You are on page 1of 5

6 (IJCNS) International Journal of Computer and Network Security,

Vol. 2, No. 1, January 2010

An Improved Implementation of Brain Tumor


Detection using Soft Computing
Marcus karnan1 and T.logeswari2
1
Tamilnadu College of Engineering
Coimbatore, India
drmkarnan@gmail.com
2
Research Scholar, Dept of Computer Science
Mother Theresa women’s University
Kodaikkanal, India
saralogu4uin@gmail.com

Abstract: Ant Colony Optimization (ACO) metaheuristic is a tools are expected to be helpful to the radiologist. Thangavel
recent population-based approach inspired by the observation of et al., [12,13] presented a study on methods of various
real ants colony and based upon their collective foraging stages on automatic detection of micro calcification in
behavior. In ACO, solutions of the problem are constructed digital mammograms. According to those studies it is noted
within a stochastic iterative process, by adding solution that the ACO has not been implemented in the field of
components to partial solutions. Each individual ant constructs mammogram analysis. In this paper, meta-heuristic
a part of the solution using an artificial pheromone, which
algorithms such as ACO are implemented to extract the
reflects its experience accumulated while solving the problem,
and heuristic information dependent on the problem. In this
suspicious region in the MRI Brain Image. The textural
paper, the proposed technique ACO hybrid with Fuzzy and features can be extracted from the suspicious region to
Hierarchical self organizing Map(HSOM) Hybrid with Fuzzy classify them into benign or malign. two different methods
describe segmentation consists of two steps. In the first step, the ACO hybrid with Fuzzy and HSOM Hybrid with Fuzzy is
MRI brain image is Segmented using HSOM Hybrid with Fuzzy used for segmentation
and the second step ACO Hybrid with Fuzzy method to extract
the suspicious region from Background tissue Both techniques 2. Segmentation
are compared and performance evaluation is evaluated
Segmentation is the initial step for any image analysis.
Keywords: MRI Brain Image analysis, ACO, HSOM, Fuzzy Segmentation is an important process to extract information
C-Means. from complex medical images. Segmentation has wide
application in medical field [1,4,10].The main objective of
1. Introduction the image segmentation is to partition an image into
mutually exclusive and exhausted regions such that each
Brain tumors are abnormal growths in the brain that can be region of interest is spatially contiguous and the pixels
either cancerous (malignant) or non cancerous (benign). within the region are homogeneous with respect to a
The effects on the brain of malignant and benign brain predefined criterion.[15] Widely used homogeneity criteria
tumors are very similar and can cause the same types of include values of intensity, texture, color, range, surface
problems depending upon the type of tumor and where it is normal and surface curvatures. During the past many
located in the brain. Each year more than 2Lakhs people in researchers in the field of medical imaging and soft
the United States are diagnosed with a primary or metastatic computing have made significant survey in the field of
brain tumor. Primary brain tumors comprise approximately image segmentation [5,9,10]
40,000 of these diagnoses. Brain tumors are the leading
cause of solid tumor cancer death in children under the age 2.1 Hierarchical Self Organizing Map (HSOM)
of 20, now surpassing Acute Lymphoblastic Leukemia Image segmentation techniques can be classified as based on
(ALL). They are the second leading cause of cancer death in edge detection, region or surface growing, threshold level,
male adults ages 20-29 and the fifth leading cause of cancer classifier such as HSOM, and feature vector clustering or
death in female adults ages 20-39. Metastatic brain tumors, vector quantization. Vector quantization has proved to be a
cancer that spreads from other parts of the body to the brain, very effective model for image egmentation process [2].
are the most common types of brain tumors. They occur in Siyal et al described a new m method on Fuzzy C-means for
10-15% of people with cancer. Primary brain tumors segmentation purpose Multiforme[10]. Murugavalli1 and
generally do not metastasize to other parts of the body. Rajamani , A high speed parallel fuzzy c-mean algorithm
Although computer-aided Magnetic Resonance for brain tumorsegmentation [11]. Murugavalli1, and
Imaging(MRI) brain image has been studied over the last Rajamanil An Improved Implementation of Brain Tumor
two decades, automated interpretation of segmentation still Detection Using Segmentation Based on Neuro Fuzzy
remains very difficult. Double readings, as carried out, for Technique Self Organizing Map (SOM) is a member of the
example, by two radiologists, usually improve the quality of Competition learning Neural network (CLNNs) and thiscan
diagnostic findings, thus, greatly reducing the probability of be the best choice when implementing vector quantization
misdiagnosis. On these grounds, adequate computational
(IJCNS) International Journal of Computer and Network Security, 7
Vol. 2, No. 1, January 2010

using neural network. The importance of SOM for vector of iterations (N), number of ants (K), initial pheromone
quantization is primarily due to the similarity between the value(T0).
competitive learning process employed in the SOM and the Pheromone Initialization: The initial pheromone value T
vector quantization procedure .The HSOM directly address 0 has been initialized for each ant and a random pixel is
the aforesaid shortcomings of the SOM. HSOM is the chosen from the image, which has not been selected
combination of self organization and graphic mapping previously. To find out the pixels is been selected or not, a
technique. HSOM combine the idea of regarding the image flag value is assigned for each pixel. Initially the flag value
segmentation process as one of data abstraction where the is assigned as 0, once the pixel is selected the flag is
segmented image is the final domain independent changed to 1. This procedure is followed for all the ants. For
abstraction of the input image.. The neuron with weight each ant a separate column for pheromone and flag values
vector most similar to the input is called the Best Matching are allocated in the solution matrix.
Unit (BMU). The weights of the BMU and neurons close to Local Pheromone Update: Update the pheromone values
it in the SOM lattice are adjusted towards the input vector. for all the randomly selected pixels using the following
The magnitude of the change decreases with time and with equation:
distance from the BMU. The update formula for a neuron T new= (1 – ρ) * Told + ρ * T 0(2) (6)
with weight vector Wv(i) is where Told and T new are the old and new pheromone values,
and ρ is rate of pheromone evaporation parameter in local
Wv(i+ 1) = Wi(i+1)=wi(i)+hci(I)*[(x(i)-w(i))] (1) pdate, ranges from [0,1] i.e., 0 < ρ < 1. Calculate the
Here hci is neighborhood function to calculate it posterior energy function value for all the selected pixels by
h(i)=h(rc-r1)*a(i)*alpha (2) the ants from the solution matrix.
Here rc-r1=current neuron-next current neuron Global Pheromone Update: Genetic algorithm [7,8,14] is
a(i)=sigma 0 * exp(-i/nsm) (3) used to compare the posterior energy function value for all
Initialize the variable the randomly selected pixels from each ant, to select the
Sigma=number of neighborhood pixels (8 or 24 or 48 or minimum value from the set, which is known as ‘Local
80 or 120) Minimum’ (Lmin) or ‘Iterations best’ solution. The
if sliding window size(3*3=8),(5*5=24),(7*7=48),) subsequent algorithm implements genetic operators to find
(9*9=80),(11*11=120) out the local minimum
Sigma N= Sigma 0 * exp(-i/taul) (4)
Taul= total number of pixels / log(neighborhood number 3. Implementation Of ACO And HSOM With
of pixel) Fuzzy
Similarly find the sigma value for each and every pixel This
process is repeated for each input vector for a (usually large) Fuzzy C-Means Based on the previous equation,
number of cycles λ. The network winds up the first step of the regularized FCM-HSOM algorithm is
associating output nodes with groups or patterns in the input the following:
data set. If these patterns can be named, the names can be Step 2: Calculate the cluster centers.
attached to the associated nodes in the trained net. C = (N/2)1/2 (7)
The winning neuron formula is Step 3: Compute the Euclidean distances
||x-wc||=maxi{[=x-wi]} (5) Dij = CCp – Cn (8)
X is a neuron ,wi is the weight vector Step 4: Update the partition matrix
1
2.2 Ant Colony Optimization (ACO) U ij = 2 / (m-1)
Ant Colony Optimization (ACO) is a population-based c
 d ij 
approach first designed by Marco Dorigo and coworkers, ∑  
k=1  d kj 
inspired by the foraging behavior of ant colonies [3]. (9)
Individuals ants are simple insects with limited memory and (Repeat the step 4)
capable of performing simple actions[6,13]. However, the
collective behavior of ants provides intelligent solutions to Until Max[ │Uij(k+1)-Uijk│] <€ is satisfied
problems such as finding the shortest paths from the nest to Step 5: Calculate the average clustering points.
a food source. Ants foraging for food lay down quantities of
a volatile chemical substance named pheromone, marking c c n
their path that it follows. Ants smell pheromone and decide Ci = ∑ J i = ∑∑ Uij n dij 2
to follow the path with a high probability and thereby i =1 i =1 j =1
(10)
reinforce it with a further quantity of pheromone. The
probability that an ant chooses a path increases with the Step 6: Compute the adaptive threshold Adaptive threshold
number of ants choosing the path at previous times and with =max (Adaptive threshold, ci ) i=1...n
the strength of the pheromone concentration laid[3,5,16]. In
this work, the labels created from the MRF method and the In the first step ,the algorithm select the initial cluster from
posterior energy function values for each pixel are stored in HSOM Clustering algorithm. Then, in later step after
a solution matrix. The goal of this method is to find out the several iteration of the algorithm, the final result converges
optimum label of the image that minimizes the posterior to actual cluster center achieved and it is very important for
energy function value. Initially assign the values of number an FCM algorithm.
8 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 1, January 2010

Table2 shows the result of image segmentation of ACO .In


4. Result And Analysis any computer aided analysis ,the execution time is one of
the important parameters of medical image segmentation .In
Table1 shows the result of image segmentation of HSOM .In these result, we have calculated the number of tumor cells of
any computer aided analysis ,the execution time is one of different neighborhood pixel of 3×3, 5×5, 7×7, 9×9, 11×11
the important parameters of medical image segmentation .In windows. In that 3×3 window is chosen based on the high
these result, we have calculated the number of tumor cells of contrast than 5×5, 7×7, 9×9, and 11×11.
different neighborhood pixel of 3×3, 5×5, 7×7, 9×9, 11×11
windows. In that 3×3 window Table 2: Adaptive threshold Number of segmented pixel,
is chosen based on the high contrast than 5×5, 7×7, 9×9, Execution time, weight

Value / Value / 3x3 5x5 7x7 9x9 11x11


Neighbor
Neighborhood
hood
pixels 3x3 5x5 7x7 9x9 11x11 pixels
Adaptive 188.338 164.752 161.138 150.749 139.38
Adaptive Threshold 6 4 5 1 62
threshold 184.07 152.09 149.70 141.15 133.22 Number of
Segmente
Number of
d 765 1894 2260 4050 8299
segmented Execution 19.292
Time 40.9446 27.4551 24.8271 24.163 5
pixels 883 3732 4319 7417 12025
Weight 8 40 50 87 122
Execution
time 31.83 33.14 29.50 27.02 28.364
Figure 2 shows the tested segmented image with various
Weight 28.28 51.22 55.05 74.56 93.91 neighborhood pixels .the original image of(256x256).
and 11×11.

Table 1: Adaptive threshold Number of segmented pixel,


Execution time, weight

Figure 1 shows the tested segmented image with various


neighborhood pixels .the original image of (256x256)

Figure 2. Input image (256x256), output 3x3, output 5x5,


output 7x7

Figure 1. input image (256x256), output 3x3, output 5x5,


output 7x7, output 9x9, output 11x11
5. Performance Analysis
(IJCNS) International Journal of Computer and Network Security, 9
Vol. 2, No. 1, January 2010

It is very difficult to measure the performance of


enhancement objectively. If the enhanced image can make
observer perceive the region of interest better, then we can
say that the original image has been improved. Here we are
giving input image in that neighborhood pixel of 3×3, 5×5,
7×7, 9×9, 11×11 windows are analyzed. In that 3×3 window
is chosen based on the high contrast than 5×5, 7×7, 9×9,
and 11×11

Figure3 shows the wining neuron for HSOM is 3x3 is 209,


5x5 is 201, 7x7 is 194, 9x9 is 186 and 11x11 is 177. Figure
4 shows the Execution time in HSOM of 3x3 is 13.76 , 5x5
is 14.96 , 7x7 is 15.20, 9x9 is 11.05 and 11x11 is 11.53.
Figure 5 shows the weight vector for HSOM is 3x3 is 14 ,
5x5 is 8 , 7x7 is 15, 9x9 is 23 and 11x11 is 32 .. Figure 6
shows the number of segmented pixel in HSOM of 3x3 is Figure 5. relationship between weight and neighborhood
795, 5x5 is 1073, 7x7 is 1285, 9x9 is 1594and 11x11 Pixels
is1881. Below the figure shows the performance analysis of
given input image.
Figure 3. relationship between Adaptive threshold and
neighborhood pixels

Figure 6. relationship between no of segmented pixel


and neighborhood pixels
In above 3×3, 5×5, 7×7, 9×9, 11×11 windows are analyzed
in that 3×3 window is chosen based on the high contrast
than 5×5, 7×7, 9×9, and 11×11.

6. Conclusion

In this work, a novel approach applied to MRI Brain Image


segmentation based on the combination of HSOM with
fuzzy and Ant Colony System with Fuzzy Algorithm is
Analysised. In that The self-organizing map describes a
mapping from a higher dimensional input space to a lower
dimensional map space. The procedure for placing a vector
from data space onto the map is to find the node with the
closest weight vector to the vector taken from data space and
to assign the map coordinates of this node to our vector. Ant
Colony Optimization (ACO) with Fuzzy Algorithm have
been used to find out the optimum label that minimizes the
Maximizing a Posterior estimate to segment the image. The
ACO search is inspired by the foraging behavior of real
ants. Each ant constructs a solution using the pheromone
information accumulated by the other ants. In each iteration,
Figure 4. relationship between execution time and local minimum value is selected from the ants’ solution and
neighborhood pixels the pheromones are updated locally. Fuzzy Algorithm is
used to find out the local minimum. If this value is less than
global minimum, the local minimum is assigned to global
minimum. The pheromone of the ant that generates the
global minimum is updated. At the final iteration global
10 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 1, January 2010

minimum returns the optimum label for image [12] S. Murugavalli1 , V. Rajamani,” An Improved
segmentation. In the above 3×3, 5×5, 7×7, 9×9, 11×11 Implementation of Brain Tumor Detection Using
windows are analyzed the ACO with Fuzzy of 3×3 window Segmentation Based on Neuro Fuzzy Technique”
is chosen based on the high contrast than 5×5, 7×7, 9×9, Journal of Computer Science 3 (11): 841-846, 2007.
and 11×11. [13] K. Thangavel, M. Karnan, R. Siva Kumar, and A. Kaja
Mohideen. Automatic Detection of Microcalcification in
ACKNOWLEDGEMENT Mammograms-A Review. International Journal on
Graphics Vision and Image Processing, 5(5):31-61, 2005
The author wishes to thank Doctor Pankaj Metha for his [14] K.Thangavel, M.Karnan, P.Jaganathan, Pethalakshmi,
helpful suggestion on tumor recognition with his knowledge R.Siva Kumar and Geetha Ramani, “Ant Colony
and experience in medicine area. The MRI Image data Algorithms In Diverse Combinational Optimization
obtained from KMCH Hospital.in Coimbatore Problems –A Survey,” International Journal on
INDIA Automatic Control System and Engineering, vol. 6, no.
1, pp: 7-26, 2005.
References [15]K.Thangavel, M.Karnan, R.Siva Kumar and
A.Kajamohideen, “ Segmentation and Classification of
[1] Aaron Lefohn, Joshua Cates, Ross Whitaker,”Interactive Microcalcification in Mammograms Using the Ant
GPU-Based level sets for 3D Brain Tumor Colony System,” International Journal on Artificial
Segmentation”, April 16, 2003. Intelligence and Machine Learning, vol. 5, no. 3, pp: 29-
[2] Bilbro, G., M. White and W. Snyder, 1987. “Image 40, 2005.
segmentation with neurocomputers”, In: R. Eckmiller [16] Y. Ge, Q. C. Meng, C. J. Yan and J. Xu. A Hybrid Ant
and C.van der Malsburg (eds.), NeuralComputers, Colony Algorithm for Global Optimization of
NATO AS1 Series, (Springer-Verlag, Berlin, Germany), Continuous Multi-Extreme Functions. Proceedings of the
41: 7 l-79. Third International Conference on Machine Learning
[3] M. Dorigo, V. Maniezzo and A. Colorni. The Ant and Cybernetics, Shanghai, 2427-2432, 2004.
System:
Optimization by a Colony of Cooperating Agents. IEEE Authors Profile
Transactions on Systems, Man and Cybernetics-Part B,
1(26):29-41, 1996 Marcus Karnan received the BE
[4] M. Dorigo and T. Stuztle, Ant Colony Optimization, Degree in Electrical and Electronics
PHI ed, 2005. Engineering from Government
[5] J. A. Gamez and J. M. Puerta. Searching for the best College of Technology,Bharathiar
elimination sequence in Bayesian networks by using ant University, India. Received the ME
colony optimization. Pattern Recognition Letters, Degree in Computer Science and
Engineering from Government
23:261- 277, 2002.
College of Engineering,Manonmaniam
[6] Haralick, R.M. and L.G. Shapiro, 1985. “Survey ,image Sundaranar University in 2000.
segmentation techniques”, Computer Vision, Graphics Received the PhD degree in CSE from Gandhigram Rural
Image Process, 29: 100-132[9] University, India in 2007, Currently he is working as Professor,
[7] M.Karnan and K.Thangavel, Automatic Detection of the Department of Computer Science & Engineering Department,
Breast Border and Nipple Position onDigital Tamilnadu College of Engineering, India. He has been in teaching
Mammograms Using Genetic Algorithm, International since 1998 and has more than eleven years in industrial and
Journal on Computer Methodsand Programs in research experience. His area of interests includes medical image
Biomedicine (Elsvier). VOL 87, pp.12-20 2007 processing, artificial intelligence, neural network, genetic
algorithm, pattern recognition and fuzzy logic
[8] M.Karnan, R. Sivakumar, M. Almelumangai, K.
Selvanayagi and T. Logeswari,” Hybrid ParticleSwarm
T.Logeswari received MCA degree
Optimization for Automatically Detect the Breast Border
in the year 2006 and Mphil degree
and Nipple position toIdentify the Suspicious Regions on
in the year 2007 from Alagappa
Digital Mammograms Based on
university , Karakudi, India. She
ymmetries”,International Journal of Soft Computing 3
has registered Phd In Mother
(3): 220-223, 2008
Terasa Women’s University,
[9] Kohonen, T., “Self-Organization and Associative
Kodaikanal ,India in the year 2007.
Memory”, 2nd Edition (Springer-Verlag, Berlin,
She has 3 year of teaching Experience. She has presented 5
Germany, 1988.
paper in International Conferences and published 2 paper in
[10]sai.C,Manjunath .S, Jagadeesan.R, ”Automated
International Journals.
Segmentation of brain MR Images”,Pergamon, Pattern
Recognition, Vol 28, No 12, March 1995.
[11] S. Murugavalli1 , V. Rajamani,” A high speed parallel
fuzzy c-mean algorithm for brain tumor
segmentation” ,BIME Journal, Volume (06), Issue (1),
Dec., 2006.

You might also like