International Journal of Fuzzy Logic Systems (IJFLS) Vol.2, No.

3, July 2012
DOI : 10.5121/ijfls.2012.2305 51
A Hybrid Approach for DICOMImage Segmenta-
tion Using Fuzzy Techniques
J. Umamaheswari
1
and G.Radhamani
2
,
1
Research Scholar, Department of Computer Science, Dr.G.R.D. College of Science,
Coimbatore,Tamilnadu,India.
umamugesh@yahoo.com
2
Director, Department of Computer Science, Dr.G.R.D. College of Science,
Coimbatore,Tamilnadu,India.
radhamanig@hotmail.com
ABSTRACT
This paper provides an hybrid method for segmentation using FCM thresholding with fuzzy levelset . FCM
thresholding gives fine segmented results when compared to edge detection method. The optimization prop-
erty of FCM is improved when it is combined with local thresholding. The application of Fuzzy levelset
gives enhanced segmentation results. The experimental result proves that the hybrid method is efficient in
statistical metrices that enhances the segmented results with fine regions.
KEYWORDS
Medical Image segmentation, FCM, Local thresholding, Levelset, Fuzzy levelset, Adaptive thresholding.
1. INTRODUCTION
Segmentation of images holds an important position in the area of image processing and widely
used in medical applications which includes surgical planning, abnormality detection and treat-
ment progress monitoring. The purpose of segmentation is to partition an image into distinct, se-
mantically meaningful entities by defining boundaries between features and objects in an image
based on some constraint, or homogeneity predicate. Computer aided detection of abnormal
growth of tissues is primarily motivated by the necessity of achieving maximum possible accura-
cy. The various hybrid models for medical image segmentation is explained in [1,4].
There are many methods available using levelset segmentation [9,10,13,14]. Such algorithms em-
ploy fuzzy clustering, based on image intensity, for initial segmentation and employ levelset me-
thods for object refinement by tracking boundary variation. The previous work on liver tumor
segmentation [9] has based on fuzzy clustering. These methods approximately delineate a tumor
boundary which not only relieves manual intervention, but also accelerates levelset optimization.
Ho and Suri, on the other hand, proposed to regularize levelset evolution locally by fuzzy cluster-
ing, in order to deviate the problems of noises sensitivity and weak boundaries of medical images
[10,13,14].
An optimized approach that uses a variational method of image segmentation is described to mi-
nimize energy by weighted Total Variation (TV) method is discussed in [2,3]. This model de-
scribes the Fuzzy C-means method (FCM) to obtain segmentation via fuzzy pixel classification.
FCM allows pixels to fit in multiple classes with varying degrees of membership than hard classi-
International Journal of Fuzzy Logic Systems (IJFLS) Vol.2, No.3, July 2012
52
fication methods that normally makes pixel to fit in one class. This approach allows additional
flexibility in many applications and has recently been used in processing of magnetic resonance
image (MRI) and Computer Topography (CT) image. Threshold segmentation is widely used in
many fields because of its simplicity and efficiency which is most frequently used for image seg-
mentation. In our work the integration of FCM thresholding with fuzzy levelset is analyzed. The
segmented results are proved to be efficient compare to other traditional method.
The paper is organized as follows, section 2 discusses about the optimal method for DICOM im-
age segmentation. Section 3 describes the experimentation and results. Finally the section 4 in-
cludes the conclusion and references.
2. OVERVIEW OF THE OPTIMAL APPROACH FOR DICOM
IMAGE SEGMENTATION
The optimal approach consists of FCM thresholding with fuzzy levelset process. The FCM al-
gorithm that incorporates spatial information into the membership function is used for clustering,
while a conventional FCM algorithm does not fully utilize the spatial information in the image.
The advantages of the algorithm are its less sensitivity to noise and consider regions more homo-
geneous compared to other methods. So fuzzy levelset by Bing Nang Li [15] is used for the en-
hancement of the segmentation results. The optimal approach gives better results for DICOM im-
age segmentation.
2.1 FCM Thresholding
The fuzzy c means clustering method with thresholding approach is based on the principles of
fuzzy algorithm. It consists of three components namely, Fuzzy clustering, C -Means and Thre-
sholding
a. Fuzzy clustering:
The goal of a clustering analysis is to divide a given set of image data or objects into a cluster,
which represents subsets or a group. The partition should have two properties such as homogenei-
ty and heterogeneity. The homogeneity cluster consists of similar data whereas the heterogene-
ity cluster contains different data. The membership functions do not reflect the actual data distri-
bution in the input and the output spaces. They may not be suitable for fuzzy pattern recognition.
To build membership functions from the data available, a clustering technique may be used to
partition the data, and then produce membership functions from the resulting clustering. In
our method C-means clustering is used. It is a simple unsupervised learning method which
can be used for data grouping or classification when the number of the clusters is known. It con-
sists of the following steps:
Step 1: Choose the number of clusters K
Step 2: Set initial centers of clusters C1, C2,, C.
Step 3: Classify each vector
1 2 n
[ , ,.... ]
T
l l l l x x x x

into the closest center

ci
using Euclidean
distance measure:
c x c x i i i i
min
Step 4: Recompute the estimates for the cluster centers
ci
.
Let
ci
=
1 2 ln
[ , ,.... ]
T
l l c c c cim
be computed by:
N
i x
c
i
x
li
im
cluster

) (
lim
where
Ni
is the number of vectors in the i-th cluster.
International Journal of Fuzzy Logic Systems (IJFLS) Vol.2, No.3, July 2012
53
Step 5: If none of the cluster centers (
ci
=1, 2,, k) changes in step 4 stop process, otherwise
go to step 3.
b. C-means algorithm: The criterion function used for the clustering process is:
, ) (
2
1
v x c k i k
v J
i
n
k
x

where
vi
is the sample mean or the center of samples of cluster i, and v={v1,v2,,vc}.
Clusters are not completely disjoint and the data could be classified as belonging to one cluster
almost as well to another. Therefore, the separation of the clusters becomes a fuzzy notion, and
representation of the data can be more accurately handled by fuzzy clustering methods. It is ne-
cessary to describe the data in terms of fuzzy clusters. The criterion function used for fuzzy C-
means clustering is
, ) (
2
1 1
v x u
i k
v J
n
k
m
ik
c
i

where:

x x n
,......
1
n data sample vectors;

v v c
,......
1
c denotes cluster centers (centroids);
u u ik
cxm matrix, where
uik
is the i-th membership value of the k-th input sample x
k
, and the
membership values satisfy the following conditions:
1
1
0 1 ; 1 , . . . . . . . ; 1 , . . . . . . , ;
1 ; 1 , . . . . . ;
0 1 ; 1 , . . . . . . ;
i k
C
i k
i
n
i k
k
i c k n
k n
i c
U
U
U

j ) 1, m is an exponent weight factor

c. Thresholding
Local threshold method is used to automatically perform histogram shape-based image threshold-
ing or, the reduction of a graylevel image to a binary image. The algorithm assumes that the im-
age to be threshold contains two classes of pixels or bi-modal histogram (e.g. foreground and
background) then calculates the optimum threshold separating those two classes so that their
combined spread is minimal. Through FCM method the segmented part cannot be seen visibly.
Here FCM is used based on local thresholding. The segmentation image is made visible through
optimized method.
2.2 Fuzzy Levelset for Image Segmentation
This approach is based on the active [5-8] fuzzy model with the integration of adaptive region in-
formation to obtain a robust segmentation model. The level set method was first introduced by
Osher and Sethian [17]. The level set method is a numerical and theoretical tool for propagating
interfaces. Both FCM algorithms and level set methods are general-purpose computational mod-
International Journal of Fuzzy Logic Systems (IJFLS) Vol.2, No.3, July 2012
54
els that can be applied to problems of any dimension. However, when applied to medical image
segmentation, it helps to consider the specific circumstances for better performance. A new fuzzy
level set algorithm is thereby proposed for automated medical image segmentation [15]. It begins
with spatial fuzzy clustering, whose results are utilized to initiate level set segmentation, estimate
controlling parameters and regularize level set evolution. It employs an FCM with spatial restric-
tions to determine the approximate contours of interest in a medical image. Benefitting from the
flexible initialization as in levelset, the enhanced level set function can accommodate FCM re-
sults directly for evolution.
Suppose the component of interest in an FCM results, the level set function of the closed front
C is defined as follows, [16]
, ) , ) , ) , , , x y d x y C r
Where d((x, y),C) is the distance from point (x, y) to the contour C, and the sign plus or minus are
chosen if the point (x, y) is inside or outside of interface C. The interface is now represented im-
plicitly as the zero level set (or contour) of this scalar function
, ) , ) [ }
, , C x y x y
Such an implicit representation has numerous advantages over a parametrical approach. The level
set evolution equation is given by
, )
, ) , )
, ) , )
, ) , ) , ) , )
2 2
,
,
,
1 0
, ,
c
k x y
x y
x y
t
I x y I x y

' '

`

' '

Where 0 and 1 are the mean of the image intensity within two subsets inside or outside the con-
tours respectively. The final segmented image can be represented as a set of piece-wise constants.
2.3 Fuzzy Thresholding and Fuzzy Levelset Method
While applying normal thresholding to the edges, the region of the image is not segmented prop-
erly and hence the excess regions are removed by applying FCM method. An FCM algorithm
which is the general-purpose computational model is applied to DICOM CT image segmentation,
for the better results. An image can be represented in various feature spaces, and the FCM algo-
rithm classifies the image by grouping similar data points in the feature space into clusters. The
resultant image obtained by FCM consists of less spatial information. So to improve that and to
separate the segmented part from the spatial region, the local thresholding is applied. The image
obtained by fuzzy thresholding consist of some deblurred edges and some incorrect segmented
part. To improve the segmentation the fuzzy levelset method is used. The levelset function will
automatically slow the evolution down and will become totally dependent on the smoothing term.
Since a conservative levelset evolution is adopted here, it stabilizes automatically. For robust
segmentation a comparatively large iteration of evolution is adopted. The levelset evolution is ap-
plied in order to avoid insufficient or excessive segmentation.
3. EXPERIMENTATION AND RESULTS:
DICOM medical images are taken as test images for evaluating results. Here an average of ten
images is taken for evaluation. The algorithm is tested in MATLAB. The reconstruction of an im-
International Journal of Fuzzy Logic Systems (IJFLS) Vol.2, No.3, July 2012
55
age has the dimensions of 256 pixel intensity. The DICOM CT images in this contain a wide va-
riety of subject matters and textures. Most of the images used are brain images with normal and
abnormal images. The results are estimated statistically based on The Energy, Entropy and Mu-
tual Information (MI)
To test the accuracy of the segmentation algorithms, four steps are followed.
i) First, a DICOM CT image is taken as test input.
ii) Second, the different segmentation method is applied to a DICOM image.
iii)Third, the performance evaluation is obtained based on the statistical measures like Energy,
Entropy and MI.
a. Energy: The gray level energy indicates how the gray levels are distributed. It is formulated
as,

x
i
x p x E
1
) ( ) (
where E(x) represents the gray level energy with 256 bins and p(i) refers to the probability distri-
bution functions, which contains the histogram counts. The larger energy value corresponds to the
lower number of gray levels, which means simple. The smaller energy corresponds to the higher
number of gray levels, which means complex. The following table 1 shows the different parame-
tric evaluation for segmentation algorithm.
Table 1. Parametric Evaluation for Different Segmentation algorithms
Method Energy Entropy MI
Adaptive threshold 0.6719 0.5097 0.2003
Edge 0.8987 0.2087 0.0671
Fuzzy Threshold 0.6932 0.4851 7.38E-01
Fuzzy Levelset 0.6934 0.4952 6.24E-01
HM 0.90E-01 1.48E-00 1.43E+00
b) Entropy: Suppose that two discrete probability distributions of the images have the probabili-
ty functions of p and q, the relative entropy of p with respect to q is then defined as the summa-
tion of all possible states of the system, which is formulated as,
) (
) (
log ) (
2
1
i q
i p
i p d
k
i

The following figure 1 shows the energy value for segmentation algorithm. The hybrid method
gives high energy value compare to other methods.
International Journal of Fuzzy Logic Systems (IJFLS) Vol.2, No.3, July 2012
56
Fig. 1 Energy Value of Different Segmentation Algorithm
Fig. 2 Entropy Value for Different Segmentation Algorithm
The above figure 2 shows the entropy value for segmentation algorithm. The hybrid method gives
less entropy value compared to other methods.
c) Mutual Information (MI) The notion of the mutual information can be applied as another ob-
jective metric. The mutual information acts as a symmetric function, which is formulated as,
) | ( ) , (
) ) (
) , (
log ) , ( ) ( log ) (
) ( ) (
) , (
log ) , ( ) , (
2
.
2
2
Y X H X H
Y P X P
Y X P
Y X P X P X P
Y P X P
Y X P
Y X P Y X I
y x
xy
y x
xy
x
x
y x
xy
X Y
X Y

-

where I(X; Y) represents the mutual information; H(X) and H(X| Y) are entropy and conditional
entropy values. It is interpreted as the information that Y can tell about X and the measure of re-
duction in uncertainty of X due to the existence of Y. At the same time, it also shows the relation-
ship of the joint and product distributions. The following figure 4 shows the MI value for segmen-
tation algorithm. The hybrid method gives high MI value compared to other methods.
International Journal of Fuzzy Logic Systems (IJFLS) Vol.2, No.3, July 2012
57
Fig.3. MI Value for Different Segmentation Algorithm
(a) Input Image (b) Adaptive Threshold (c) Edge detection
(d) Fuzzy Threshold (e) Fuzzy Level Set (f) Hybrid Method
Fig. 5 Image Results for Different Segmentation Algorithm
The above figure 5 shows the image results for different segmentation algorithm. The hybrid me-
thod gives the suitable segmented results with fine regions.
4. CONCLUSION
In this paper an optimal method for DICOM CT image segmentation is explored with the integra-
tion of FCM thresholding with fuzzy levelset for medical image processing FCM thresholding
gives fine segmented results when compared to Otsu method. The optimization property of FCM
is improved when it is combined with local thresholding. The application of Fuzzy levelset gives
enhanced segmentation results. The experimentation results based on the statistical metrices
proves that the optimal approach enhances the segmented results with fine regions.
International Journal of Fuzzy Logic Systems (IJFLS) Vol.2, No.3, July 2012
58
REFERENCES:
[1] Yang Feng, Xiaohuan, An Improved Hybrid Model for Medical Image Segmentation, Proceeding of
IEEE conference (ICCS 2008), P.367-370, 2008.
[2] Indah Soesanti, Adhi Susanto, Thomas Sri Widodo, Maesadji Tjokronagoro, Optimized Fuzzy Logic
Application for MRI Brain Images Segmentation, International Journal of Computer Science & In-
formation Technology (IJCSIT) Vol 3, No 5, P.137-146, 2011.
[3] Indah Soesanti, Adhi Susanto, Thomas Sri Widodo, Maesadji Tjokronagoro, MRI Brain Images Seg-
mentation Based on Optimized Fuzzy Logic and Spatial Information, International Journal of Video
& Image Processing and Network Security IJVIPNS-IJENS Vol: 11, No: 04, P.6-10, 2011.
[4] S. Zulaikha Beevi, M. Mohamed Sathik, An Effective Approach for Segmentation of MRI Im-ages:
Combining Spatial Information with Fuzzy C-Means Clustering, European Journal of Scien-tific Re-
search Vol.41, No.3, P.437- 451,2010.
[5] Youness Aliyari Ghassabeh, Nosratallah Forghani, Mohamad Forouzanfar, Mohammad Teshnehlab,
MRI Fuzzy Segmentation of Brain Tissue Using IFCM Algorithm with Genetic Algorithm Optimiza-
tion, proceeding of IEEE International Conference, P.665 - 668,2007.
[6] N. Senthilkumaran and R. Rajesh, Brain Image Segmentation, International Journal of Wisdom Based
Computing, Vol. 1, No.3, P.14-18, 2011.
[7] Shafaf Ibrahim, Noor Elaiza Abdul Khalid, and Mazani Manaf, Seed-Based Region Growing (SBRG)
vs Adaptive Network-Based Inference System (ANFIS) vs Fuzzy C-Means (FCM): Brain Abnormali-
ties Segmentation, International Journal of Electrical and Computer Engineering, Vol. 5, No.2, P.94-
104, 2010.
[8] G.Wiselin Jiji, L.Ganesan, Unsupervised Segmentation using Fuzzy Logic based Texture Spectrum
for MRI Brain Images, World Academy of Science, Engineering and Technology, Vol. 5, 2005,P.155-
157.
[9] Jun Zhang, and Jinglu Hu, Image Segmentation Based on 2D Otsu Method with Histogram Analy-sis,
2008 International Conference on Computer Science and Software Engineering, IEEE 2008, P.105-
108.
[10] J.S.Suri, K.Liu, S.Singh, S.N.Laxminarayan, X.Zeng, L.Reden, Shape recovery algorithms using le-
velset sin 2-D/3-D medical imagery: a state-of-the-art review, IEEE Transactions on Informa-tion
Technology in Biomedicine P.828, 2002.
[11] N.Paragios, A levelset approach for shape-driven segmentation and tracking of left ventricle, IEEE
Transactions on Medical Imaging, P.773776, 2003.
[12] I.M.Mitchell, The flexible, extensible and efficient toolbox of levelset methods, Journal of Scien-tific
Computing, P- 300-339, 2008.
[13] J.S.Suri, Two-dimensional fast magnetic resonance brain segmentation, IEEE Engineering in Med-
icine and Biology Magazine Vol.20, P.8495, 2001.
[14] S.Ho, E.Bullitt, G.Gerig, Levelset evolution with region competition: automatic 3-D segmentation of
brain tumors, In Proceedings of the International Conference on Pattern Recogni-
tion(ICPR02).,P.532535, 2001.
[15] Bing NanLi, CheeKongChui, StephenChang, S.H.Ong, Integrating spatial fuzzy clustering with level
set methods for automated medical image segmentation, Vol.41, P.1-10, 2011.
[16] B.N.Li, C.K.Chui, S.H.Ong, S.Chang, Integrating FCM and levelsets for liver tumor segmentation, in
Proceedings of the 13th International Conference on Biomedical Engineering,(ICBME2008), P.202
205,2009.
[17] J. A. Sethian and Peter Smereka, Levelset Methods for Fluid Interfaces, Annu. Rev. Fluid Mech.
Vol.35, P.341372, 2003.