Brain Tumor
Segmentation using Contextual Clustering and Object Labeling in MRI
G.
Merlin Sheeba*, M. S. Godwin Premi, G. Mary Valantina, Z. Mary
Livinsa
School of Electrical and Electronics, Sathyabama
Institute of Science and Technology, Chennai
*Corresponding Author E-mail: merlinsheebu@gmail.com
ABSTRACT:
Brain tumor segmentation in MRI images is one of the challenging
tasks in medical field. In this work, an efficient brain tumor is detection
methodology is proposed for the MRI images. Image preprocessing filters are
used to remove noises from the images. Contextual clustering extracts the tumor
regions from the images Object labeling is also performed following the
contextual clustering. The simulation results show that the proposed technique
has achieved 94% of efficiency.
KEYWORDS: Contextual clustering, object labeling,
Segmentation.
INTRODUCTION:
The major challenge in analyzing MRI brain images is to classify
the pixels of images into homogeneous regions. This type of problems is termed
as clustering or segmentation problem. Segmentation is the process of dividing
the image into multiple segments. The goal of the segmentation is to simplify
or change the representation of the image into something easier to analysis.
There are many segmentation methods are followed, such as, compression
based method, histogram based method, region growing method, edge
detection method, split and merge method, etc. This project depends about the
clustering.
Clustering deals with finding a structure in a collection of unlabelled data. A clustering is a process of the objects
is made into groups whose members are similar in some way. A collection of object which is similar and as similar in between them is
called as cluster. The segmentation of the human brain image from the MRI
(magnetic resonance imaging) is made into three different tissues. They are
cerebrospinal fluid, gray matter and white matter. Fuzzy c-means clustering is
widely (FCM) used for image segmentation. But the FCM clustering are not
accurate in dealing with local spatial property of images which leads to strong
noise sensibility. In most of the medical images are not proper and also some
unknown noise will leads to degrade in segmentation
quality. For this purpose we are making use of this
contextual clustering to get a clear view of the brain tumor. Our method is to
detect the object (MRI images) to give a clear quality for the operation
purposes.
RELATED WORKS:
M. Mohammed and D.B. Mohammed [1] presented to produce the
segmentation of the brain MRI images for the extraction by using the K-means
clustering and Perona-Malik anisotropic diffusion
model. Segmentation by manually of these abnormal tissues cannot be compared
now which enable us to observe the volume and location of unwanted tissues. The
segmentation of the brain will be divided according to their tissue type which
includes White Matter (WM), Gray Matter (GM) and Cerebrospinal Fluid (CSF).
These tissues are not properly shown .In this method
T1,T2 and PD weighted gray level intensity images are used. J. Selvakumar et .al [2] have performed the Brain tumor
segmentation by using k-means clustering and fuzzy c-mean algorithm (FCM). They
have implemented FCM Algorithm for detection of the range and shape of tumor in
brain MRI images. Uncontrolled growth of tissues in any part of the body is
called as tumor. As it is known, brain tumor is inherently serious and
life-threatening because of its character in the short space formed inside the
skull. The scans or MRI that is directed into the cavity produces a complete
image of brain. This image of tumor is examined by the physician and made to be
detected. This method of detects the stage and size of tumor. To avoid that,
this project uses the segmentation of brain tumor based on the combination of
both algorithms. In addition, it also reduces the time in analyzing. The tumor
is taken from the MRI image and its actual position and shape is determined.
The Brain MRI segmentation for tumor detection is done by cohesion based self merging algorithm (CSM) was explained by S. Koley et .al [3] deals with the segmentation of brain MRI
for determining the exact location of brain tumor by CSM with partitioned K-
means clustering algorithm. The CSM is made to be compared and the noise is
also made to be reduced. This method is much simpler and computationally less
complex and computation time is also less. M. Shasidhar
et. al [4] have recommended the MRI brain image segmentation using modified
fuzzy c-means clustering (FCM) algorithm. In this process the Clustering
approach is widely used in biomedical applications especially for brain tumor
detection in MRI images. Fuzzy clustering using fuzzy C-means (FCM) algorithm
is superior over the other clustering approaches in terms of segmentation
efficiency. The effectiveness of the FCM algorithm rate is improved by
modifying the cluster center and membership value updating criterion. In this
method, the FCM algorithm for MR brain tumor detection is explored. Comparative
analysis in terms of efficient segmentation and the convergence rate is
performed between the conventional FCM and the modified FCM. This process show superior results for the modified FCM algorithm in
terms of the performance measures. For the brain tumor pattern image processing
technique is used which was presented by Kimmi Verma
et .al [5]. The research was done which made by the use of software with
edge detection and segmentation methods, which gave the edge pattern and
segment of brain and brain tumor itself. Medical image will play a vital point
in research, as it is complex problems for the proper diagnosis of brain
disorders. In this research, it is a foundation of segmentation and edge
detection, as the first step towards brain tumor grading. The segmentation is
reviewed with an emphasis placed on revealing the advantages and drawbacks of
these methods for medical imaging applications.
MATERIAL AND METHODS:
System Description:
The block diagram of the proposed method is shown in figure 1.Intially the MRIimage is
preprocessed wherein the filtering techniques are used.
The MRI brain tumor image is given as the input of
the proposed system. As shown in the figure 1, the image preprocessed is
the first step in this process. The conversion of the image from RGB to
grayscale image is done.While
in the conversion process there may be some kind of noise might be added in the
image .Even though the image is in good quality the noise should be added
in the process. The image is added with 0.05% of salt and pepper noise.
Now the image will be a noisy image and this noise will be removed by the
median filter .These are the preprocessed
methods in the given image. The salt and pepper noise are removed by the Median
filter.
Morphological Operations on Binary Images:
The Morphological operations will affect the form, structure or
shape of an object. After the median filter the process of morphological
opening is done as shown in the fig.1.Here in this process the Erosion and
Dilation is done. The morphological opening is first done with Erosion and then
only with Dilation. But in the morphological closing method the process of
Dilation is done first then only with Erosion. The range of the grayscale image
is 0-255.
Erosion:
The erosion is process of eroding the image ,it
turns the black to white pixel. The structuring element is made to be done with
erosion process. The pixels will be altered by this process. The outer part of
the mri images are eroded in this process. The
erosion will make the image more clear for the doctor
purpose. As shown in the fig.2. the tumor part will be clearly distinguished in
this method. After this process is done then the image is made to be diluted.
Dilation:
The dilation process is performed by laying the structuring
element on the image. After the erosion phase is completed the
image is diluted. This will be done in the inner part of the tumor, hence a clear view of the tumor is possible. The
dilution is just the opposite of the erosion process. This
two process will make the image to be complete in the morphological
opening process as shown in figure 2.
Contextual Clustering:
The recognition of the active region is the process of contextual
clustering. According to the statistical property the activation class is
unknown and the background class is known. A testing of hypothesis is used to
deal with this kind of classification problem. In the present case, the null
hypothesis is that the pixel under consideration is zero from the background
class. In a standard one-tailed test of a null hypothesis, the zero hypothesis
would be rejected, and if the statistical value are
smaller than zero.
In order to develop a contextual testing method, a hypothetical
activation class, or rejection class, is defined. The class determines a
difficult region corresponding to an “different event”. Pixels whose
characteristics are “closer” to the rejection class than the background class
violate the null hypothesis and are put into the rejection class and considered
active [6,7]. The standard normal distribution is obtained by the background
class. Suppose that there is no contextual support toward either accepting or
rejecting the null hypothesis. The null hypothesis is rejected in this
conventional hypothesis because the pixel value is smaller than a user
specified threshold. It is assumed, therefore, that the rejection class
follows a distribution, which guarantees that the rejection probability will be
a single increased function. The definition of the rejection class is required
to formulate the testing problem as a classification problem. The detection
problem is now reduced to a two class-segmentation problem [8] that can be
solved using standard methods. The cluster variation is shown in Figure 3.
Object Labeling:
The main motive of this project to detect
the tumor from the brain. So the object labeling
algorithm is used to detect accurate part of brain tumor. Object Labeling is
the process of labeling the given image. Here in this process the MRI image are
labeled. The binary image of the tumor will be labeled accordingly. The
simulation results are compared different tumors in Table 1.
TABLE 1: Experimental results
|
MRI Images |
Types of Tumor |
MSE |
PSNR |
|
|
Maligant |
13.2955 |
36.9277 |
|
|
Benign |
8.0933 |
39.0836 |
|
|
Benign |
27.591 |
33.7571 |
CONCLUSION:
The main contribution of the work is to
explore various techniques to detect brain tumor in an efficient way. This work
has proposed a new object based brain tumor detection
along with the decision based median filtering. The design and implementation
of the object labeling algorithm is done in MATLAB using GUI (graphical user
interface). The experimental results of the proposed technique has achieved up to 94% accuracy. This method can detect
tumor in high noisy images as well.
REFERENCES:
1. Revathy, M. (2012). Image classification with
application to MRI brain using 2nd order moment based
algorithm. International Journal of Engineering Research and Applications
(IJERA), 2(3), 1821-1824.
2. Sucharita, V., Rao, P. V., Bhattacharyya, D., and
Kim, T. H. (2016). Classification of Penaeid Prawn Species using Radial basis
Probabilistic Neural Networks and Support Vector Machines. International
Journal of Bio-Science and Bio-Technology, 8(1), 255-262.
3. Padma, A., and Sukanesh,
D. R. (2011). Automatic diagnosis of abnormal tumor region from brain computed
tomography images using wavelet based statistical texture features. arXiv Preprint arXiv:1109.1067.
4. De Boer, R., Van Der Lijn,
F., Vrooman, H. A., Vernooij,
M. W., Ikram, M. A., Breteler, M. M., and Niessen, W. J. (2007, April). Automatic Segmentation of
Brain Tissue and White matter Lesions In MRI. In 2007 4th IEEE
International Symposium on Biomedical Imaging: From Nano to Macro (pp.
652-655). IEEE.
5. Anbeek, P., Vincken, K.
L., and Viergever, M. A. (2008). Automated MS-lesion
segmentation by k-nearest neighbor classification. MIDAS Journal.
6. Qurat-Ul-Ain, G. L., Kazmi, S. B., Jaffar, M.
A., and Mirza, A. M. (2010). Classification and segmentation of brain tumor
using texture analysis. Recent advances in artificial intelligence,
Knowledge Engineering and Data Bases, 147-155.
7. Mancas, M., Gosselin, B., and Macq, B. (2006,
January). Tumor detection using airways asymmetry. In 2005 IEEE
Engineering in Medicine and Biology 27th Annual Conference (pp.
6528-6531). IEEE.
8. Sheeba, G. M., and Memala,
A. (2018, May). Detection of Gaze Direction for Human–Computer Interaction.
In International Conference on ISMAC in Computational Vision and
Bio-Engineering (pp. 1793-1803). Springer, Cham.
Received on 21.06.2019
Modified on 09.09.2019
Accepted on 29.10.2019
© RJPT All right reserved
Research J. Pharm.
and Tech. 2020; 13(4):1833-1836.
DOI: 10.5958/0974-360X.2020.00330.3