Brain Tumor Segmentation and 3D Visualization

using Fluid Vector Flow

B.Vijayakumar, B.E., M.Tech.,(PhD) K.Muthu Kumar,B.Tech.,(M.Tech)
Department of computer science and engg Department of computer science and engg
AP, PSN Engineering College Student, PSN College of Engineering
vijaymtech29@gmail.com conferancemuthu@gmail.com

Abstract Medical Imaging is perceived to designate the set of techniques that
noninvasively produce images of the internal aspect of the body. MRI is more suitable for
imaging the brain due to its ability of discriminating soft tissues with no side effects.
Manual segmentation of brain tumors by medical practitioners is a time consuming task and
has inability to assist in accurate diagnosis. Several automatic methods have been developed
to overcome these issues. Tumors doesnt exhibit any shape and causes deformations of
nearby structures of the brain. Active contour models play a vital role in accurately
delineating the boundary of the tumor. Several active contour models such as traditional
snake, Gradient Vector Flow, Boundary Vector Flow, Magnetostatic Active Contour
models, Fluid Vector Flow are in existence of which Fluid Vector Flow addresses problems
of insufficient capture range and poor convergence for concavities faced by the other
models. Fluid vector flow is insensitive to initialization that make it attractive one for
segmenting brain tumor. Fluid Vector Flow is utilized for segmentation of two dimensional
brain tumor MR images to extract the tumor and that tumor can be projected into the
three dimensional plane to analyze the depth of the tumor.

Keywords Active contour models, brain tumor, snakes, vector flow,
projection.



I. Introduction

Medical imaging is a technique and process used to create images of the human body for
clinical purposes such as medical procedures seeking to reveal, diagnose or examine disease.
Medical Science incorporates the study of normal anatomy and physiology. In the clinical
context, medical imaging is generally equated to radiology or clinical imaging and the
medical practitioner responsible for interpreting the images is a radiologist. As a field of
scientific investigation, medical imaging constitutes a sub-discipline of biomedical
engineering, medical physics or medicine depending on the context. Research and
development in the area of instrumentation, image acquisition, modeling and quantification
are usually the preserve of biomedical engineering, medical physics and computer science. [1]
Medical imaging constitutes various modalities to create images of t he human
body such as X-RAY, Comput ed Tomography (CT), Magnetic Resonance Imaging
(MRI), and Ultrasound. X-RAY, invented by Winhelm in 1895 is the oldest source of
electromagnetic radiation used for imaging having wavelength in the range of 0.01 to 10
nanometers. It emits high radiation to create images of the human body thereby causing side
effects such as cancer and eye cataract. Then Computed Tomography (CT) came into
existence. CT scan utilizes X-ray to produce two dimensional images of the structures in the
thin section of the body. CT produces a volume of data which can be manipulated through a
process known as windowing, in order to demonstrate various bodily structures based on
their ability to block the X-ray beam. Although most common in medicine, CT is also used
in other fields, such as nondestructive materials testing. Since it has a greater ionizing
radiation dose burden, repeated scans must be limited to avoid health effects.
Active Contour Models or Snakes are effective tools for segmentation. There are two
types of active contour models, parametric and level set active contour models. Parametric
active contour models are able to extract a single object and are often faster than level set
models. Parametric snakes have two weaknesses; the capture range is limited and inability to
capture acute concave shapes. Level set active contour models are able to capture multiple
objects and complex geometries and are slower than parametric methods. In the presence
of noise level set snakes capture multiple false objects. In the existing work parametric active
contour models such as Gradient Vector Flow (GVF), Boundary Vector Flow (BVF) and
Fluid Vector Flow (FVF) and level set active contour models such as Magnetostatic Active
Contour model (MAC) were used for extracting the boundary of the brain tumor but these
models differ by their capture range and their ability to capture acute concavities. FVF
demonstrated improvements over other techniques with its large capture range and better
convergence for concavities. It is proposed to utilize Fluid Vector Flow for extracting the
tumor in brain MR image and projection of extracted tumor into the three dimensional space
i.e. 3D visualization to analyze the depth of the tumor.

The rest of this paper is organized as follows. In Section II, related work is described.
Section III introduces segmentation and projection of tumor. Discussions are presented in
Section IV, the study of conclusion in Section V.
II. RELATED WORK
Soft Computing techniques such as fuzzy logic and neural networks have been used for
segmentation of brain tumor. Segmentation of brain tumor using HSOM (Hierarchical Self
Organizing Map), an unsupervised clustering technique maps high dimensional inputs to one
or two dimensional discrete lattice of neuron units [2]. HSOM method aids physicians in
tumor diagnosis and monitoring. Another segmentation technique which is an extension to
traditional Fuzzy C-Means (FCM) clustering algorithm which considers two influential
factors: feature difference between neighboring pixels and the relative locations of
neighboring pixels in the image [4].
FCM, by considering those factors greatly restrained the influence of noise. Fuzzy
algorithms for Learning Vector Quantization (FALVQ), an unsupervised learning process
discriminated normal tissues and abnormalities of the brain by partitioning the set of feature
vectors obtained from an MR image into a relatively small number of clusters [6]. Yan Zhu
and Hong Yan proposed an optimization framework for detection of tumor boundaries
using a Hopfield neural network [8]. The tumor boundary was detected in consecutive
slices in such a way to extensively making use of the aspect of connectivity to speed up the
process. Since a tumor doesnt have accurate shapes, detection of tumor boundary is a
difficult task. Active Contour Models play a vital role in accurately capturing the boundary of
the tumor. One such is the traditional snakes or active contours. They are nothing but curves
defined within an image domain that can move under the influence of internal forces coming
from within the curve itself and external forces
computed from the image data [5], [7]. There
are two key difficulties with snakes. First,
the initial contour must, in general, be close to
the true boundary or else it will likely
converge to the wrong result.
The second problem is that active contours
have difficulties progressing into boundary concavities. Then a new class of external forces
for active contour models that addresses both problems called Gradient Vector Flow (GVF)
[9] emerged. It diffuses the edge information from the object contour to its neighborhood.
Particular advantages of the GVF snake over a traditional snake are its insensitivity to
initialization and its ability to move into boundary concavities. Its initializations can be
inside, outside, or across the objects boundary and it does not need prior knowledge about
whether to shrink or expand toward the boundary. Then a new external force which is
generated by a novel interpolation scheme which reduced the computational requirement
significantly called Boundary Vector Flow (BVF) [10] came into existence. It has improved
the capture range and concave object extraction capability than Gradient Vector Flow.
BVF provided large downward force thus reducing the evolution steps required to converge
to the object. Magnetostatic Active Contour model (MAC) utilized bidirectional force field
with the ability of dynamically updating itself while the contour evolves [3]. MAC being a
level set active contour model may capture multiple false objects in the presence of noise. S.
Lobregt described a new technique called discrete dynamic contour model [7] for defining
contours in 2D images.
It consists of a set of connected vertices as the initial model and undergoes
deformations to extract the boundary of the object. It incorporated elegant solutions to
shrinking and clustering problems. Tao Wang developed a new concept called Fluid
Vector Flow (FVF) [1] for addressing the problems of limited capture range and poor
convergence for concavities. FVF is insensitive to initialization by taking advantage of the
boundary map generated from the image. The capture range of FVF extends to the whole
image because the vector flow energy spreads around the entire image.
Figure-1 Proposed system flow diagram
III .Proposed Method
The Objective is to segment the brain tumor using Fluid Vector Flow (FVF) and
three dimensional visualization of tumor. The contour evolves to capture the target object i.e.
the brain tumor. The methods used are preprocessing, binary boundary map generation,
vector flow initialization, FVF and Depth Analysis. Fig.1 shows the flow diagram of the
proposed method which starts with smoothing the input image using a Gaussian smoothing
filter and edge finding using a gradient operator. Then binary boundary map is generated,
and vector flow is initiated, and FVF is computed iteratively until the object i.e. the tumor is
extracted. Then the extracted tumor is projected onto the three dimensional plane for
analyzing the depth or volume of the tumor

A. Preprocessing
Gaussian Filter: A Gaussian filter [34] is a filter whose impulse response is a Gaussian
function. Gaussian filters are created to shun overshoot of step function input while reducing
the rise and fall time. Gaussian filter has the minimum possible group delay. In mathematical
terms, a Gaussian filter changes the input signal by convolution with a Gaussian function and
this change is also called as Weierstrass transform. The input image undergoes smoothing
using Gaussian smoothing filter for elimination of noise. Gaussian filter is a linear spatial
filter which is used for reducing the high frequency components of an image as a result it
smooths the edges of the input image. Gaussian Smoothing is performed by convolving the
input image with the Gaussian function i.e.

( , ) * ( , ) G x y I x y
o


2 2
2
2
2
1
( , )
2
x y
G x y e
o
o
to
+

=

Where I(x, y) is the input image, G

(x, y) is Gaussian smoothing filter with standard
deviation , x and y are the spatial coordinates, and * is the convolution operator.
Gradient operator is then applied to the smoothened image to find edges in the image which
have been suppressed by the Gaussian filter i.e.
( ( , )* ( , )) G x y I x y
o
V
Where is the gradient operator which calculates the directional changes in intensity values.

B. Binary Boundary Map Generation
The boundary map is defined as
( , ) ( ( , )* ( , )) MB x y G x y I x y
o
= V
Where G

(x, y) is the Gaussian smoothing filter with standard deviation , * is the

convolution operator, and is the gradient operator. The normalized boundary map is
defined as

( , ) min( ( , ))
( , )
max( ( , ) min( ( , ))
B B
NB
B B
M x y M x y
M x y
M x y M x y

( , ) 1 ( , )
0
BB NM
M x y if M x y T
else
= >


The choice of suitable threshold value varies depending on the intensity distribution and
contrast associated with the set of i mages bei ng anal yzed. For t he br ai n MR
i mages a threshold of 0.1 is used to achieve object continuity and for extracting the low
intensity region. The extracted boundary provides an envelope to ensure that final
convergence is not out of bound.
C. Vector Flow I nitialization

The contour must be initialized to initialize the external force field. The initial contour can
be inside, outside or overlapping the target objects. FVF is insensitive to initialization by
taking advantage of the binary boundary map. Contour C can be represented as
( ) ( ) 0,1,..... 1
i i
c i x y i P ={ , } e| |
Where P is the number of points on the contour. An external energy function is defined as
( , ) ( cos , sin )
e x y
E x y f f _ o u o u = + + ---------------------------------------- (8)
When M
BB
(x, y) = 0 , E
e
(x,y) holds the first value in (8), otherwise holds zero.
Where is a normalization operator, = 1 (controls the inward or outward direction when
the contour is outside or inside),
( ) ( ( , ))
x y
f f I x y _
,
= V and [ , 2 ] o u t e

D. FVF Computation and Contour Extraction
A trace method is applied to the binary boundary map to get a list of control points. The
control points are used to generate the external force fields. FVF has directional and gradient
forces. The directional force attracts the evolving contour toward the control points even for
control points in a concave region. When the contour is close to the object, the gradient force
fits the contour onto the object. One control point is sequentially selected and this point flows
freely along the o b j e c t b o u n d a r y a n d g e n e r a t e s e x t e r n a l f o r c e f i e l d
dynamically. A parameter is used to manage the selection of control point. The method
selects 1 out of control points to achieve better time efficiency. Once the control point
moves to its new location it generates new external force field for further evolution of contour
until convergence is achieved. FVF computation is iteratively performed to extract the tumor.

Figure -2
Comparision of the segmenting image



E. Volumetric Projection and Depth Analysis
Volumetric projection of brain tumor is needed for diagnostic application such as
analyzing the depth of the tumor. The extracted tumor enclosed by the contour can be
projected on to the three dimensional plane (i.e. x, y, z plane). Mesh is used for three
dimensional visualization or projection of t u mo r . Three Di me n s i o n a l mode l i n g
a l s o k n o wn a s meshing is the process of developing a mathematical representation of
any three dimensional surfaces or object. To perform meshing, a grid must be set up in X and
Y directions and then create a two dimensional grid of coordinate[x, y] with mesh-grid. It
generates x and y arrays for three dimensional projection. As a result of this projection
the tumor is visualized in three dimensions i.e. length, breadth and depth. Depending upon the
gray variations of the projected tumor the depth of the tumor is analyzed. This projection
method requires considerable interactive guidance for satisfactory results since it is sensitive
to image noise and pseudo edges which are common problems in medical images. Compared
to traditional methods the model based segmentation i.e. Fluid vector flow aids in accurate
projection and analysis of depth of tumor.
TABLE -1
Comparative results on segmented images (%)
ACTIVE
CONTOUR
Segmentation validation
Correct
detection
False
detection Pf
MODELS
Pc (%) (%)
BVF 65.67 6.48
MAC 78.53 9.34
FVF 87.38 4.79

Discussion
This paper presented an active contour model based approach for automatic
segmentation of tumor in brain MRI images and three dimensional visualization to
analyze the depth of the tumor. Since MRI is the best modality for imaging the brain
which is composed of soft tissues, this method utilizes two dimensional brain MR images.
Compared to traditional segmentation methods active contour models aids in efficient
segmentation of tumor. There are several active contour models such as Gradient vector flow,
Boundary vector flow and Magnetostatic active contour models. All these models have
difficulties regarding capture range and convergence. Since Fluid Vector Flow model has the
ability of capturing the entire object i.e. the tumor and acute convergence into boundary
concavities. Since segmentation aids in diagnosing the area of the tumor, volumetric
projection is performed to visualize the depth of the tumor. Since manual segmentation of
tumors by physicians is a time consuming
task, this segmentation method with projection is time preserving and also provides accurate
diagnosis.

V. CONCLUSIONS
A parametric Fluid Vector Flow (FVF) active contour model is utilized for automatic
segmentation of tumor in brain MR images and the segmented tumor is visualized in three
dimensions for depth analysis. Since a tumor doesnt exhibits any prior shape, delineating the
tumor accurately is a difficult task. FVF is utilized for segmentation because it can deform
in all directions for capturing the tumor. It also addresses the issues of limited capture range
and the inability to extract complex contours with acute concavities.Segmentation aids in
visualization of area of tumor. Volumetric projection is proposed to analyze the third
dimension i.e. the volume or depth o f the tumor. This automatic method for brain tumor
segmentation and three dimensional visualization will help physicians in accurate diagnosis.

REFERENCES

[1] Tao Wang, Irene Cheng and Anup Basu, "Fluid Vector Flow and Applications in Brain Tumor
Segmentation," IEEE Transactions on Biomedical Engineering., vol. 56, no. 3, March. 2009.
[2] T.Logeswari and M.Karnan, "An improved implementation of brain tumor detection based on soft
computing," Journal of Cancer Research and Experimental Ontology on Medical Imaging., vol. 2(1),
March.2010.
[3] Xianghua Xie, and Majid Mirmehdi, "MAC: Magnetostatic Active Contour Model," IEEE Transactions on
Pattern Analysis and Machine Intelligence., vol. 30, no. 4, April. 2008.
4] Shan Shen, William Sandham, Malcolm Granat, and Annette Sterr, "MRI Fuzzy Segmentation of Brain
Tissue Using Neighbourhood Attraction With Neural Network Optimization," IEEE Transactions on Information
Technology in Biomedicine., vol. 9, no. 3, September.2005.
[5] Chenyang Xu, and Jerry L. Prince, "Snakes, Shapes and Gradient Vector Flow," IEEE Transactions on
Image Processing., vol. 7, no. 3, March. 1998.
[6] Nicolas B.Karayiannis, and Pin-I Pai, "Segmentation of Magnetic Resonance Images Using Fuzzy
Algorithms for Learning Vector Quantization," IEEE Transactions on Medical Imaging., vol. 18, no. 2, February.
1999.
[7] M. Kass, A.Witkin and D.Terzopoulus, "Snakes: Active Contour Models," International Journal of
Computer Vision on Medical Imaging., vol. 1,no. 4, 1988.