
4
0.09
7
0.08
9
0.082 0.161 0.094 0.084 0.08 0.10
2
0.079 0.079
Tweets_Da
ta 6.50
3
5.70
6
6.13
3
6.12
4
6.4 5.787 6.118 6.102 5.822 20.9
37
5.768 5.611
Traffic_Da
ta 15.5 4.83
3
4.36
7
4.39
3
15.13 4.35 4.32 4.251 4.104 4.14
7
4.059 4.027
Avocado_D
ata 0.14
5
0.22
7
0.12
8
0.13
7
0.145 0.225 0.122 0.1
3
0.123 0.21 0.115 0.132
Real_Esta
te 5.65
4
5.31 4.76
8
4.57
1
5.611 5.258 4.315 4.314 4.558 5.42
9
4.62 4.277
AutoMobi
le 484
8.5
364
9.4
225
3.3
219
9.3
4234.
3
3641.1 2217.
8
2101.
1
2184.
8
2209
.5
2199.
9
2101.0
German_C
red 7.55 10.5
44
7.36 8.00
4
7.617 10.256 7.181 7.225 7.725 8.53
1
7.854 7.276
Supermar
ket 75.7
32
81.9
08
80.9
74
86.2
9
75.62
7
80.93 79.41
8
81.09
3
74.80 75.1
6
74.47 71.458
INTRODUCTION AND MOTIVATION
Pneumonia and Malaria has accounted for numerous premature
death especially in developing countries in Africa and Asia where
access to timely and good detection technologies is limited.
WHO estimated that about 4 million death occur annually from air
pollution diseases including pneumonia [3] and over 150 million
people get infected annually especially children under 5 years old
[4]. Malaria on the other hand was reported to cause approximately
438,000 deaths from 214 million infections in 2015 [5].
Automating the detection of infected patients with either
pneumonia or malaria will ensure accurate diagnosis and will
greatly improve health care in resource-scarce areas like Africa.
REFERENCES
CNN ARCHITECTURE
Malaria Cell Images
size : 27,558
Classes: 2
Channel: 3
Sample Images
Fig 2. Infected Malaria Cell
Fig 3. Uninfected Malaria Cell
Infected (Class 1)
Uninfected (Class 0)
Fig 6. CNN Architecture for Pneumonia Chest X-ray Model
Infected (Class 1)
Uninfected (Class 0)
Fig 7. CNN Architecture for Malaria Cell Model
Pneumonia Chest X-rays
size : 5,863
Classes: 2
Channel: 3
Sample Images
Fig 4. Infected Pneumonia Image
Fig 5. Uninfected Pneumonia Image
RESULTS
[1] Rajaraman S, Antani SK, Poostchi M, Silamut K, Hossain MA, Maude, RJ, Jaeger S,
Thoma GR. (2018) Pre-trained convolutional neural networks as feature extractors toward
improved Malaria parasite detection in thin blood smear images. PeerJ6:e4568
https://doi.org/10.7717/peerj.4568
[2] Kermany, Daniel; Zhang, Kang; Goldbaum, Michael (2018), “Labeled Optical
Coherence Tomography (OCT) and Chest X-Ray Images for Classification”, Mendeley
Data, v2
[3] World Health Organization, Household Air Pollution and Health [Fact Sheet], WHO,
Geneva, Switzerland, 2018, http://www.who.int/newa-room/fact-sheets/detail/household-
airpollution-and-health.
[4] I. Rudan, L. Tomaskovic, C. Boschi-Pinto, and H. Campbell, “Global estimate of the
incidence of clinical pneumonia among children under five years of age,” Bulletin of the
World Health Organization, vol. 82, pp. 85–903, 2004
[5] Chan M. World Malaria Report. Geneva: World Health Organization; 2015
Fig 8. Training,validation and confusion matrix of Pneumonia Classification Model
5
Fig 9. Training,validation and confusion matrix of Malaria Cell Classification Model
Fig 1. Interface of our prediction application (webodoc)
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.