Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Look up keyword
Like this
2Activity
0 of .
Results for:
No results containing your search query
P. 1
A MULTIPLE KERNEL FUZZY C-MEANS CLUSTERING ALGORITHM FOR BRAIN MR IMAGE SEGMENTATION

A MULTIPLE KERNEL FUZZY C-MEANS CLUSTERING ALGORITHM FOR BRAIN MR IMAGE SEGMENTATION

Ratings: (0)|Views: 12|Likes:
Published by P Singh Ijaet
In spite of its computational efficiency and wide spread popularity, the FCM algorithm does not take the spatial information of pixels into consideration. In this paper, a multiple kernel fuzzy c-means clustering (MKFCM) algorithm is presented for fuzzy segmentation of magnetic resonance (MR) images. By introducing a novel adaptive method to compute the weights of local spatial values in the objective function, the new multiple kernel fuzzy clustering algorithm is capable of utilizing local contextual information to impose local spatial continuity, thus improving the classification accuracy and reduces the number of iterations. To estimate the intensity in homogeneity, the global intensity is introduced into the coherent local intensity clustering algorithm. Our results show that the proposed MKFCM algorithm can effectively segment the test images and MR images. Comparisons with other FCM approaches based on number of iterations and time complexity demonstrate the superior performance of the proposed algorithm.
In spite of its computational efficiency and wide spread popularity, the FCM algorithm does not take the spatial information of pixels into consideration. In this paper, a multiple kernel fuzzy c-means clustering (MKFCM) algorithm is presented for fuzzy segmentation of magnetic resonance (MR) images. By introducing a novel adaptive method to compute the weights of local spatial values in the objective function, the new multiple kernel fuzzy clustering algorithm is capable of utilizing local contextual information to impose local spatial continuity, thus improving the classification accuracy and reduces the number of iterations. To estimate the intensity in homogeneity, the global intensity is introduced into the coherent local intensity clustering algorithm. Our results show that the proposed MKFCM algorithm can effectively segment the test images and MR images. Comparisons with other FCM approaches based on number of iterations and time complexity demonstrate the superior performance of the proposed algorithm.

More info:

Categories:Types, Research, Science
Published by: P Singh Ijaet on Nov 07, 2012
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

12/04/2012

pdf

text

original

 
International Journal of Advances in Engineering & Technology, Nov. 2012.©IJAET ISSN: 2231-1963406 Vol. 5, Issue 1, pp. 406-415
A
 
M
ULTIPLE
K
ERNEL
F
UZZY
C-M
EANS
C
LUSTERING
A
LGORITHM FOR
B
RAIN
MR
 
I
MAGE
S
EGMENTATION
 
M. Ganesh
1
and V. Palanisamy
2
 
1
Assistant Professor, Department of ECE
,
Info Institute of Engineering, Coimbatore, India
2
Principal, Info Institute of Engineering, Coimbatore, India
 A
 BSTRACT 
 
 In spite of its computational efficiency and wide spread popularity, the FCM algorithm does not take the spatialinformation of pixels into consideration. In this paper, a multiple kernel fuzzy c-means clustering (MKFCM)algorithm is presented for fuzzy segmentation of magnetic resonance (MR) images. By introducing a noveladaptive method to compute the weights of local spatial values in the objective function, the new multiple kernel fuzzy clustering algorithm is capable of utilizing local contextual information to impose local spatial continuity,thus improving the classification accuracy and reduces the number of iterations. To estimate the intensity inhomogeneity, the global intensity is introduced into the coherent local intensity clustering algorithm. Our results show that the proposed MKFCM algorithm can effectively segment the test images and MR images.Comparisons with other FCM approaches based on number of iterations and time complexity demonstrate thesuperior performance of the proposed algorithm.
 K 
 EYWORDS
:
 
Fuzzy C-means (FCM), Image segmentation, Kernel function, Multiple kernel
 
 fuzzy c-meansclustering (MKFCM), Magnetic resonance (MR) imaging.
I.
 
I
NTRODUCTION
 
Magnetic resonance (MR) imaging has several advantages over other medical imaging methods,including high contrast among different soft tissues, relatively high spatial resolution across the entirefield of view and multi-spectral characteristics. Therefore, it has been widely used in quantitativebrain imaging studies. Quantitative volumetric measurement and three-dimensional (3D) visualizationof brain tissues are helpful for pathological evolution analyses, where image segmentation plays animportant role. The size alterations in brain tissues often accompany various diseases, such asschizophrenia [1]. Thus, estimation of tissue sizes has become an extremely important aspect of treatment which should be accomplished as precisely as possible. This creates the need to properlysegment the brain MR images into gray matter (GM), white matter (WM) and cerebrospinal fluid(CSF) and also to identify tumors or lesions, if present [2]. The main difficulties in brain segmentationare the intensity inhomogeneities and noise. In fact, intensity inhomogeneity occurs in many real-world images from different modalities [3, 4]. In particular, it is often seen in medical images, such asX-ray radiography/ tomography and MR images. For example, the intensity variation across theimage, which arises from radio-frequency (RF) coils or acquisition sequences. Thus the resultantintensities of the same tissue vary with the locations in the image. The noise in MR images is Riciandistributed and can affect significantly the performances of classification methods. The best solutions
 
International Journal of Advances in Engineering & Technology, Nov. 2012.©IJAET ISSN: 2231-1963407 Vol. 5, Issue 1, pp. 406-415
consist of either filtering the image prior to classification or embedding spatial regularization insidethe classifier itself.This paper is organized into six sections. Section 1 gives introduction to the proposed work, Section 2deals with Conventional FCM clustering algorithm, Section 3 deals with Kernel Fuzzy C-meansalgorithm, Section 4 describes the proposed method of Multiple kernel fuzzy c-means clustering,Section 5 describes the quantitative comparison of the accuracy of those segmentation results withother segmentation algorithms and finally conclusion is presented in section 6.
II.
 
R
ELATED
W
ORK
 Image segmentation is an important and challenging problem and a necessary first step in imageanalysis as well as in high-level image interpretation and understanding such as robot vision, objectrecognition, and medical imaging. The goal of image segmentation is to partition an image into a setof disjoint regions with uniform and homogeneous attributes such as intensity, colour, tone or texture,etc. Many different segmentation techniques have been developed and detailed surveys can be foundin references [21
 – 
23].Segmentation subdivides an image into different regions or objects based on the information foundabout objects in imaging data. In the segmentation of medical images, the objective is to identifydifferent regions, organs and anatomical structures from data acquired via MRI or other medicalimaging technique. Initially segmentation has been done based manually by human experts. Butmanual segmentation is a difficult and time consuming task, which makes an automated breast cancersegmentation [15] method desirable. The automated segmentation [16] of MR images into anatomicaltissues, fluids, and structures is an interesting field in medical image analysis. Automatedsegmentation methods based on artificial intelligence techniques were proposed by [17], [18]. Amethod that detects deviations from normal brains using a multilayer Markov random field framework [19]. In the last decades, fuzzy segmentation algorithms, especially the fuzzy c-means algorithm(FCM), have been broadly used in the image segmentation [9] and such a success mostly attributes tothe introduction of fuzziness for the belongingness of each image pixel. Fuzzy c-means [14] allowsfor the ability to make the clustering methods able to retain more information from the original imagethan the crisp or hard segmentation methods [20]. Clustering is used to panel a set of given observedinput data vectors or image pixels into clusters so that components of the same cluster is similar toone another than to members of other clusters where the number of clusters is usually predefined orset by some weight criterion or a priori knowledge. Fuzzy c-means segmentation methods are havingsignificant profit in segmentation of medical images [10], because they could retain a moreinformation from the original image than hard c-means segmentation methods. The main advantage infuzzy c-means algorithm is it allows pixels to belong to multiple clusters with reasonable degrees of membership grades. However, there are some disadvantages in using fuzzy c-means; the membershipof an object has not strong enough or significantly high for a particular cluster, it means that theequation of calculating membership is not an effective, and sometimes the equation for updatingprototypes has incapable to work with data which greatly affected by noise. Thus the equation forupdating prototypes leads the result of clustering might be uncorrected. The main reason forunderlying drawbacks of above is, fuzzy c-means employs based on existed Euclidean distancemeasures.Computer aided brain tumor segmentation system is an important application in medical imageanalysis [27]. Developing a medical image analysis system not only can lighten the workload anddecrease the errors of the doctors, but also can provide a quantitative measure about variation of thebrain tumor throughout its whole therapeutic treatment. However, it is still a difficult problem toautomatically segment brain tumor regions from MRI multi-sequences because of many existing typesof tumors with morphological variability, a variety of shapes and appearance properties amongindividuals, the deformation near the structures in the brain which results in an abnormal geometryalso for healthy tissues, and lack of prior knowledge about them [26]. Therefore, it is practicallymeaningful to focus on semi-automatic or fully-automatic segmentation methods on multiple MRIscans for medical research, disease monitoring, therapeutic control and so on. Different MRIsequences from different excitations can respectively provide different and partly independentinformation about different tissues, and reflect pathologic information about the tumors in the brain.
 
International Journal of Advances in Engineering & Technology, Nov. 2012.©IJAET ISSN: 2231-1963408 Vol. 5, Issue 1, pp. 406-415
As a tumor consists of different biologic tissues, one type of MRI cannot give complete informationabout abnormal tissues. Combining different complementary information can enhance thesegmentation of the tumors. Therefore, radiology experts always combine the multi-spectral MRIinformation of one patient to make a decision on the location, extension, prognosis and Clustering is aprocess of classifying objects or patterns in such a way that the samples in the same group are moresimilar than the samples in different groups. Based on the fuzzy theory, the fuzzy clustering method[5], which produces the idea of partial membership of belonging. As a soft clustering method, fuzzyclustering has been extensively studied and successfully applied to image segmentation. One of themost important and widely used fuzzy clustering methods is the fuzzy c-means (FCM) algorithm [6],and promoted as the general FCM clustering algorithm [7]. The main purpose of the FCM algorithmis to make the vector space of a sample point be divided into a number of sub-spaces in accordancewith a distance measure [8]. However, the FCM algorithm does not take the local spatial property of images into consideration, and hence suffers from high sensitivity to noise. To improve its robustness,many modifications to the FCM algorithm that incorporate spatial information into clustering havebeen proposed. The FCM objective functions with a penalty term and resulted in spatially smoothedmembership functions.
III.
 
C
ONVENTIONAL
FCM
 
C
LUSTERING
A
LGORITHM
 
Multiresolution segmentation is a bottom up region merging technique starting with one-pixel objects.In numerous subsequent steps, smaller image objects are merged into bigger ones. Throughout thispair wise clustering process, the underlying optimization procedure minimizes the weightedheterogeneity of resulting image objects, where n is the size of a segment and h an arbitrary definitionof heterogeneity [3]. The algorithm is developed by modifying the objective function of the standardFCM algorithm with a penalty term that takes into account the influence of the neighboring pixels onthe centre pixels [25]. In each step, that pair of adjacent image objects is merged which stands for thesmallest growth of the defined heterogeneity. If the smallest growth exceeds the threshold defined bythe scale parameter, the process stops. Doing so, multi-resolution segmentation is a local optimizationprocedure. The entropy based methodology for segmentation of satellite images is performed asfollows. Images are divided into square windows with a fixed size L, the entropy is calculated foreach window, and then a classification methodology is applied for the identification of the category of the respective windows. The classification approach can be supervised or non-supervised. Supervisedclassification needs a training set composed by windows whose classes are previously known(prototypes), such as rural and urban areas.Given a data set X
= {x1,x2,…….x
n
}, where the data point
x
 j
,R
p
(j = 1, . . . , n),
n
is the number of data, and
 p
is the input dimension of a data point, traditional FCM [3] groups
 X 
into
c
clusters byminimizing the weighted sum of distances between the data and the cluster centers or prototypesdefined as
 
21 1
c nmij j ii j
Q u x o
 
(1)
Here, is the Euclidean distance. u
ij
is the membership of data
 j
 x
belonging to cluster
i
, which isrepresented by the prototype
i
o
.
The constraint on
ij
u
is
1
1
ciji
u
and m is the fuzzificationcoefficient.
IV.
 
K
ERNEL
F
UZZY
C-M
EANS
A
LGORITHM
 
When applying the KFCM framework in image-segmentation problems, the multiresolutionsegmentation may end up with local optimization procedure. Global mutual fitting is the strongestconstraint for the optimization problem and it reduces heterogeneity most over the scene following apure quantitative criterion. Its main disadvantage is that it does not use the treatment order and buildsfirst segments in regions with a low spectral variance leading to an uneven growth of the imageobjects over a scene. It also causes an unbalance between regions of high and regions of low spectral

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->