A COMPARITIVE STUDY OF IMAGE SEGMENTATION TECHNIQUES

FOR BRAIN TUMOR DETECTION

MANJUNATHA C
M.Tech, Department of ECE, VTU University, MITE Moodbidre, Karnataka

LAXMI G GULAPPAGOL
Assistant Professor Department of ECE, MITE Moodabidre, Karnataka

ABSTRACT: This paper deals with the implementation of simple algorithm for detection of brain
tumor in MRI images based on pillar k-means algorithm. The complete image of brain is produced
by directing CT or MRI scan in to intracranial cavity. The obtained image is examined by
physician for detection and diagnosis. But this approach is not providing the accurate size and
stage of the tumor. To cope up with this inaccuracy of the tumor detection so applying a new
method that is pillar k-means algorithm. This pillar k-means approach provides the accurate and
reproducibility as compare to manual method. The time taken for analysis is less. At the end tumor
can be extract and analyse exact position from the MRI image. By calculating amount of area from
clusters shows the stages of tumor.

KEYWORDS: MRI, Brain tumor, Image segmentation, Detection, Centroids,

INTRODUCTION

Image segmentation refers to the process of dividing an image into groups of pixel which are
homogeneous with respect to some criterion. The segmentation of magnetic resonance images
plays a very important role in medical field because it extracts the required area from the image. In
medical fields image segmentation done by different algorithms and different procedure to
segment and classification of image. Image segmentation is the first step and also one of the most
difficult task of analysing the image, which has independent of removing information and which is
represented in the form of data from an image via image segmentation, feature extent and object
illustration.

Image segmentation process that subdivides an image into its constituent parts and extracts those
parts of interest or objects. Automatic image segmentation also done but the most critical task is
that the segmentation result affect all the subsequent processes of image analysis [10]. Usually
brain anatomy can be viewed by the CT or MRI scan. The use of the scanning can be adversely
affect the normal structure and normal behaviour of the brain. This problem is called as brain
tumor. Brain tumor causes the abnormal growth of the cells in the brain. The MRI scanned image
is taken for the entire process. Images obtained by the MRI are used for analysing and studying the
behaviour of the brain. Image intensity of MRI depends on four parameters. The MRI scan is more
comfortable than CT scan for finding. It is based on the radio waves and magnetic field. Different
types of algorithms are used for detection of brain tumors. But the existing algorithms contains
certain disadvantages in detection and extraction. In this Paper, four algorithms are used for
segmentation. The proposed method gives the exact result for detecting the tumor. The tumor may
be primary or secondary. The initial stage of tumor is called as primary. If the portion of the tumor
is spread to any other place and grown as its own then it is called as secondary. Usually brain
tumor affects CSF (Cerebral Spinal Fluid) and causes strokes. Rather than giving the treatment for
tumor physicians treats for stroke. So detection of tumor is important for that treatment. As early
as detecting the tumor life span of a person is increased like 1 to 2 years. Normally tumor cells are
of two types benign and malignant. Malignant tumors are cancerous and thus mortal but benign are
non-cancerous mild type. In this Paper we focused on detection of tumor and its area using four
different algorithms.

CLASSIFICATION OF SEGMENTATION TECHNIQUES

Classification of segmentation techniques can broadly classified into two types:

1. Local segmentation
2. Global segmentation
Global segmentation is concerned with segmenting a whole image and it deals with segments
consisting of relatively large number of pixels [7]. So it makes estimated parameter values for
global segments most forceful. Image segmentation can be considered as an approach from three
different philosophical perspectives. They are termed as region approach, edge approach and
boundary approach as illustrated in Figure1.

Figure 1. Image segmentation Approach

THRESHOLD BASED SEGMENTATION DETECTION

Threshold is one of the widely used method for image segmentation. It is useful in distinguish
objects of interest from the background. Selection of threshold value is necessary here. Lets take
the threshold value as T which can convert the gray level image to binary image. The binary image
contain characters like position and shape of the objects of interest (foreground). By selecting the
first binary image we can reduce the complexity of the data and get the process of recognition and
classification simplified. The widely accepted way to convert a gray-level image to a binary image
is to select a single threshold value (T). Then all the gray level values below this T will be
classified as black (0), and those above T will be white (1). The segmentation problem becomes
one of selecting the proper value for the threshold T. A regular method used to choose T is by
analysing the histograms of the type of images that want to be segmented. The ideal case is when
the histogram presents only two dominant modes and a clear valley (bimodal). In this case the
value of T is selected as the valley point between the two modes. Threshold technique is one of the
important techniques in image segmentation. This technique can be stated as:

T=T[x, y, k(x, y), f(x, y] . (1)

Where: T is the threshold value. x, y are the coordinates of the threshold value point. k(x, y), f(x,
y) are points the gray level pixels of the image. Then g(x, y) is the threshold image can be defined
as:

599
 (2)
REGION BASED SEGMENTATION METHOD

A region denoted by R of an image is defined as a connected homogenous subset of the image

with respect to some criterion such as gray level or texture. Pixels having similar properties can be
grouped as regions. In the region approach, each pixel is assigned to a particular object or region.
Segmentation algorithms based on region are relatively simple and more immune to noise
Compared to edge detection method [5]. Edge based methods partition an image based on rapid
changes in intensity near edges whereas region based methods, divides an image into regions that
are similar according to a set of predefined criteria [6][8]. In the region-based segmentation, pixels
corresponding to an object are collectively take and marked. Region-based segmentation also
involves the use of appropriate thresholding techniques. The important principles are value
similarity (which includes gray value differences and gray value variance) and spatial proximity
(which consists of Euclidean distance and compactness of a region).

There are many advantages for this algorithm over other color image segmentation algorithms.
Region based segmentation is simple and the border of regions found by are perfectly thin and
connected and it is very stable with respect to noise. It has certain disadvantages include it requires
a seed point, which generally means manual collaboration. Thus, each region to be segmented, a
seed point is needed.

K-MEANS Clustering

K-Means is the one of the unsupervised learning algorithm for clusters. Clustering the image is
collecting the pixels according to the some characteristics. The k-means algorithm initially starts
with defining the number of clusters k. Then k-cluster [11] center are chosen randomly. The
distance between pixels to pixel in each cluster centers are calculated. And the obtained distance
may be of simple Euclidean function. By using distance formula Single pixel is compared to all
cluster centers. The pixel is relocated to particular cluster which has shortest distance among all.
Then the centroid is re-estimated. Yet again each pixel is compared to all centroids. The procedure
continuous until the center converges.
Algorithm
1. Give the no of cluster value as k.
2. Randomly choose the k cluster centers3. Calculate mean or Center of the cluster
4. Calculate the distance between each pixel to each cluster Center.
5. If the distance is near to the center then move to that cluster.
6. Otherwise move to next cluster.
7. Re-estimate the center.
8. Repeat the process until the center doesn't move.

PILLAR ALGORITHM

The capacity of K-means algorithm is to cluster huge data and outlier. An initial starting points
produced randomly so K-means algorithm is difficult to reach global optimum. Barak bah and
Helen performed that the error ratio of K-means is more than 60% for well-separated datasets. To

600
get global optimum result K-means uses Pillar algorithm. This algorithm is very superior and
robust for initial centroids optimization for K-means by positioning all centroids far separately
among them in the data distribution. The algorithm constructed by the use of set of pillars
locations in order to make a stable house or building. Fig. 2 denotes the location of two, three, and
four pillars, in support to withstand the pressure distributions of some different roof structures
composed of separate points. By distributing the pillars as far as possible from each other within
the pressure distribution of a roof, the pillars can resist the roofs pressure and stabilize a house or
building.

Figure 2: The location of set of pillars (white point) with standing against different pressure
distribution of roofs.

The Pillar algorithm is represented as follows. Let X={xi |i=1,,n} be data, k be number of
clusters, C={ci |i=1,,k} be initial centroids, SX _ X be identification for X which are already
selected in the sequence of process, DM={xi |i=1,,n} be accumulated distance metric, D={xi |
i=1,,n} be distance metric for each repetition, and m be the outstanding mean value.

EXPERIMENTAL RESULTS

We have taken the sample image of the brain tumor for estimating the performance of threshold
based segmentation, Region based approach and unsupervised k-means clustering algorithm are
compared with the Pillar K-means algorithm for MRI Brain tumor detection is shown in table 1.
We found the area and time consumption for different segmentation it is observed that result
obtained by pillar k-means is more accurate than the other segmentation methods.

Table 1: Comparison of Area and consumed time of the Tumor

Types of Segmentation Area of the Tumor Time Consumed

Threshold based segmentation 6.95 mm2 1.75783 Sec

Region growing Approach. 22.08 mm2 2.244560 Sec

k-means clustering (cluster=3) 6.36 mm2 0.990927 Sec

Pillar k-means 6.34 mm2 0.885756 Sec

601
Figure 3: original and converted gray scale image. Figure 4: Threshold based segmentation.

Figure 5: Region based segmentation

602
Figure 6: Segmentation using k-means Figure 7: Pillar k-means. (Cluster=3).
clustering (Cluster=3).

Threshold Threshold
25 based 2.5 based
segmentation Time consumed (Seconds) segmentation
20 2
Region based Region based
Area of the tumor

15 Segmentation 1.5 Segmentation

10 K-means 1 K-means based

based Segmentation
5 Segmentation 0.5

0 Pillar K-means 0 Pillar K-means

Segmentation techniques
Segmentation techniques

Figure 8: Graphical representation for table 1 Figure 9: Graphical representation for table 1

CONCLUSION AND FUTURE WORK

In this Analysis of image segmentation, the overview of various segmentation methodologies

applied for digital image processing is explained briefly. By using the different algorithms the
current methodologies of image segmentation is reviewed so that user interaction is possible for
images. This paper describes the review of image segmentation, by using different techniques are
explained. Type of the tumor will be categorized in future. By Using all these techniques observed
that pillar k-means is the best result for the detection of brain tumor.

REFERENCES

W. X. Kang, Q. Q. Yang, R. R. Liang, (2009) The Comparative Research on Image

Segmentation
Algorithms, IEEE Conference on ETCS, pp. 703-707.
H. Zhang, J. E. Fritts, S. A. Goldman, Image Segmentation Evaluation: A Survey of
unsupervised
Methods, computer vision and image understanding, pp. 260-280, 2008.

603
J.selvakumar, A.Lakshmi, T.Arivoli (2012) Brain Tumor Segmentation and Its Area Calculation
in Brain MR Images using K-Mean Clustering and Fuzzy C-Mean Algorithm IEEE-
International Conference on Advances in Engineering, Science and Management (ICAESM -
2012)
K. Atsushi, N. Masayuki, (1998) K-Means Algorithm Using Texture Directionality for Natural
Image Segmentation, IEICE technical report. Image engineering, 97 (467), pp.17-22, 1998.
S. Ray, R.H. Turi, (1999) Determination of number of clusters in K-means clustering and
application in the image segmentation, Proc. 4th ICAPRDT, pp. 137-143,
P. Dhanalakshmi, T. Kanimozhi, (2013) Automatic segmentation of brain tumor using k means
clustering and its area calculation International Journal of advanced electrical and Electronics
Engineering, volume 2, issue 2, 2013.
Shally Hr, Chittaranjan K, (2013) Tomor volume calculation of Brain from MRI Slices,
International journal of Computer science and Engineering Technology, Vol 4 No. 08.
T. Kanungo, D. M. Mount, N. Netanyahu, C. Piatko, R. Silverman, & A. Y.Wu (2002) , "An
efficient k-means clustering algorithm: Analysis and implementation", Proc. JEEE Con!
Computer Vision and Pattern Recognition, pp.881-892.

604