fMRI image Segmentation using conventional methods versus Contextual Clustering by ijcsis


									                                                  (IJCSIS) International Journal of Computer Science and Information Security,
                                                  Vol. 11, No. 10, October 2013

     fMRI image Segmentation using conventional
        methods versus Contextual Clustering
                                  Suganthi D., and 2Purushothaman S.,
1                                                   2
    Suganthi D., Research Scholar,                      Dr.Purushothaman S,
Department of Computer Science,                     Professor, PET Engineering College, Vallioor, India-627117.
Mother Teresa Women’s University,
Kodaikanal, Tamilnadu, India-624101,

Abstract- Image segmentation plays a vital role in               methods that exist to segment the brain. Conventional
medical      imaging    applications.  Many      image           methods like sobel, canny, log, zerocross, and prewitt
segmentation methods have been proposed for the                  use pure image processing techniques that need
process of successive image analysis tasks in the last           human interaction for accurate and reliable
decades. The paper has considered fMRI segmentation              segmentation. Unsupervised methods, can segment
inspite of existing techniques to segment the fMRI               the brain with high precision. For this reason,
                                                                 unsupervised     methods     are    preferred     over
slices. In this paper an fmri image segmentation using
                                                                 conventional methods. Many unsupervised methods
contextual clustering method is presented. Matlab
                                                                 such as Fuzzy c-means, Finite Gaussian Mixture
software ‘regionprops’ function has been used as one of          Model, Artificial Neural Networks, etc. are available.
the criteria to show performance of CC. The CC
segmentaion shows more segmented objects with least                                II. RELATED WORK
discontuinty within the objects in the fMRI image.                         Bueno et al. 2000, described an image-based
From the experimental results, it has been found that,           method founded on mathematical morphology to
the Contextual clustering method shows a better                  facilitate the segmentation of cerebral structures on
segmentation when compared to other conventional                 3D magnetic resonance images (MRIs). Jose et al,
segmentation methods.                                            2003, descried parametric image segmentation that
                                                                 consists of finding a label field which defines a
Keywords:    Contextual clustering; segmentation; fMRI           partition of an image into a set of non overlapping
image.                                                           regions and the parameters of the models that
                                                                 describe the variation of some property within each
                   I. INTRODUCTION                               region. A Bayesian formulation is presented, based
    Image segmentation is the process of partitioning            on the key idea of using a doubly stochastic prior
/ subdividing a digital image into multiple                      model for the label field, which allows one to find
meaningful regions or sets of pixels regions with                exact optimal estimators for both this field and the
respect to a particular application. The segmentation            model parameters by the minimization of a
is based on measurements taken from the image and                differentiable function.
might be grey level, color, texture, depth or motion.                      Liu, 2006, presented a new level set based
The result of image segmentation is a set of segments            solution for automatic medical image segmentation.
that collectively cover the entire image. All the pixels         Wee et al, 2006, described accurate segmentation of
in region are similar with respect to some                       magnetic resonance (MR) images of the brain. They
characteristic or computed property, such as color,              broadly divided current MR brain image
intensity, or texture. For any object in an image, there         segmentation algorithms into three categories:
are many 'features' which are interesting points on the          classification based, region-based, and contour-based.
object that can be extracted to provide a "feature"
                                                                 They showed that by incorporating two key ideas into
description of the object. Image segmentation is done
                                                                 the conventional fuzzy C-means clustering algorithm,
using various edge detection techniques such as
Sobel, Prewitt, Roberts, Canny, LoG,                             they are able to take into account the local spatial
                                                                 context and compensate for the intensity non
    MRI Segmentation provides great importance in                uniformity (INU) artifact during the clustering
research and clinical applications. There are many               process. Xiangrong et al, 2010, described clustering

                                                                                             ISSN 1947-5500
                                                 (IJCSIS) International Journal of Computer Science and Information Security,
                                                 Vol. 11, No. 10, October 2013

algorithms in tissue segmentation in MRI. The                   obtain the gradient magnitude of the image from the
authors proposed an approach to tissue segmentation             original.
of 3D brain MRI using semi-supervised spectral                  Table 2. Sobel Mask
clustering. Yan Li and Zheru Chi, 2005, described
magnetic resonance imaging (MRI) as an advanced
medical imaging technique providing rich
information about the human soft tissue anatomy.
          The goal of magnetic resonance (MR) image                 B. Prewitt Operator: The prewitt operator is an
segmentation is to accurately identify the principal            approximate way to estimate the magnitude and
tissue structures in these image volumes. A new                 orientation of an edge. The convolution mask of
unsupervised MR image segmentation method based                 Prewitt operator is shown in Table 3.
on self-organizing feature map (SOFM) network has
                                                                Table 3. Prewitt Mask
been presented. Yongyue et al, 2001, stated that the
finite mixture (FM) model is the most commonly
used model for statistical segmentation of brain
magnetic resonance (MR) images because of its
simple mathematical form and the piecewise constant
nature of ideal brain MR images. They proposed a                    C. Roberts Operator: It performs 2-D spatial
hidden Markov random field (HMRF) model, which                  gradient measurement on an image. It highlights
is a stochastic process generated by a MRF whose                regions of high spatial frequency which often
state sequence cannot be observed directly but which            correspond to edges. The cross convolution mask is
can be indirectly estimated through observations.               shown in Table 4
          Zavaljevski et al, 2000 described MR brain            Table 4 Roberts Mask
image segmentation into several tissue classes is of
significant interest to visualize and quantify
individual anatomical structures. Zhang, 2004 stated
that image segmentation plays a crucial role in many
medical imaging applications. They presented a novel               D. Laplacian of Guassian (LoG) Operator: It is a
algorithm for fuzzy segmentation of magnetic                    second order derivative. The digital implementation
resonance imaging (MRI) data                                    of the Laplacian function is made using the mask
                                                                given in Table 5.
       III. MATERIALS AND METHODOLOGY                           Table 5Laplacian of Guassian (LoG) Operator
    The Internet Brain Segmentation Repository
(IBSR)      provides       manually-guided     expert
segmentation results along with magnetic resonance
brain image data. fMRI slice images have been
obtained from IBSR for use in this research work.
                                                                    E. Canny Operator: It is a method to find edges
Table 1 presents two figures (full slice and cropped            by isolating noise from the image without affecting
slice used for segmentation analysis).                          the features of the edges in the image and then
                   Table 1 fMRI slice                           applying the tendency to find the edges and the
                                                                critical value for threshold.
                                                                    F. Contextual clustering
                                                                    Image segmentation plays an important role in
                                                                image analysis and computer vision and it is
                                                                considered as one of the major obstruction in the
                             Region cropped used     for
                             segmentation analysis
                                                                development of image processing technology.
                                                                Recently there has been considerable interest among
                                                                researchers in statistical clustering techniques in
Full slice image                                                image segmentation was inspired by the methods of
                                                                statistical physics. These methods were developed to
   A Sobel Operator: It performs 2-D spatial                    study the equilibrium properties of large, lattice
gradient measurement on an image and so                         based systems consisting of interacting components
emphasizes regions of high spatial frequency that               as identical. In a clustering technique for image
correspond to edges. The convolution masks of Sobel             segmentation, each pixel is associated with one of the
operator are as shown Table 2, which are used to                finite number of categories to form disjoint regions.

                                                                                            ISSN 1947-5500
                                                  (IJCSIS) International Journal of Computer Science and Information Security,
                                                  Vol. 11, No. 10, October 2013

    The contextual clustering based algorithms are
assumed to be drawn from standard normal
distribution. It segments a data into category 1 (ω0)
and category 2 (ω1).
   The following are the steps adopted for
implementing the contextual clustering algorithm for
segmenting the fMRI slice image
   (i) Define decision parameter Tcc (positive) and              Fig. 1. Sample FMRI              Fig. 2 Segmentation by
weight of neighborhood information β (positive). Let             image                            ‘Sobel’ method
Nn be the total number of data in the neighborhood.
Let Zi be the data itself, i.
   (ii) Classify data with zi>Tα to ω1 and data to ω0.
Store the classification to C0 and C1.
   (iii) For each data i, count the number of data ui,
belonging to class ω1 in the neighborhood of data i.
Assume that the data outside the range belong to ω0.

    (iv) Classify data with                      to ω1           Fig 3.Segmentation by’           Fig 4.Segmentation by’
and other data to ω0. Store the classification to                Prewitt’ method                  Roberts’ method
variable C2.
    (v) If C2 ≠C1 and C2 ≠ C0, copy C1 to C0, C2 to
C1 and return to step iii, otherwise stop and return to
    The contextual clustering implementation is as
   Step 1: Read a Pattern (fmri image feature).
   Step 2: Sort the values of the pattern.                       Fig. 5 Segmentation by           Fig 6.Segmentation by
   Step 3: Find the Median of the Pattern Cm.                    ‘Log’ method                     Zero crossing method
    Step 4: Find the number of values greater than
the Median Values, Um.
    Step 5: Calculate CC using Cm + (beta/Tcc) * (Um
– (bs/2)).
   Step 6: Assign CC as the segmented values.

Earlier researchers had used different metrics to                Fig. 7 Segmentation by           Fig. 8 Segmentation by
evaluate the segmentation accuracy. In this paper, we            ‘Canny’ method                   Contextual clustering
have used ‘bwlabel’ and ‘Regionprops’ to evaluate
the accuracy of segmentation and it has been found
that CC segmentation is much better when compared
                                                                 Table 2 Outptuts of ‘bwlabel’ function of ‘Matlab 2010’
to that of remaining segmentations mentioned in this
                                                                 No        Method                            Objects detected
    Figure 1 shows fmri slice. Figures (2-8) show the
segmentation by ‘Sobel’, ‘Prewitt’, ‘Roberts’, ‘Log’,            1         Sobel                             12
‘Zero crossing’, ‘Canny’, ‘CC’ methods. Except CC
method, in all other segmentation methods, the                   2         Prewitt                           11
number of objects are more and , there are some
                                                                 3         Robertz                           8
objects segmented are not clear. Matlab ‘bwlabel’
function has been used and the number objects for                4         Log                               64
each method is shown in Table 6. In addition to
.bwlabel’, the ‘Regionprops’ command has been used               5         Zerocross                         64
to find out correct number of segmented objects

                                                                                               ISSN 1947-5500
                                                        (IJCSIS) International Journal of Computer Science and Information Security,
                                                        Vol. 11, No. 10, October 2013

6        Canny                           32                                                      Suganthi D., is pursuing her PhD
                                                                                                 from Mother Teresa Women’s
                                                                                                 University, Kodikanal, India.
7        Contextual Clustering           24

                    V. CONCLUSION
    The main purpose of proposing contextual                                                     Dr.S.Purushothaman completed his
clustering method is to improve segmentation of                                                  PhD from Indian Institute of
fMRI images. The supervised contextual clustering                                                Technology Madras, India in 1995.
                                                                                                 He has 133 publications to his credit.
extracts features from the fMRI slice that represents                                            He has 19 years of teaching
information in a given window. The algorithm                                                     experience. Presently he is working
involves least computation in the segmentation of                                                as Professor in PET Engineering
fMRI slice. The advantages of CC segmentation is                                                 college , India
that this method uses neighboring information and
assured segmentation of minimum one object of fmri
image is possible.

    1.   Bueno, G., Musse, O., Heitz F, Armspach J.P., 2000,
         3D Watershed-based segmentation of internal structures
         within MR brain images. Medical Images 2000: Image
         processing, Proc. SPIE, Vol.3979, pp.284-293.
    2.   Jose, L. Marroquin, Edgar Arce Santana, and Salvador
         Botello, 2003, Hidden Markov Measure Field Models
         for Image Segmentation, IEEE Transactions On Pattern
         Analysis And Machine Intelligence, Vol.25, No.11,
    3.   Liu, S. and Li, J., 2006, Automatic medical image
         segmentation using gradient and intensity combined
         level set method, The 28th IEEE Engineering in
         Medicine and Biological Society, Vol. 1, pp. 3118–
    4.   Wee, A., Liew, C., and. Yan, H., 2006, Current
         Methods in the Automatic Tissue Segmentation of 3D
         Magnetic Resonance Brain Images, Current Medical
         Imaging Reviews, Vol.2, No.1, pp.91-103.
    5.   Xiangrong Zhang, Feng Dong, Gordon Clapworthy,
         Youbing Zhao and Licheng Jiao, 2010, Semi-
         supervised Tissue Segmentation of 3D Brain MR
         Images, 14th International Conference Information
         Visualization, pp.623-628.
    6.   Yan Li and Zheru Chi, 2005, MR Brain Image
         Segmentation Based on Self-Organizing Map Network,
         International Journal of Information Technology,
         Vol.11, No. 8, pp.45-53.
    7.   Yongyue Zhang, Michael Brady, and Stephen Smith,
         2001, Segmentation of Brain MR Images Through a
         Hidden Markov Random Field Model and the
         Expectation-Maximization         Algorithm,      IEEE
         Transactions On Medical Imaging, Vol. 20, No. 1,
    8.   Zavaljevski, A., Dhawan, A.P., Gaskil, M., Ball, W.,
         Johnson J.D., 2000, Multi-level adaptive segmentation
         of multi-parameter MR brain images. Comput Med
         Imag Graphics, Vol. 24, Issue 2, pp. 87–98
    9.   Zhang, D.Q., Chen, S.C., 2004, A novel kernelized
         fuzzy c-means algorithm with application in medical
         image segmentation, Artificial Intelligence Medicine,
         Vol. 32, Issue 1, pp. 37–50.

                                                                                                   ISSN 1947-5500

To top