International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org
Volume 3, Issue 4, April 2014
ISSN 2319 - 4847
Brain Tumor Detection Using Multiple Kernel
Fuzzy C-Means on Level set Method
Mrs.D.Manju1 and K.Pavani2
1
Assistant Professor, Department of CSE, G. Narayanamma Institute of Technology and Science, Hyderabad
2
P.G Scholar, Department of CSE, G. Narayanamma Institute of Technology and Science, Hyderabad
Abstract
Image segmentation is the process of dividing an image into its constituent parts. Segmentation in medical image community is
still a challenging and difficult task because of the variety of possible shapes, locations and image intensities. Brain tumors are
one of the most common brain diseases, so detection and segmentation of brain tumors in MRI are important in medical to
diagnosis and treatment of the disease. The proposed multiple kernel fuzzy c-means (MKFCM) algorithm and adaptive level set
method for image segmentation can be used to detect brain tumor. MKFCM combines different information of image pixels in
the kernel space by using different kernel functions. Edge indicator function was redefined in MKFCM, which is used for
segmentation. Level set methods are used for contour evaluation and shape recovery. The result of MKFCM segmentation was
used to obtain the initial contour of level set method. This proposed algorithm can be implemented on the brain MRI images to
detect tumors.
Keywords: MRI, Brain tumor, Image segmentation, KFCM, MKFCM, Level set method
1. INTRODUCTION
1.1 Brain Tumor
Tumor or brain tumor is an intracranial solid neoplasm. Tumors are created by an unusual and uncontrolled cell division,
usually in the brain itself, also in blood vessels, lymphatic tissue, brain envelopes and cranial nerves. Tumors can also
spread from cancers initially situated in other organs. A brain tumor can cause damage in the brain by increasing
pressure, by pushing or by shifting the brain against the skull and by occupying and damaging nerves and healthy brain
tissue. The location of a brain tumor effects the type metastasize (spread) to other parts of the body outside of the central
nervous system (CNS). The CNS includes the brain and spinal cord.
1.2 Diagnose:
Brain tumor identification usually involves a brain scans, neurological examination, and/or an analysis of the brain tissue.
A brain scan is a picture that shows the internal structures of the brain. A specialized machine takes a scan in much the
same way a digital camera takes a photograph. The most common scans used for diagnosis are as follows:
1.2.1 MRI
Magnetic Resonance Imaging is a scanning device that captures images of the brain on film using magnetic fields and
computers. It does not use x-rays. It produces pictures from various planes, which permit doctors in creating a threedimensional image of the tumor. The MRI detects signals emitted from normal and unusual tissue, providing clear
images of most tumors.
1.2.2. CT or CAT Scan
Computed Tomography (CT) combines enlightened computer technology and x-ray. CT scan shows a combination of soft
tissue, blood vessels and bone. CT images can determine some types of tumors, as well as help detect swelling, bleeding,
bone and tissue clot.A biopsy is a surgical procedure in which a sample of tissue is taken from the tumor site and
examined under a microscope. The biopsy will provide information on types of unusual cells present in the tumor. The
purpose of a biopsy is to discover the type and grade of a tumor. A biopsy is the most accurate method of obtaining a
diagnosis.
1.2.3. PET images
A brain positron emission tomography (PET) scan is an imaging test of the brain. It uses a radioactive substance called a
tracer to look for disease or injury in the brain. A PET scan shows how the brain and its tissues are working. Other
imaging tests, such as magnetic resonance imaging (MRI ) and computed tomography (CT ) scans only reveal the
structure of the brain.
1.3 Image Segmentation
Image segmentation plays very important role in medical image analysis tasks. Image segmentation denotes a process by
which an image is partitioned into number of homogeneous segments, such that the union of any two neighborhood
segmentation yields different segments. The most important goal of segmentation is to simplify and convert the
representation of an image into new form that is more meaningful and easier to analyze [4]. There are many research
have done on this topic and number of methods has been proposed [5]-[1]. Image segmentation plays an important role in
image processing and it is necessary task for image analysis, understanding and pattern recognition.
Volume 3, Issue 4, April 2014
Page 197
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org
Volume 3, Issue 4, April 2014
ISSN 2319 - 4847
Different segmentation algorithms have been proposed those are edge based, histogram based and hybrid methods.
Among them fuzzy segmentation methods have more benefits, because they have much more information from the
original image than hard segmentation methods [3]. The fuzzy C-means (FCM) algorithm [10], assign pixels to fuzzy
clusters without labels. Unlike the hard clustering methods which force pixels to belong exclusively to one class, FCM
allows pixels to belong to multiple clusters with varying degrees of membership. Because of the additional flexibility,
FCM has been widely used in MR image segmentation applications recently. However, because of the spatial intensity in
homogeneity induced by the radio-frequency coil in MR image, conventional intensity-based FCM algorithm has proven
to be problematic, even when advanced techniques such as non-parametric, multi-channel methods are used [9]. To deal
with the in homogeneity problem, many algorithms have been proposed by adding correction steps before segmenting the
image [4,5] or by modeling the image as the product of the original image and a smooth varying multiplier field [16,6].
Recently, many researchers have incorporated spatial information into the original FCM algorithm to better segment the
images [16]. The kernel FCM (KFCM) algorithm is an extension of FCM, which maps the original inputs into a much
higher dimensional Hilbert space by some transform function. After this reproduction in the kernel Hilbert space, the data
are more easily to be separated or clustered. Previous research on transformation to the kernel space has already been
studied. Liao et al. [8] have directly applied the KFCM in the image-segmentation problems, where the input data
selected for clustering is the combination of the pixel intensity and the local spatial information of a pixel represented by
the mean or the median of neighboring pixels. Chen and Zhang [11] applied the idea of kernel methods in the calculation
of the distances between the examples and the cluster centers. They compute these distances in the extended Hilbert
space, and they have demonstrated that such distances are more robust to noises. To keep the merit of applying local
spatial information, an additional term about the difference between the local spatial information and the cluster centers
(also computed in the extended Hilbert space) is appended to the objective function. More kernel methods, the
kernelization of clustering algorithms besides FCM and their applications in the problems of image segmentation and
classification can be found in [12]. Recently, developments on kernel methods and their applications have emphasized the
need to consider multiple kernels or composite kernels instead of a single fixed kernel [12].With multiple kernels, the
kernel methods gain more flexibility on kernel selections and also reflect the fact that practical learning problems often
involve data from multiple heterogeneous or homogeneous sources [12],[13].Specifically, in image-segmentation
problems, the inputs are the properties of image pixels, and they could be derived from different sources. For example, the
intensity of a pixel is directly obtained from the image itself, but some complicated texture information is perhaps gained
from some wavelet filtering of the image [14]. Multiple-kernel methods provide us a great tool to fuse information from
different sources [15]. It is necessary to clarify that, in this paper, we use the term “multiple kernel” in a wider sense than
the one used in machine learning community. In the machine learning community, “multiple-kernel learning” refers to
the learning using an ensemble of basis kernels (usually a linear ensemble), whose combination is optimized in the
learning process. In this paper, we focus more on the flexible information fusion by applications of composite kernels
constructed by multiple kernels defined in different information channels. The combination of the ensemble kernel can be
automatically adjusted in the learning of multiple-kernel FCM (MKFCM) or it can be settled by trial and error or crossvalidation.
2. PROPOSED M ETHOD (MKFCM)
Multiple kernel fuzzy c-means provides us to combine different information of image pixels in the kernel space by
combining different kernel functions defined on specific information domains.
Fuzzy c-means clustering method will be largely limited to spherical clusters. kernel fuzzy c-means algorithm maps data
with nonlinear relationships to accurate feature spaces. Kernel combination, or selection, is effective for kernel
clustering. In this paper a multiple kernel fuzzy c-means (MKFC) algorithm which extends the Kernel fuzzy c-means
algorithm with a multiple kernel learning setting. Multiple kernels automatically adjusts the kernel weights, MKFCM is
more important to ineffective kernels and irrelevant features. It makes the choice of kernels less crucial. An experiment
on the medical images shows the effectiveness of the proposed MKFCM algorithm [2]. MKFCM minimizes the objective
function as.
c n m
2
J (U , V )     u ij  (xi )   (vi )
m
i 1 j 1
Where xi = [xi,vi] €R2 i=(1,2,3,4….n), Uij membership matrix,
Φ(xi) is the transformation defined by the combined kernel.
k
com
( x , y )  
com
( x ), 
com
(2)
( y) 
Where k=k1+αk2 .k1= Gausian kernel for pixel intensity.K2= Gausian kernel for spacial information.K1 ,k2 defined as
2
k ( x , x )  exp(  r x  x
i
i
1 i j
Volume 3, Issue 4, April 2014
)
Page 198
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org
Volume 3, Issue 4, April 2014
ISSN 2319 - 4847
2
k ( x , x )  exp(  r x  x
i
j
2 i j
)
Proposed MKFCM algorithm replaces the kernel function as
K
com
 K  K
1
2
(3)
Kernel function selection can be increased by a linearly combined kernel function is proposed here
k
(4)
b
b
b
 w *k w
* k  ............w * k
L
1
1
2
2
l
l
where
KL= Combined Kernel
If b>1 is coefficient similar to the fuzzy coefficient m,Here b value is chosen as 10.
The learning rule for membership values is
u
c
2
2 1
 (  ( d
/ d
))
ij
ij
kj
k 1
(5)
Where
n
d
2
ij
  ( x )   (v )
L j
L j
n
2  (u lk ) m k L ( x h , x j )
2

kL (x j , x j ) 
h 1
n
 (u ik ) m
h 1
n
6)
m
m
  (u lh ) (u il ) k L ( x h , x l )

h 1 l 1
n
(  (u ik ) m ) 2
h 1
Start from the objective function involving cluster in the membership matrix falls below a given threshold. The
computational complexity of MKFCM is O (N2). MKFCM algorithm, starts by initializing a random membership matrix
satisfying non-negative and unity constraints. Optimal weights are calculated by fixing the memberships, and optimal
memberships are updated assuming fixed weights. The process is repeated until the amount of change per iteration
excluding construction of the kernel metrics
3. ADAPTIVE LEVEL SET METHOD
In image segmentation, Level set methods are numerical techniques designed to track the interfaces between two different
regions. Active contours are dynamic curves that move toward the object boundaries. Adaptive Level set method is able to
deal with non sharp segment boundaries and defines an edge boundary and segmentation of the image. Level set method
handles topological changes in the edge contour that would be difficult to handle with a model that directly evolve the
contour. Level set provides a straight way to estimate the geometric properties of the evolving structure [18] and addresses
the problem of curve and surface [17].Level set method uses a signal function, where its zero level corresponds to the
actual contour. The zero level set function at the front as follows:
φ (x, y, t)= 0 ,if (x, y)€1
(7)
Figure 1 Zero level set of the surface is a square
The curve inside and outside regions defined as
φ (x, y, t)> 0 inside the boundary line
φ (x, y, t)= 0 on the boundary
φ (x, y, t)< 0 outside the boundary
An energy function defined as
E (φ) = µ Eint (φ) + Eout (φ)
An Edge Indicator function can be defined as fallows
g  1 / 1  G * I
(8)
Where I is the image, g the edge indicator function [17].
Volume 3, Issue 4, April 2014
Page 199
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org
Volume 3, Issue 4, April 2014
ISSN 2319 - 4847
4. DISCUSSIONS AND CONCLUSION
The original image is partitioned with MKFCM, and then the Adaptive Level set method is used to obtain the boundary
of the image. The proposed algorithm extracts the corresponding boundaries and improves the accuracy of the
segmentations. MFKCM can be applied to a variety of images, including dataset images, face clustering and satellite
images to compare the performance of the proposed algorithm.
References
[1] T.Saikumar and P.Yugadner, " An adaptive threshold algorithm for MRJ Brain image segmentation on level set
method," International Journal of Advances in Computer Networks and its Security,Vol. 1, pp 367-370, 2011.
[2] Hsin-Chien Huang, Yung-Yu Chuang and Chu-Song Chen,” Multiple Kernel Fuzzy Clustering”, IEEE Transactions
on fuzzy systems ,2011.
[3] Bezdek JC, Hall LO, Clarke LP. Review of MR image segmentation techniques using pattern recognition. Med Phys
1993;20:1033—48.
[4] S.K Pal and J.e. Bezdek, Fuzzzy Models for Pattern Recognition. New York: I EEE Press, 1992.
[5] J.e. Bezdek, J. Kellet, R.Krishnapuram and N.R Pal," Fuzzy model and algorithms for Pattern Recognition and
Image Processing, Kluwar, Boston 1999.
[6] K.Lung," A Cluster validity index for fuzzy clustering," Pattern Recognition Letters, Vol. 25, 1275-1291,2005.
[7] K.L.Wu, M.S.Yang, " Alternative c-means clustering algorithms, " Pattern Recognition Vo1.35, pp.2267-2278,
2002.
[8] L. Liao, T. S. Lin, and B. Li, “MRI brain image segmentation and bias field correction based on fast spatially
constrained kernel clustering approach,” Pattern Recognit. Lett., vol. 29, no. 10, pp. 1580–1588, Jul. 15, 2008.
[9] Wells WM, Grimson WEL, Kikinis R, Arrdrige SR. Adaptive segmentation of MRI data. IEEE Trans Med Imaging
1996;15: 429—42.
[10] Bezdek JC. Pattern recognition with fuzzy objective function algorithms. New York: Plenum Press; 1981.
[11] S. C. Chen and D. Q. Zhang, “Robust image segmentation using FCM with spatial constraints based on new kernelinduced distance measure,” IEEE Trans. Syst., Man, Cybern. B, Cybern, vol. 34, no. 4, pp. 1907– 1916, Aug. 2004.
[12] G. Camps-Valls, L. Gomez-Chova, J. Munoz-Mari, J. Vila-Frances, and J. Calpe-Maravilla, “Composite kernels for
hyperspectral image”.
[13] S. A. Rojas and D. Fernandez-Reyes, “Adapting multiple kernel parameters for support vector machines using
genetic algorithms,” in Proc. IEEE Congr. Evol. Comput. Edinburgh, U.K., 2005, vol. 1–3, pp. 626–631.
[14] K. Muneeswaran, L. Ganesan, S. Arumugam, and K. R. Soundar, “Texture image segmentation using combined
features from spatial and spectral distribution,” Pattern Recognit. Lett., vol. 27, no. 7, pp. 755–764, May 2006.
[15] G. Camps-Valls, L. Gomez-Chova, J. Munoz-Mari, J. L. Rojo-Alvarez, and M. Martinez-Ramon, “Kernel-based
framework for multitemporal and multisource remote sensing data classification and change detection,” IEEE Trans.
Geosci. Remote Sens., vol. 46, no. 6, pp. 1822–1835, Jun. 2008.
[16] Tolias YA, Panas SM. On applying spatial constraints in fuzzy image clustering using a fuzzy rule-based system.
IEEE Signal Proc Lett 1998;5(10):245—7.
[17] S.Osher and R. Fedkiw," Level set methods and Dyna,ic implicit surfaces, Springer, 2002, pp.112-113
[18] J.Gomes and O. Faugeras," Reconciling distance functions and level sets J. Visual communic. And Img.
Representation, 20000, pp.209-223.
AUTHOR
D.Manju currently working as Assistant Professor, at Department of CSE, G. Narayanamma Institute of
Technology and Science, Hyderabad. Obtained her Master’s degree in Artificial Intelligence from University
of Hyderabad in 2004. Areas of interest are Image processing, Distributed systems & Artificial Intelligence.
K.Pavani P.G Scholar, Department of CSE, G. Narayanamma Institute of Technology and Science,
Hyderabad. Obtained her B.Tech degree from TKR College Of Engineering and Technology
Medbowli,Meerpet , CSE Stream,2012 passed out.
Volume 3, Issue 4, April 2014
Page 200