Diagnosis of Tuberculosis Using Combined Blur and Affine Moment Invariants

ISSN 2394-3777 (Print)
ISSN 2394-3785 (Online)

Available online at www.ijartet.com
International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)

Vol. 4, Issue 12, December 2017
Diagnosis of Tuberculosis using Combined Blur and

Affine Moment Invariants
K. Durga Prasad1, Dr. M.B. Ramana Murthy2, Dr. A. Venkataramana3
Associate Professor, E.C.E Dept., J.P.N. College of Engg , Mahabubnagar, Telangana,India1
mail ID: dpkolluru@gmail.com
Professor , E.C.E Dept. , IIIT Basara, Telangana , India 2
Lecturer in ECE, Q.Q. Government Polytechnic, Hyderabad, India 3
Abstract: Tuberculosis is a major infectious disease in many regions of the World. Detection of Tuberculosis is the most
efficient solution for prevention of this disease. Chest radiography is the most common method adopted for screening this
disease and the success of this method depends on the experience and interpretation of the skilled radiologist. This paper
presents Computer Aided Diagnosis (CAD) system for detection of Tuberculosis using Combine Blur and Affine Moment
Invariants (CBAMI). The CAD method not only provides second opinion to radiologist, but also accelerates the process of
active case finding. The CBAMI are selected as features in this paper because they are invariant to both affine distortions as
well as blur. Simulation results are carried out by considering The Montgomery County X ray dataset and the results are
compared with Geometric Moment Invariants. It is observed from the results that the proposed method provided better
accuracy.
Keywords: Tuberculosis, Computer Aided Diagnosis , Combine Blur and Affine Moment Invariants, Geometric Moment
Invariants
recommended chest radiography (CXR) for screening

I. INTRODUCTION proposes for early detection of TB. Early detection leads to
Tuberculosis (TB) is one of the public health problem success of anti TB therapy. It also helps to control over
which causes major health threat worldwide. The TB is an transmission of infection as well as development of drug
infectious disease caused by bacillus mycobacterium resistant TB. CXR diagnostic method improves TB detection
tuberculosis. It affects the lung (pulmonary TB) portions. It [3,4] can be used for screening large population and can
is a communication disease [1] which spreads through identify active TB cases with a reasonable amount of time.
coughing and sneezing of the infected person. If the person This method is also low cost effective manner. Further, in
is living with HIV, the probability to attach TB disease is areas having limited resources, the confirmatory diagnostic
very high. The most efficient solution for prevention of this test like sputum smear test are not available , CBX can be
disease is to detect its presence as early as possible in the used as most powerful diagnostic tool for TB detection.
affected person. There are different methods available for Especially in developing countries, because of lack of skilled
screening of TB namely (i) The Sputum smear microscopy radiologists and other resources, it becomes difficult for
(ii) Tuberculin skin test (iii) Rapid molecular test and (iv) effective diagnosis. Hence, Computer Aided Diagnosis
Chest radiography test. Sputum smear microscopy is a (CAD) tool can gain lot of significance because they not
screening method in which sputum samples are examined only reduce diagnostic errors but also increases the
under a microscope to see if bacteria are present or not. Its efficiency of mass screening in poor resource setting. CAD
sensitivity is reduced in patients with HIV co-infection [2]. method provides second opinion to the radiologists for their
In Tuberculin skin test, a small amount of tuberculin is finding. It provides better diagnosis of cancer and other
injected into the skin and the size of swelling is observed. diseases including TB.
This method is not reliable as it causes misclassification [1].
Latest rapid molecular tests are fast and accurate. But, their In the literature, for object recognition applications,
availabilities and cost is a big primary concerned. The WHO Histogram of Oriented Gradients (HOG) are used as
All Rights Reserved © 2017 IJARTET 20


features. Sivaranjani [6] proposed a method for analysis of II. OVERVIEW ON MOMENT INVARIANTS
TB with HOG as features and SVM as classifier. Ankitha et In this section, we present details about Geometric,
al [7] proposed a method to extract the lung field by an moments, Geometric Moment Invariants and Combined
interactive lung field segmentation based on active contour Blur and Affine Moment Invariants.
models. After segmentation process, they analyzed the pixel
data within region of interest using first order statistics to 2.1 Geometric Moments and Their Invariants
extract textural properties in classifying image as TB and
non TB. A method using HOG Features in Computer-Aided The Geometric moments of order (p, q) for an image
Diagnosis of Tuberculosis without Segmentation was f ( x, y ) are defined [10] as
proposed in [8]. Pallavi et al. [9] used first order statistical
properties such as mean and entropy for diagnosis of TB.  
Hu [10] introduced geometric moments based on monomials

M pq  
 
x p y q f ( x, y ) dx dy (1)
. He derived a set of seven moment invariants which are

invariant with respect to change in image translation, where p,q  0,1, 2, ........., 
rotation and scale. Jan Flusser et al [11] derived blur
invariants which are invariant with respect to blurring on the To keep the dynamic range of M pq consistent for different
acquired image. Blur invariants are required because real
imaging systems as well as imaging conditions are size images, the N x M image plane is first mapped onto a
imperfect. Blur may occur in the captured image during the
image acquisition process because of the factors such as lane square defined by x [1,  1], y [1,  1] . Then M pq
aberration, diffraction, wrong focus and atmospheric
can be written as
turbulence. They proposed an approach which describes
images by features that are invariant with respect to blur and 1 1
recognize images in the feature space. In order to obtain M pq 
features which are invariant to blur as well as affine

1 1
x p y q f ( x, y ) dx dy (2)
distortion, The same authors [12] derived combined blur and

affine moment invariants. Combined invariants are capable The discrete form representation of the above expression is
of recognizing objects in the degraded scene without any given by
restoration.
N 1 M 1
Most of the methods discussed above for diagnosis of M pq    xi y j f ( xi , y j ) x y

p q
(3)
i 0 j 0
Tuberculosis is based on histogram of oriented features,
calculation of mean, variance, third moment and entropy
where ( xi , y j ) is the centre of (i, j) pixel and
features as well Geometric moment invariants. All these
methods do not take care into account of blur as well as
affine distortion present in the Tb X-ray image. In order to x  xi  xi 1 , y  y j  y j 1 are the sampling intervals
solve this problem and to obtain better accuracy, we propose
in the ‘x’ and ‘y’ directions respectively. N x M
an efficient method for detection of Tuberculosis in this
Paper. The proposed method computes Combined Blur and corresponds to the size of the image.
Affine Moment Invariants (CBAMI) on X-ray image and
uses them as feature for detection of tuberculosis. The Geometric central moments [10] are defined as
 
  ( xx
This paper is organized into five sections. Details about the
Geometric moment invariants as well as CBAMI are  pq  
) p ( y  y ) q f ( x, y ) dx dy (4)
presented in section II. Proposed method is presented in 
section III. Simulation results are presented in section IV.

The last section presents the conclusions about the work.


M 10 M 01 restoration. The Combined Blur and Affine Moment

where x  
and y  (5) Invariants (CBAMI) are given below [12].
M 00 M 00 Set1 = µ230 µ203 – 6 µ30 µ21 µ12 µ03
+ 4 µ30 µ312 + 4 µ321 µ03 – 3 µ221 µ212)/ µ1000.
For a digital image, the Geometric central moments can be
(15)
written as
Set2 = µ250 µ205 – 10 µ50 µ41 µ14 µ05 +4 µ50 µ32 µ23 µ05
N 1 M 1
 pq   ( xi  x ) p ( y j  y ) q f ( xi , y j ) x y
+ 16 µ50 µ32 µ214 – 12 µ50 µ14 µ223 + 16 µ241 µ23 µ05
(6)
+ 9 µ241 µ214 – 12 µ41 µ232 µ05 -72 µ32 µ41 µ14 µ23
i 0 j 0
+ 48 µ323 µ41 +48 µ14 µ332 - 32 µ232 µ223) / µ1400 .
The normalized central moments [10] are defined as (16)
Set3 = (µ230µ05 µ12 - µ230µ03 µ14 - µ30µ221 µ05
 pq
 pq  
(7) - 2 µ30 µ21 µ12 µ14 +4 µ30 µ21 µ23 µ03 + 2 µ30µ212 µ23
00 – 4 µ30 µ12 µ32 µ03 + µ30µ203 µ41 + 3 µ321 µ14
- 6 µ221µ12 µ23 – 2 µ221µ03 µ32 + 6 µ21µ212 µ32
pq + 2 µ12 µ21 µ41 µ03 - µ21µ203 µ50 – 3 µ312 µ41+ µ212µ03 µ50) /
where   1 for p  q  2, 3,..... µ1100.
2 (17)
Hu [10] derived moment invariants using Geometric Set4 = (2 µ30 µ12 µ41 µ05 - 8 µ30 µ12 µ32 µ14 + 6 µ30 µ12 µ223
moments which are invariant to image rotation, translation -µ30 µ03 µ50 µ05 + 3 µ30 µ03 µ41 µ14 – 2 µ30 µ03 µ23 µ32
- 2 µ221µ41 µ05 + 8 µ221µ32 µ14 - 6 µ221 µ223 + µ21 µ12 µ50 µ05
and scale. These invariants are given below [10]. - 3 µ21 µ12 µ41 µ14 +2 µ21 µ12 µ32 µ23 + 2 µ21 µ03 µ50 µ14
– 8 µ21 µ03 µ41 µ23 +6 µ21 µ03 µ232 -2 µ212µ50 µ14 +8 µ212µ41
Ø1 = Ƞ20 + Ƞ02 (8) µ23 – 6 µ212 µ232) / µ1200. (18)
Ø2 = (Ƞ20 - Ƞ02)2 + 4 Ƞ211 (9)

Set5 = (µ30 µ41 µ23 µ05 - µ30 µ41 µ214 - µ30µ232 µ05 +2 µ30 µ32
(10) µ23 µ14
Ø3 = (Ƞ30 -3 Ƞ12)2 + (3Ƞ21 - Ƞ03)2
- µ30 µ323 - µ21 µ50 µ23 µ05 + µ21 µ50 µ214 + µ21 µ41 µ32 µ05
- µ21 µ41 µ23 µ14 - µ21µ232 µ14 + µ21 µ32 µ223 + µ12 µ50 µ32 µ05
(11)
Ø4 = (Ƞ30 + Ƞ12)2 + (Ƞ21 + Ƞ03)2 - µ12 µ50 µ23 µ14 - µ12µ241 µ05 + µ12 µ41 µ32 µ14 + µ12 µ41 µ223
- µ12µ232 µ23 - µ03 µ50 µ32 µ14 + µ03 µ50 µ223 + µ03µ241 µ14
Ø5 = (Ƞ30 -3 Ƞ12) (Ƞ30 + Ƞ12)[ (Ƞ30 - Ƞ12)2 – 3 (Ƞ21 - Ƞ03)2 -2 µ03 µ41 µ23 µ32 + µ03 µ332) / µ1300. (19)
+ (3Ƞ21 + Ƞ03) (Ƞ21 + Ƞ03) [ 3(Ƞ30 - Ƞ12)2 - (Ƞ21 + Ƞ03)2 ]
(12)
Ø6 = (Ƞ20 - 3 Ƞ02) [(Ƞ30 + Ƞ12)2 - (Ƞ21 + Ƞ03)2 + 4 (Ƞ30 + Ƞ12)

Set6 = (µ270 µ207 – 14 µ70 µ61 µ16 µ07 + 18 µ70 µ5 2 µ25 µ07
(Ƞ21 + Ƞ03) (13)
+ 24 µ70 µ52 µ216 – 10 µ70 µ43 µ34 µ07 – 60 µ70 µ43 µ25 µ16
–234 µ61 µ52 µ25 µ16 + 40 µ61µ243 µ07 + 50 µ61 µ43 µ34 µ16
Ø7 = (3Ƞ21 - Ƞ03) (Ƞ30 + Ƞ12) (Ƞ30 - Ƞ12)2 -3(Ƞ21 + Ƞ03)2 +
+ 360 µ61 µ43 µ225 – 240 µ61µ234 µ25 + 360 µ252µ34 µ16
(Ƞ21 + Ƞ03) (Ƞ21 + Ƞ03)[3 (Ƞ30 + Ƞ12)2 - (Ƞ21 + Ƞ03)2
(14)
+ 81 µ252 µ225 - 240 µ52µ243 µ16 – 990 µ52 µ43 µ34 µ25
+ 600µ52 µ334 +600µ343 µ25 – 375 µ243 µ234 ) / µ1800. (20)
2.2 Combined Blur and Affine Moment Invariants
Jan Flusser et al. [12] derived combined blur and affine
moment invariants. Combined invariants are capable of
recognizing objects in the degraded scene without any


III. PROPOSED METHOD
The proposed technique for detection of TB in CXR is

shown in Fig.1.
The first step is proposing which is used to remove

background noise. In this work, we used median filtering for Image 1 Image 2
smoothing. Combine blur and affine moment invariants are
computed on preprocessed image. Combine blur and affine
moment invariants are selected as features because they are Fig.2 Some sample images used in the simulations .
invariant with respect to image blur and affine distortions. In
presence of these distortions also, the computed features are
invariants. The extracted features are then applied for In order to test the invariance property of CBAMI under
classification with and without TB features using Euclidean different distortions, we selected image as shown in Fig.3.
distance classifier. Finally, the result is evaluated by expert
radiologist.
IV. SIMULATION RESULTS
In order to test the proposed approach, we considered The

Montgomery County dataset. This dataset is collected online
freely from the Department of Health and Human Services,
Montgomery County, Maryland, USA. It consists of 138 Original Test Image Original Image Scaled by 50%
frontal chest X-rays from Montgomery Countys
Tuberculosis screening program, out of which 80 are normal
cases and 58 are cases with manifestations of TB. These X-
rays were captured with a Eureka stationary X-ray machine
(CR), and are provided in Portable Network Graphics (PNG)
format as 12-bit gray level images. The size of the X-rays is
either 4,020x4,892 or 4,892x4,020 pixels. These images are
resized to 450 x 450 pixels. Some of the database images are
shown in Fig.2.
Original Image rotated by 200 Original Image Blurred
CBAMI Original Half Rotated Blurred

Image size by 20 by LPF
degrees
Image 1 Image 2
Set1 5.58e-22 5.58e-22 5.58e-22 5.59e-22
Fig 1
Set2 5.03e-32 5.02e-32 5.03e-32 5.02e-32
Set3 -5.27e-26 -5.27e-26 -5.25e-26 -5.27e-26
Set4 -5.46e-27 -5.46e-27 -5.46e-27 -5.46e-27
Set5 -1.14e-31 -1.14e-31 -1.14e-31 -1.13e-31
All Rights Reserved © 2017Set6

IJARTET
2.86e+02 2.83e+02 2.84e+02 2.89e+02 23

Fig.3. Images used for computing CBAMI [4] A. Harries, N. Hargreaves, J. Kwanjana, F. Salaniponi et al.,
“Clinical diagnosis of smear-negative pulmonary tuberculosis: an
audit of diagnostic practice in hospitals in malawi,” The international
journal of tuberculosis and lung disease, vol. 5, no. 12, pp. 1143–
The computed CBAMI are entered in Table.1. 1147, 2001.
[5] J. Day, S. Charalambous, K. Fielding, R. Hayes, G. Churchyard, and
A. Grant, “Screening for tuberculosis prior to isoniazid preventive
therapy among hiv-infected gold miners in south africa,”The
international journal of tuberculosis and lung disease, vol. 10, no. 5,
Table1. CBAMI invariants computed for images shown in Fig.3. pp. 523–529, 2006.
[6] Sivaranjani, “ Analysis of Tuberculosis in chest using SVM
From the results, it is observed that these features are classifier”, National Conference on Research Advances in
Communication, Computation, Electrical Sciences and Structures,
invariant with respect to affine distortions such as rotation, pp. 18-20, 2015
scale, translation as well as blur. The proposed approach is [7] Ankitha B M , Sanjeev Kubakaddi and Dr.Bharathi S H, “Computer
applied on the database images and the the accuracy is Aided Diagnosis of Tuberculosis usingFirst order Statistical
calculated. The accuracy is 92%. For comparision, the same approach”, IPASJ International Journal of Electronics &
Communication (IIJEC), Volume 3, Issue 4, pp. 22-28, April 2015
experiment is repeated with Geometric moment invariants as [8] Arun Chauhan, Devesh Chauhan and Chittaranjan Rout, “Role of
features and observed that the accuracy is 88%. Hence, the Gist and PHOG Features in Computer-Aided Diagnosis of
CBAMI can be effectively used for extracting features of Tuberculosis without Segmentation”, PLOS ONE www.plosone.org
chest X ray images. 1, Volume 9, Issue 11, , November 2014
[9] Pallavi T. P., Praveen Kumar P. S., “ Diagnosis of Tuberculosis by
using first order statistics applied to chest radiograph”, International
Journal of Sciences & Applied Research, IJSAR, 2(8), pp. 16-22,
2015
[10] Hu. M.K, “ Visual Pattern Recognition by Moiments Invariants”,
IRE Transaction on Information Theory, 8, 179-187, 1962
V. CONCLUSIONS [11] Jan Flusser and Tomas suk et al., “ Recognition of Blurred Images
by Method of Moments”, IEEE Transaction on Image Processing,
vol.5, no.3, pp.533-537, March 1996
This paper presents a method for detection of Tuberculosis [12] Tomas Suk and Jan Flusser, “Combined Blur and Affine Moment
Invariants and their use in
using CBAMI as features and Euclidean Distance as
classifier. The proposed method computes Combined Blur BIOGRAPHY
and Affine Moment Invariants (CBAMI) on X-ray image K.Durga Prasad (mail ID:
and uses them as feature for detection of tuberculosis. The dpkolluru@gmail.com) received
CBAMI are selected as features in this paper because they his M.Tech(DSCE) from
are invariant to both affine distortions as well as blur. It is J.P.N.C.E, and B.Tech(ECE)
observed from the simulation results that the detection from NimraCollege of Engg.,
accuracy is better. Our future work will be directed towards Currently he is working as
use of neural network classifier or support vector machines Associate Professor at
for classification. Jayaprakash narayan college of
engineering and has 16 years of Experience in teaching . His
REFERENCES
areas of interest include Digital Image Processing, Pattern
[1] Anuj, Rahul Hooda and Ajay Mittal, “ TB Detection in Chest
Radiograph Using Deep Learning Architecture”, 5th International
Recognition, Electro Magmetic Field,VLSI design, Antenna
Conferenceon Emerging Trends in Engineering, Technology, Science and wave propagation, Communication system .
and Managementpp.136-140,, IETE, August 2017
[2] P. Quinco, S. B uhrer-Śekula, W. Branddao, R. Monte, S. L. Souza,
V. Saraceni, M. Palaci, R. Dietze, and M. Cordeiro-Santos,
“Increased sensitivity in diagnosis of tuberculosis in hiv-positive
patients through the small-membrane-filter method of microscopy,” 2. M B Ramamurthy is in the academic filed for the past 36
Journal of clinical microbiology, vol. 51, no. 9, pp. 2921– 2925, years. Currently he is Professor ECE and Dean Academics
2013. in Jayaprakash Narayan College of Engineering,
[3] J. Pouchot, A. Grasland, C. Collet, J. Coste, J. M. Esdaile, and P.
Vinceneux, “Reliability of tuberculin skin test measurement,” Annals
Mahabubnagar India 509001. He has 66 publicaions to his
of internal medicine, vol. 126, no. 3, pp. 210–214, 1997. credit in International Journals and Conferences. He is senior


member IEEE, life fellow IETE, life member IEI and life
member ISOI. His areas of interest are Speech Processing,
Signal Processing, Digital Image Processing and
Communications.
3. Dr. A. Venkataramana received his B.E, M.E and Ph.D.

degrees from Osmania University, Hyderabad. He has total
19 years of teaching experience. He worked as a Junior
Telecom Officer and ITS officer in the Department of
Telecom, G.O.I for a period of 4 years. He has published
various papers in IEEE, international, national conferences
and journals. He has received a number of prestigious
awards A.P. State Government Best Lecturer in Polytechnics
award, ISTE Best Polytechnic Teacher award, Bharath
Siksha Ratan and Rashtra Ratna awards. He is presently
working as faculty in the department of Electronics and
Communication Engineering at Q.Q Government
Polytechnic, Hyderabad. His areas of interests are Digital
Image Processing, Pattern Recognition, Neural Networks
and Applications of Moments for Image Processing
Problems

Diagnosis of Tuberculosis Using Combined Blur and Affine Moment Invariants

Enviado por

Dados do documento

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Diagnosis of Tuberculosis Using Combined Blur and Affine Moment Invariants

Enviado por

Direitos autorais:

Formatos disponíveis

ISSN 2394-3777 (Print)

ISSN 2394-3785 (Online)

International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)