Proceedings of The 19th Iranian conference on Biomedical Engineering (ICBME 2012), Tehran, Iran, 21-22 December 2012
Fuzzy c-means clustering based on Gaussian spatial
information for brain MR image segmentation
Abbas Biniaz,Ataollah Abbassi
Computational Neuroscience Laboratory, Sahand University
of Technology
Tabriz, Iran
CNLab@sut.ac.ir
Asac-Conventional fuzzy c-means (FCM) algorithm is
highly vulnerable to noise due to not considering the spatial
information in image segmentation. This paper aims to develop
a Gaussian spatial FCM (gsFCM) for segmentation of brain
magnetic resonance (MR) images. The proposed algorithm
uses fuzzy spatial information to update fuzzy membership
with a Gaussian function. Proposed method has less sensitivity
to noise speciically in tissue boundaries, angles, and borders
than spatial FCM (sFCM). Furthermore by the proposed
algorithm a pixel which is a distinct tissue from anatomically
point of view for example a tumor in preliminary stages of its
appearance, has more chance to be a unique cluster. The
quantitative assessment of presented FCM techniques is
evaluated by conventional validity functions. Experimental
results show the eiciency of proposed algorithm in
segmentation of MR images.
ewos-comonen; Segmenaion; I; FM; saial
infoaion.
I. INTRODUCTION
Magnetic resonance imaging (MRI) is most common
imaging modalities employed as a diagnostic technique [1].
Segmentation of medical images infered to prtition
pixels/voxels in an image into the number of 2D/3D tissues,
each with unique features and similar properties.
Segmentation process could be based on numerous features
of input data. Therefore a vriety of edge based techniques
has been developed in image segmentation. Here is a list of
edge operators which commonly is used in the image
segmentation rials: Sobel, Roberts, Prewitt, Cnny, Zero
crossing,Laplacian,and Laplacian of Gaussian( LoG ) [2,3].
There re the lrge number of gray level based approaches
for segmentation of medical images using both local and
universal image intensity information. Thresholding is one of
the image segmentation techniques and has two common
types: Global thresholding ,nd Local thresholding [4].
Mousa Shamsi, Afshin Ebrahimi.
Department of Elecrical Engineering, Sahand University of
Technology
Tabriz,Iran
{shamsi,aebrahimi}@sut.ac.ir
Region based approaches are popular segmentation
procedures. A well-developed region based method is region
growing. Based on some predeined criteria, a connected
rea is porrayed by region growing. Disadvntageous of
these methods re creation of holes and disconnectedness in
segmented image [5, 6]. Other mehods like deformable
models and active contours models (ACMs) or level set are
applied as numerical methods for racking boundries and
borders in an image [7].
Fuzzy clustering has many applications in medical image
segmentation, because they cn preserve more information
about original image using uzziness membership thn other
mehods [8]. However standrd FCM doesn't exploit spatial
information of neighborhood pixels in image segmentation.
In order to develop a modiied FCM algorithm compred
with sFCM approach [9], this paper presents a modiied
sFCM algorithm based on Gaussin spatial information as
gsFCM. New approach exracts tissue boundaries, borders,
ngles, and small organisms successully. The rest of this
paper is organized as follows: Section 2 inroduces
mehodology of his paper. Section 3 describes quantitative
validity unctions; and Section 4 presents experimental
results. Section 5 smm arizes conclusions of this paper.
II. M ETHODOLOGY
A Fuy c-Means Clustering
Fuzzy c-Mens clustering algorithms,developed in 1970s
nd optimized later [10]. Let X = { i, 2, ... , xn} denotes an
input vector with n number of elements to be ptitioned into
c (2; Sn) clusters, nd ; denotes the feature value. The
FCM algorithm is an iterative optimization process that
minimizes the following cost unction:
978-1-4673-3130-2/12/$31.00 ©2012 IEEE 154