You are on page 1of 4

Asian Journal Of Computer Science And Information Technology 3 : 4 (2013) 56 - 59.

Contents lists available at www.innovativejournal.in

Asian Journal of Computer Science And Information Technology

Journal Homepage: http://www.innovativejournal.in/index.php/ajcsit

FIREFLY ALGORITHM BASED MAMMOGRAPHIC IMAGE ANALYSIS


Goutam Das, Md. Iqbal Quraishi, Manisha Barman

Dept. of Information Technology Kalyani Govt. Engineering College, Nadia, India

ARTICLE INFO ABSTRACT


Corresponding Author: Mammography images are used to detect tumor formation within breast
Goutam das tissues. It works as a monitoring stage where a medical practitioner judges
Dept. of Information Technology, that the probability of cancer. Several image processing techniques in
Kalyani Govt Engineering conjunction with various algorithms shows ways to enhance and detect large
College, Kalyani, Nadia, India tissue lumps. But the domain of new nature inspired metaheuristic algorithms
remains untouched. Generally these algorithm are meant to solve optimization
Keywords: - Mammographic problems but this study attempts to involve one of the algorithms which is
Images, Firefly Algorithm, Image known as Firefly algorithm, to enhance the quality of mammographic image by
Enhancement, K-Means, enhancing the edge details and contrast. As this shown, next the enhanced
Segmentation, Self Organizing result is segmented using self organizing map (SOM) to properly visualize the
Map, Clustering, Peak Signal To large tissue lump formation within breast, which is then clustered using K-
Noise Ratio, Histogram, means algorithm. The results of enhancement also shown via pick signal to
Computer Vision noise ratio and histogram difference. Whereas segmentation results are
presented for visual analysis only, because of the reason that mammographic
images needs to be analyzed by medical practitioners as appears on screen.

2013, AJCSIT, All Right Reserved.


INTRODUCTION
The challenge of computer vision remains within decomposition and B-spline had been introduced [3], with
the fact that visually clear representation of image after encouraging results. Histogram equalization based
processing. Researchers already been developed a variety approach in mammographic images in comparison with
of useful tools to achieve this objective. But still the local histogram enhancement method shows that root
challenge remains open because of enormous development mean squared error (RMSE) and pick signal to noise ratio
in the field of machine intelligence in combination with (PSNR) remains within the range of 25 and 12 respectively
computer vision. Denoising and enhancement remains a as compared with 12 and 58 for the other [4]. A mixture of
core task to be performed before to meet an objective. histogram processing and fuzzy to reduce noise and
Mammography images suffer from the same problem too. enhance the quality of mammographic image was proposed
Digitization of mammogram is not new, but analysis of and tested across 50 test images [5] proved to be an
which make the task critical. In general mammograms are effective solution in terms of segmentation accuracy. In the
used to identify cancer probability, in the context of lump paper [6] a modified mathematical morphology and Bi-
or a mass called tumor, which is known as malignant orthogonal wavelet transform had been used. The results
tumor. The severity of the problem can be identified from shows, contrast Improvement Index (CII) and Edge
the fact that more than 400,000 women die every year Preservation Index (EPI) shows that the proposed method
from this health problem [1]. In most cases of cancer, if yields significantly better image quality when compared
detected early, there remains a high probability to be with other contemporary methods. Testing the effects of
cured. Early detection of malignant glandular tissues is a several preprocessing segmentation methods [7], it was
critical task. In computer vision the objective should be to suggested that segmentation can be applied on
build a system that can highlight the regions of highly mammographic images before further processing or as a
dense tumor like structure within breast. The final decision preprocessing step. Combining Gabor algorithm with fast
should depend upon medical practitioners, rather than Fourier transform and overlay mask segmentation [8],
machine intelligence and the system should work like a shows enhancement of edges which increase visibility of
decision support system to the professionals of the job. To thin tissues far better. For mass screening of
achieve this objective enhancement of the mammographic mammographic images a wavelet-based multiresolution
image is trivial, after which further processing can be analysis [9] was performed to show an efficient adaptively
implied. [2] Based on morphological operation such as top- enhance local edge features, suppress noise, and improve
hat transform and bottom-hat transform had been used to global visualization. In an early paper on mammography
increase the contrast of mammographic image, to help in [10], region based contrast enhancement been performed
detecting cancer probable regions. Micro-calcification to show visibility of microcalcification clusters. An edge-
clusters enhancement and identification based on wavelet based segmentation method based on region of interest

56
Das et.al/Firefly Algorithm based Mammographic Image Analysis

was proposed using contrast stretching and filtering Define the objective function of f(x), where x=(x1,........,xd)T
respectively [11] to detect contours. However the Generate the initial population of fireflies or xi (i=1, 2,.., n)
performance regarding various breast density levels found Determine the light intensity of Ii at xi via f(xi)
to be not so encouraging. A review of existing processing While (t<MaxGen)
techniques on mammographic images shows [12] how For i = 1 to n (all n fireflies);
decomposition methods such as Quad tree Decomposition, For j=1 to n (n fireflies)
Discrete Wavelet Transform, Dual Tree Complex Wavelet if (Ij > Ii), move firefly i
Transform, Pyramid-Structured Wavelet Transform, Gabor towards j;
Wavelets been used till now, though it does not prescribe end if
any general trend in this area. The implementation of Attractiveness varies with distance r via Exp
adaptive contrast enhancement using multiscale [-r2];
representation [13] proves to show wavelet based Evaluate new solutions and update light
algorithm can produce better results. Wavelet based local intensity;
feature enhancement of mammographic images on End for j;
multiscale basis could result in desired outcome but End for i;
clinically cannot be extrapolated directly [14] [15]. The Rank the fireflies and find the current best;
application of metaheuristics in either small or large scale End while;
has not been found. As this is said, nature inspired Postprocess results and visualisation;
metaheuristics not been considered well enough, as these End procedure;
algorithms are opening doors for a complete new chapter Alpha determines random percentage in firefly moving. It
of computer vision and machine intelligence. Algorithms includes value between zeros to one. Absorption coefficient
such as Harmony search, Firefly algorithm, Artificial Bee is named gamma. The constraint varies between zeros to
Colony, Cuckoo Search are yet to be tested. In this study an extreme. If Coefficient is close to zero, then =0 and this
attempt has been made to show how these nature inspired corresponds to a special case of particle swarm
algorithms can provide effective result in mammographic optimization. Besides, if absorption coefficient is close
image with some existing techniques. Firefly algorithm is particularly, this is the case where the fireflies fly in a very
implemented to enhance the mammographic image for foggy region randomly. Finally 0 is maximum
edge information enhancement and overall contrast Attractiveness value [20].
enhancement, as a preprocessing stage and this enhanced Each firefly i can move toward another more attractive
image is then feed into self organizing map (SOM) and (brighter) firefly j by equation [21].
further an adaptive K-means algorithm is implemented for xi = xi + 0er2 ij (Xj - Xi) + (rand -1/2) (1)
clustering the segmented region to show the final outcome r is the distance
of the proposed approach. 0 is attractiveness
METHODOLOGY is light absorption coefficient
A. Firefly Algorithm:
The Firefly algorithm was developed by Xin-She Yang [16]
[17], on the basis of flashing light behavior of fireflies in
nature. The fundamental function of the glowing light is to
attract mate during mating session, where male firefly uses
a brief signal pattern and female firefly respond in certain
time interval for the same species. The pattern of flashes is
often unique for a particular species of fireflies. A three rule
base of firefly was established,
1) All fireflies are unisex, so that one firefly is
attracted to other fireflies regardless of their sex. Figure 1: the original image
2) The less bright one will move towards the brighter B. Firefly Algorithm for image enhancement:
one. The value of brightness is proportional to In conjunction with the above algorithm and derivations,
brightness, which is reverse proportional to their below the strategy is given to meet the problem objective
distances. 1) Initial population: Total number of image pixels.
3) Brightness of each firefly is determined by 2) MaxGen: intensity variation through iteration.
evaluating objective function. If there is no 3) If the previous pixel value is greater than of
brighter one than a particular firefly, it will move current pixel value after considering the fitness
randomly. evaluation, which depends upon the global
The brightness of a firefly is affected or determined by the intensity values of the image, it should be replaced.
landscape of the objective function to be optimized [18] 4) As attractiveness varies with distance, the
[19]. For a maximization problem, the brightness is simply boundary value of the window size, that is 3x3 in
proportional to the value of the objective function. this case, is considered. Any value that crosses the
The algorithm itself proposes some initializations that boundary, it was ignored.
include: 5) After each iteration, the global best in
1) : the light absorption coefficient consideration in accordance to the window size
2) r: the particular distance from the light source was updated and the highest intensity value of that
3) d: the domain space. iteration was considered for the previous update.
The rank of the firefly was updated.
The general algorithm of firefly: 6) The value of gamma which is the absorption rate
was considered to give a smoothing effect for the

57
Das et.al/Firefly Algorithm based Mammographic Image Analysis

image. The value of which will be image specific. which through a iteration was increased in the consequent
0, contains the attractiveness which is then stages, that effectively draws clustering of the data
updated according to the rank matrix. patterns.
Firefly algorithm does not memorize or remember any
history of better situation for each firefly and this causes
them to move regardless of its previous better situation,
and they may end up missing their situations[ijisa].

Figure 4: After clustering


RESULTS
The results of this methodology found to very encouraging.
Quality of the enhanced image (Figure2) as compared with
Figure 2: after enhancement original image (Figure1) shows the visual improvements.
C. SOM based segmentation: PSNR of the original image was 28.996, which was
The main characteristic of SOM is that it helps to produce increased after the firefly based enhancement to 41.379.
segmentation easily the regions of the image that has The SOM segmentation proves to be very effective in terms
similar features [23]. This is the most well-known identifying the cluster centers and weight centers for
unsupervised approach using ANN, the self-organizing further processing. The probable tumor region was
feature maps (SOFM), developed by Kohonen [24]. identified in SOM. Applying K-means clustering next shows
Basically it is two layers model, which consists, the first the exact tumor region with an increased visibility.
layer as the input nodes and the second as the output nodes CONCLUSION
[25]. Based on the weight center of the input matrix, the The study attempts to provide a nature inspired
neighborhood values are determined which acts as a metaheuristic approach where firefly algorithm was
feature matrix for the input image. the Markov Random implemented with unsupervised learning that can
Field model was used to reduce the effect of noise upto a effectively identify the problem region in mammographic
limit of =0.0195. However entropy values are also image. The system discussed here act as a support
considered where no satisfactory result was found. The mechanism for decision making, where the end decision
similarity is defined by the Euclidian distance of two depends upon the medical practitioner. The overall
segmented parts. This is, performance of the system is also computationally less
SNN = [su,v] (2) expensive. However SOM is an unsupervised learning
methodology which involves increase in computational
time. This is minute problem, as it can be overcome as the
system will be independent of nature in terms of
implementation.
REFERENCES
[1] N.R. Pal, B. Bhowmick, S.K. Patel, S. Pal, J. Das,
2008. A multistage neural network aided system for
detection of microcalcifications in digitized mammograms,
Neuro Computing, pg.26252634.
[2] D. Narain Ponraj M. Evangelin Jenifer P. Poongodi J.
Samuel Manoharan, Morphological Operations for the
Mammogram Image to Increase the Contrast for the
Figure 3: After segmentation Efficient Detection of Breast Cancer, European Journal of
D. K-means clustering: Scientific Research ISSN 1450-216X Vol.68 No.4 (2012), pp.
Clustering, that is grouping of similar data point into one, 494-505.
which behaves same under some measure. In kmeans, it [3] Damir Seri and Sven Lonari, Enhancement of
finds local optima by keeping track of centroids [26]. Mamographic images for detection of Microcalcifications.
1) Input samples [4] Govind N Sarage, 2Dr. Sagar Jambhorkar, Enhancement
2) No. of clusters of Mammography Images for Breast Cancer Detection using
3) Initial cluster centre Histogram Processing Techniques, IJCST Vol. 2, Iss ue 4,
4) For every point of x there remains Oct. - Dec. 2011.
a center Ci, which is closed to x. [5] Peyman Rahmati, Ghassan Hamarneh, Doron Nussbaum
5) Calculating by every iteration and Andy Adler, A New Preprocessing Filter for Digital
and increasing the step size of Mammograms, icisp, 2010.
Ci+1 . [6] Harish Kumar.N Amutha. S, Dr. Ramesh Babu .D. R
Depending upon the above three parameters, the Enhancement of Mammographic Images using
segmented image has been taken with the window size of Morphology and Wavelet Transform, Harish Kumar N et
3x3, based upon weight center. After which based upon the al,Int.J.Computer Techology & Applications,Vol 3 (1),192-
weight center the initial cluster centers were identified, 198.

58
Das et.al/Firefly Algorithm based Mammographic Image Analysis

[7] Prof. Samir Kumar Bandyopadhyay,Pre-processing of Applications (SAGA 09), Vol. 5792 of Lecture Notes in
Mammogram Images, International Journal of Engineering Computing Sciences, pp. 169-178, Spriger, October 2009.
Science and Technology Vol. 2(11), 2010, 6753-6758. [17] X. S. Yang, (2010). Firefly Algorithm Stochastic Test
[8] Vishnukumar K. Patel, Prof. Syed Uvaid, Prof. A. C. Functions and Design Optimization. Int. J. Bio-Inspired
Suthar Mammogram of Breast Cancer detectionBased using Computation, vol.2, No. 2, pp.78-84, 2010.
Image Enhancement Algorithm, International Journal of [18] X.-S. Yang, Firefly Algorithm, Lvy Flights and Global
Emerging Technology and Advanced Engineering, Volume Optimization, Research and Development in Intelligent
2, Issue 8, August 2012 Systems XXVI (Eds M. Bramer, R. Ellis, M. Petridis),
[9] Lulin Chen, Chang W. Chen, Kevin J. Parker,Adaptive Springer, pp. 209-218, 2010.
feature enhancement for Mammographic images with [19] X. S. Yang, Engineering Optimization: An Introduction
wavelet multiresolution analysis, Journal of Electronic with Metaheuristic Applications. Wiley & Sons, New
Imaging 6(4), 467478 (October 1997). Jersey, 2010.
[10] William Mark Morrow, Raman Bhalachandra [20] Azam Amin Abshouri, Mohammad Reza Meybodi,
Paranjape,Rangaraj M. Rangayyan, Joseph Edward Leo Alireza Bakhtiary, New Firefly Algorithm based On Multi
Desautels,ReEion-Based Contrast Enhancement of swarm & Learning Automata in Dynamic
Mammograms, IEEE TRANSACTIONS ON MEDICAL Environments,978-1-4577-0174-0/11/$26.002011
IMAGING, VOL. 11, NO. 3, SEPTEMBER 1992. IEEE
[11] Yu Zhang, Noriko Tomuro, Jacob Furst, Daniela Stan [21] Babak Nasiri, M.R. Meybodi,Speciation based firefly
Raicu,Image Enhancement and Edge-based Mass algorithm for optimization in dynamic environments.
Segmentation in Mammogram, SPIE,2010. [22] Saibal K. Pal, C.S Rai, Amrit Pal Singh, Comparative
[12] D.Narain Ponraj, M.Evangelin Jenifer, P. Poongodi, Study of Firefly Algorithm and Particle Swarm Optimization
J.Samuel Manoharan,A Survey on the Preprocessing for Noisy Non-Linear Optimization Problems, I.J.
Techniques of Mammogram for the Detection of Breast Intelligent Systems and Applications, 2012, 10, 50-57
Cancer, Journal of Emerging Trends in Computing and Published Online September 2012 in MECS
Information Sciences, VOL. 2, NO. 12, December 2011 (http://www.mecs-press.org/) DOI:
[13] Andrew F. Laine, Sergio Schuler, Jian Fan, Walter 10.5815/ijisa.2012.10.06.
Huda,Mammographic Feature Enhancement by Multiscale [23] Wavelet-Based Color Image Segmentation using Self-
Analysis, IEEE TRANSACTIONS ON MEDICAL IMAGING, Organizing Map Neural Network
VOL 13, NO. 4, DECEMBER 1994 M. Arfan Jaffar, Muhammad Ishtiaq, Ayyaz Hussain and
[14] Duan Zhu Tian Hong Sun Lei,Research on Anwar M. Mirza, 2009 International Conference on
Mammogram Enhancement. Computer Engineering and Applications IPCSIT vol.2
[15] Andrew Laine, Jian Fan and Wuhai Yang,Wavelets for (2011) (2011) IACSIT Press, Singapore.
Contrast Enhancement of Digital Mammography, Specical [24] T. Kohonen, Self-organization and Associative
Issue of the EMBS Magazine entitled "Wavelets in Memory. Springer-Verlarg, 1989.
Medicine". [25] Dan Pelleg, Andrew Moore, Accelerating Exact K-
[16] X. S. Yang, Firefly algorithms for multimodal means Algorithms with Geometric Reasoning.
optimization, in Stochastic Algorithms Foundations and [26] A.Gersho, R.Grey,vector quantization and signal
Applications, Stochastic Algorithms: Foundations and compression, Kluwar Academic Pblishers: Dordrecht,
Netherlands, 1992.

59

You might also like