978-1-4799-3903-9/14/$31.00 ©2014 IEEE 511 ICALIP2014
MRI Brain Image Segmentation Based on Kerneled
FCM Algorithm and Using Image Filtering Method
Tian Lan, Zhe Xiao, Changsong Hu, Yi Ding, Zhiguang Qin
School of Computer Science and Engineering
University Of Electronic Science And Technology Of China
Chengdu, China
Lantian1029@uestc.edu.cn
Abstract—Image segmentation plays a preliminary and
indispensable step in medical image processing. Image
segmentation plays a crucial role in many medical imaging
applications. In this paper, we present a novel algorithm called
image Filtering spatial Kernel Fuzzy C-Means(FKFCM) for
fuzzy segmentation of magnetic resonance imaging (MRI)data.
The algorithm is realized by modifying the objective function in
the conventional Fuzzy C-Means (FCM) algorithm using a
kernel-induced distance metric, a spatial penalty on the
membership functions and then using the image filtering method
to correct the image. The algorithm results are compared with
standard FCM, Kerneled Fuzzy C-Means (KFCM) and Spatial
Fuzzy C-Means(SFCM). The performance of the proposed
segmentation algorithm FKFCM provides satisfactory results
compared with other algorithms.
Keywords—Image segmentation; Fuzzy C-means; Kernel
method; Spatial Kernel method; FKFCM
I. INTRODUCTION
With the increasing size and number of using medical
images, the use of computers in facilitating their processing
and analyses has become necessary. In particular, as a task of
delineating anatomical structures and other regions of interest,
image segmentation algorithms play a vital role in numerous
biomedical imaging applications such as the quantification of
tissue volumes, diagnosis, study of anatomical structure, and
computer-integrated surgery. Classically, image segmentation
is defined as the partitioning of an image into non-overlapping
constituent regions which are homogeneous with respect to
some characteristics such as intensity or texture.
Because of the advantages of magnetic resonance imaging
(MRI) over other diagnostic imaging, the majority of
researches in medical image segmentation pertains to its use
for MR images, and there are a lot of methods available for
MR image segmentation. Among them, fuzzy segmentation
methods are of considerable benefits, because they could
retain much more information from the original image than
hard segmentation methods. In particular, the fuzzy C-means
(FCM) algorithm[1], 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
inhomogeneity 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. To
deal with the inhomogeneity problem, many algorithms have
been proposed by adding correction steps before segmenting
the image or by modeling the image as the product of the
original image and a smooth varying multiplier field. Recently,
many researchers have incorporated spatial information into
the original FCM algorithm to better segment the images.
Tolias et al.[2] proposed a fuzzy rule-based system to impose
spatial continuity on FCM, and in another paper [3],they used
a small positive constant to modify the membership of the
centre pixel in a
33u
window. Pham et al.[4]modified the
objective function in the FCM algorithm to include a
multiplier field containing the first and second order
information of the image. Similarly, Ahmed et al. [5] proposed
an algorithm to compensate for the intensity inhomogeneity
and to label a pixel by considering its immediate
neighborhood. A rather recent approach proposed by Pham [6]
is to penalize the FCM objective function to constrain the
behavior of the membership functions, similar to methods
used in the regularization and Markov random field (MRF)
theory.
On the other hand, there is a trend in recent machine
learning work to construct a nonlinear version of a linear
algorithm using the 'kernel method', e.g., SVM , KPCA and
KFD. And this 'kernel method' has also been applied to
unsupervised clustering [7].However, a drawback of these
kernel clustering algorithms using the dual representation for
clustering prototypes (that is, each prototype is formulated as a
linear sum of after-mapped dataset elements, and hence the
parameters to be optimized are not original prototypes
anymore but linearly-combined coefficients) is that the
clustering prototypes lie in high dimensional feature space and
hence clustering results lack clear and intuitive descriptions as
in the original space. Then a kerneled fuzzy C-means (KFCM)
algorithm is proposed to compensate for such a lack and then
applied to the MR image segmentation[7]. It is realized by
replacing the original Euclidean distance in the FCM
algorithm with a kernel-induced distance and adding a novel
spatial penalty also.In this paper, we present a novel algorithm