Classification of Hydrocephalus using TAN

Abstract

Hydrocephalus is considered to be one of the diseases that may cause damage in children brain especially
infants. MRI (Magnetic resonance Imaging) is one source of hydrocephalus detection tools, but using MRI in
children brain diseases classification is considered to be difficult process according to the variance and complexity of
brain diseases. This paper presents a solution of detecting one of the children brain diseases which is hydrocephalus.
The proposed system consists of four stages, namely, MRI Preprocessing stage, Segmentation stage, Feature
extraction, and Classification stage. In the first stage, the main task is to eliminate the medical resonance images
(MRI) noise found in images due to light reflections or operator performance which may cause inaccuracies in the
classification process. The second stage, which is the stage where ROI is extracted (tumor region). In the third stage,
the features related with MRI images using Haar wavelet transform (HWT) will be obtained. The features of magnetic
resonance images (MRI) have been decreased using (HWT) to essential features only. And finally the fourth stages,
where new classifier will be presented and finally the result will compare the proposed classifier with six other
classifiers have been used.
Image classification is an important task in the medical field and computer vision. Image classification refers to the
process of labeling images into one of a number of predefined categories. This survey will use the Tree augmented
Naïve Bayes classification technique to detect and classify one of the children brain diseases, and classify the
hydrocephalus type depending on MRI. And it's expected to achieve a high accuracy in hydrocephalus detection to
help the radiologist in the disease detection process.

Full text

© 2015, IJARCSSE All Rights Reserved Page | 90
Volume 5, Issue 11, November 2015 ISSN: 2277 128X
International Journal of Advanced Research in
Computer Science and Software Engineering
Research Paper
Available online at: www.ijarcsse.com
Classification of Hydrocephalus using TAN
1Eman M. Ali, 2Ahmed F. Seddik, 3Mohamed H. Haggag
1Department of Computer Science, Helwan University, Cairo, Egypt
2Dean of Faculty of Computer Science, Nahda University, Professor at the Biomedical Engineering Department, Helwan
University, Cairo, Egypt
3Vice Dean for Student Affairs at Faculty of Computers and Information, Professor at Computer Science Department
Faculty of Computers and Information, Helwan University, Cairo, Egypt
Abstract— Hydrocephalus is considered to be one of the diseases that may cause damage in children brain especially
infants. MRI (Magnetic resonance Imaging) is one source of hydrocephalus detection tools, but using MRI in
children brain diseases classification is considered to be difficult process according to the variance and complexity of
brain diseases. This paper presents a solution of detecting one of the children brain diseases which is hydrocephalus.
The proposed system consists of four stages, namely, MRI Preprocessing stage, Segmentation stage, Feature
extraction, and Classification stage. In the first stage, the main task is to eliminate the medical resonance images
(MRI) noise found in images due to light reflections or operator performance which may cause inaccuracies in the
classification process. The second stage, which is the stage where ROI is extracted (tumor region). In the third stage,
the features related with MRI images using Haar wavelet transform (HWT) will be obtained. The features of magnetic
resonance images (MRI) have been decreased using (HWT) to essential features only. And finally the fourth stages,
where new classifier will be presented and finally the result will compare the proposed classifier with six other
classifiers have been used.
Image classification is an important task in the medical field and computer vision. Image classification refers to the
process of labeling images into one of a number of predefined categories. This survey will use the Tree augmented
Naïve Bayes classification technique to detect and classify one of the children brain diseases, and classify the
hydrocephalus type depending on MRI. And it's expected to achieve a high accuracy in hydrocephalus detection to
help the radiologist in the disease detection process.
Keywords— Hydrocephalus, MRI, Image Classification, Tree augmented Naïve Bayes, children brain diseases .
I. INTRODUCTION
Early detection and classification of brain diseases are very important in clinical practice. Several kinds of research
have been proposed different techniques for the classification of brain tumors using different sources of information [1],
but there is still a lake in hydrocephalus detection algorithms. In this paper, a process for hydrocephalus classification,
depending on the analysis of Magnetic Resonance (MR) images and Magnetic Resonance Spectroscopy (MRS) data
collected for patients with obstructive, Non-obstructive, and Normal Pressure Hydrocephalus is presented.
The hydrocephalus word stands for "hydro", which means water and "cephalus", meaning head, so hydrocephalus
means water in the brain. Hydrocephalus is considered to be a complex disease, which commonly affecting newborns.
The aim is to achieve a high accuracy in discriminating the three types of hydrocephalus through a combination of
several techniques for image denoising, image segmentation, feature extraction and classification. The proposed
technique has the potential of assisting medical diagnosis.
Fig.1. Hydrocephalus compresses and displaces normal brain tissue, Increasing size, pressure and swelling cause
symptoms like seizures or headaches.
Hydrocephalus can be classified according to two criteria which are pathology and etiology. According to pathology
criteria, hydrocephalus is classified into obstructive (non-communicating) or non-obstructive (communicating). And
normal pressure hydrocephalus (NPH) according to etiology criteria, which mainly affect older children, see Fig. 1[2].

Ali et al., International Journal of Advanced Research in Computer Science and Software Engineering 5(11),
November- 2015, pp. 90-97
© 2015, IJARCSSE All Rights Reserved Page | 91
Fig.2. Hydrocephalus MRI with obstructive hydrocephalus region [2].
Infants can suffer from hydrocephalus at birth and can be caused by Dandy-Walker malformations, porenchphaly,
spina bifida, Chairi I and II malformations, arachnoid cysts, and most commonly aqueductal stenosis.
Both types are potentially disabling and life-threatening. Because space inside the skull is limited, their growth
increases intracranial pressure, and may cause edema, reduced blood flow, and displacement, with consequent
degeneration, of healthy tissue that controls vital functions [3]. Brain tumors are, in fact, the second leading cause of
cancer-related deaths in children and young adults. According to the Central Brain Tumors Registry of the United States
(CBTRUS), there will be 64,530 new cases of primary brain and central nervous system tumors diagnosed by the end of
2014 among children. Overall more than 600,000 people currently live with the disease. [3]
Fig.3. Brain with obstructive hydrocephalus region [2].
Hydrocephalus goes mainly untreated in developing countries because neurosurgical care is simply not available. This
year alone, CURE conservatively estimates that nearly 400,000 newborns (3/1,000 births) will suffer from infant
hydrocephalus around the globe and over 310,000 (79%) of these children will be born in the developing world with
limited or no access to critical life-saving care see Fig. 3 [4].
Many kinds of research depending nowadays on using computer technology in medical diagnoses such as cancer-
related research in brain, breast, and liver. MRI is one of the technologies used to take a photo of the body organs
through the usage of the magnetic field. It has much higher features than other radiation tools such as x-ray and
computed tomography (CT) [4]. The researcher had proposed various features for classifying tumor in MRI. The
statistical, Intensity, Symmetry, Texture features etc., which utilize a gray value of tumors are used here for classifying
the tumor. However, the gray values of MRI tend to change due to over –enhancement or in the presence of noise [5].
Though, the main objective of all proposed classification algorithms, whether it relies on characterizing texture to its
statistics or modeling the medical image, aim to learn from how interestingly discern medical images to eventually
develop “knowledgeable ” computer systems. Thus, the objective of this paper presents an appraisal of the existing and
conventional methods for the classification of medical images and based on these observations; propose a framework for
medical image classification. The rest of the paper is structured as Section 2 to Section 5.
Section 2 presents a survey of previous brain hydrocephalus identification and classification techniques. Section 3
illustrates the proposed framework followed by a comparative analysis of the presented classification techniques in
Section 4. And finally the conclusion in Section 5.
II. SURVEY ON PREVIOUS TECHNIQUES
Most of the researchers presented in medical field focus only on tumors, in brain or breast. But there are very few
studies used to detect hydrocephalus disease. Here in this paper image processing techniques to detect and classify the
children brain hydrocephalus are used.
2.1 Javeed Hydrocephalus detection method [4]
In this method, they make use of the Fuzzy c-means algorithm for the detection of hydrocephalus in infants based on
MRI images.

Ali et al., International Journal of Advanced Research in Computer Science and Software Engineering 5(11),
November- 2015, pp. 90-97
© 2015, IJARCSSE All Rights Reserved Page | 92
2.2 Ivkovic Hydrocephalus Detection Method [5]
In this method Ivkovic depending on using image histogram analysis to detect the hydrocephalus region. Ivkovic
present a technique consist of only two phases, image acquisition and preprocessing phase, and histogram analysis phase.
In this technique, they only detect they detect the hydrocephalus region but they didn't classify the extracted region to
which kind of hydrocephalus.
2.3 Sweetman Hydrocephalus Detection Method [6]
Sweetman et al. [6] considered the problem of computational prediction of cerebrospinal fluid flow in the human brain.
By the use of image reconstruction software Mimics 12.11, the authors developed the 3D model from MRI brain images,
which reproduces pulsatile CSF motion and predicts intracranial pressures and flowrates. The model of a healthy brain
helps to predict CSF flow correctly, however, accurate prediction of pathological brain dynamics (such as hydrocephalus)
require model refinement.
III. PROPOSED TECHNIQUE
The proposed system has mainly four modules namely Pre-processing, segmentation using Contribution-Based
Clustering Algorithm, Feature extraction, and hydrocephalus classification. According to the need of the next level, the
pre-processing step converts the image. It performs filtering of noise and other artifacts in the image and sharpening the
edges in the image. RGB to gray conversion and reshaping also takes place here. It includes a median filter for noise
removal. The feature extraction is extracting the cluster, which shows the predicted hydrocephalus at the Haar wavelet
transform output. The extracted cluster is given to the threshold process. It applies a binary mask over the entire image.
In hydrocephalus detection and classification step, the hydrocephalus area is calculated using the binarization method
making the dark pixel darker and white brighter. In threshold coding, each transform coefficient is compared with a
threshold and if it's less than the threshold value, it is considered as zero or else one. In the approximate reasoning step,
the hydrocephalus area is calculated using the binarization method. That is the image having only two values either black
or white (0 or 1). Here 200x200 JPEG image is a maximum image size. The binary image can be represented as a
summation of a total number of white and black pixels. Pre-processing is done by filtering.
Segmentation is carried out by Content-based Image Retrieval (CBIR) algorithm. The feature extraction is done by
considering the threshold and finally, approximating the classification method to recognize the hydrocephalus shape and
position in MRI image using edge detection method [15] see Fig. 4.
Fig.4. Block diagram of the proposed brain diseases classification system.
3.1 Image Preprocessing
In the preprocessing stage, first transform the brain MRI image from RGB mode to Grayscale level and from eight bit
to double precision pattern to get a high-resolution image of the brain while at the same time being noninvasive. The aim
of this paper is to detect, segment and classify the hydrocephalus cells, but for the complete stage it needs the process of
noise removal.
To get the best MR image quality, a median filter was used to remove any noise from the original image see fig. 5.
(A) (B)
Fig.5. (A): The Original MRI image with noise. (B): The same image after applying the median filter.
After applying the median filter to the noisy image now, the brain MR images are ready to go to the segmentation
phase and isolate the hydrocephalus region fig. 5 state the block diagram of the proposed system.
3.2 Segmentation
In this paper, use one of the unsupervised classifications clustering forms, which are used to group the image pixels
based on the similarity between this pixels. This partitioned clustering algorithm is based on the notion of „contribution

Ali et al., International Journal of Advanced Research in Computer Science and Software Engineering 5(11),
November- 2015, pp. 90-97
© 2015, IJARCSSE All Rights Reserved Page | 93
of a data point‟. The content-based image retrieval algorithm will be applied and compare its performance with that of
the k-means clustering algorithm. In comparison with the k-means algorithm, CBIR achieves a better result in measuring
the similarity of both intra-cluster and inter-cluster [6].
The algorithm aims at partitioning a group of data points into disjoint clusters optimizing a specific criterion [2]. When
the number of data points is large, a brute-force enumeration of all possible combinations would be computationally
expensive. Instead, there are many algorithms are applied to find the optimal partitioning. The most popular criterion
function used for partition clustering is the sum of squared error function given by:

 

k
i x i
i
mxE 1
2
)(
(1)
Where k is the number of clusters, Ci is the ith cluster, x is a data point and mi is the centroid of the ith cluster. A
widely used squared-error based algorithm is the k-means clustering algorithm [2]. Here, a clustering algorithm similar to
the k-means algorithm will be used. The contribution of a data point belonging to a cluster is defined as the impact that it
has on the quality of the cluster. This metric is then used to obtain an optimal number of clusters from the given set of
data points.
While this work uses the concept of contribution to finding the optimal number of clusters, CBIR is used for optimal
partitioning of the data points into a fixed number of clusters.
The proposed outline presents contribution-based clustering algorithm. It optimizes on two measures, namely the
intra-cluster dispersion given by:



i
Cx i
mx
n2
)(
1

(2)
And the inter-cluster dispersion given by:
2
1)(
1 k
iimm
k

(3)
Where k is the number of clusters and
m
is the mean of all centroids. The algorithm tries to minimize

and
maximize

.
After founding the tumor pattern region, a resize of the tumor region will take place to be 200 × 200 to have a suitable
region to have enough features, to get a high performance in the next phase.
3.3 Feature Extractions
Once the local tumor pattern is found, which is a 200 × 200 image, now the local tumor features are found but, this
tumor feature vector is too large to use it in the classification phase so, to generate a 50 ×50 feature vector for this pattern
the Haar wavelet transform is used. Haar wavelet can be used to decompose the data in the tumor region into sub-
components that appear in different resolution. It divides the tumor image into four sub-images [7]. These resulted
images consist of two high-resolution images one image that has been high pass in horizontal and vertical directions and
one that has been low pass filtered in both directions.
In comparison with other methods by using the Haar wavelet transform, it successfully reduces the feature vector
which affects the overall performance of the system and decrease the classification process overall time.
3.4 Classification
Image classification refers to the labeling of images into one of a number of predefined categories [8-21].
Classification system consists of a database that contains predefined patterns that compare with a detected object to
classify into the proper category. Image classification is an important and challenging task in various application domains.
According to the classification process, there are three main types of hydrocephalus that the Tree augmented Naïve
Bayes is used to classify the extracted image region to one of them, which are:
1. Obstructive (Non-communicating) hydrocephalus.
2. Non-obstructive (communicating).
3. Normal Pressure Hydrocephalus.
In many kinds of research, there are a lot of classification techniques applied and most of them based on detecting a
sequence similarity between features extracted from the image and the pattern stored in the database.
Most of these classifiers use binary classifiers for classification of the image as a normal image or abnormal image
only without detecting which kind of hydrocephalus the child suffering from such as Neural Networks, Support Vector
Machines (SVMs), etc.
This paper presents a framework using the Tree-Augmented Bayesian Networks (TAN) which performs multi-
classification based on the theory of learning Bayesian Networks. In order to enhance TAN's performance, pre-
processing of data is done by feature discretization and post-processing is done by using Mean Probability Voting (MPV)
scheme. The advantage of using Bayesian approach over other learning methods is that the network structure is intuitive.
The main process of the TAN is to assign the extracted region to one of the classes of the predefined images in the
dataset.

Ali et al., International Journal of Advanced Research in Computer Science and Software Engineering 5(11),
November- 2015, pp. 90-97
© 2015, IJARCSSE All Rights Reserved Page | 94
Fig.5. Classification process using TAN.
The system focuses on calculating the probability that the extracted image q belong to one of the three classes of
hydrocephalus as shown in Fig. 5.
IV. EXPREMENTAL RESULTS
In this section, the results obtained using a database of images is presented. Start by presenting the database with
which conducted in tests, and then, present the results according to the used structure.
4.1 Database
The famous medical imaging is MRI. A magnetic resonance imaging (MRI) scanner uses powerful magnets to
polarize and excite hydrogen nuclei (single proton) in human tissue, which produces a signal that can be detected and it is
encoded spatially, resulting in images of the body.
The MRI dataset consists of 33 Children MRI in jpg format images, with 15 images containing obstructive
hydrocephalus, 13 images containing non- obstructive hydrocephalus and 5 normal pressure hydrocephalus images.
These images are divided into three categories, 70% for training, 15% for testing and 15% for validation.
Fig.6. Samples of children brain dataset containing hydrocephalus.
4.2 Comparative analysis
Next figures show the images as an output gray scale image and extracted hydrocephalus from MRI image. For this
purpose, real-time patient data is taken for analysis. As hydrocephalus in MRI image have an intensity more than that of
its background so it become very easy locate it and extract it from MRI image.
(A) (B)
Fig.7. (A): MRI image of hydrocephalus affected brain grayscale image. (B): the extracted hydrocephalus region.
After using two kinds of segmentation techniques on these images, contribution based information retrieval achieving
better performance compared with K-Means algorithm see Fig. 8 and Fig. 9.

Ali et al., International Journal of Advanced Research in Computer Science and Software Engineering 5(11),
November- 2015, pp. 90-97
© 2015, IJARCSSE All Rights Reserved Page | 95
Fig.8. Average value of

and

against the number of clusters.
Fig.9. Average Precision and Recall against the number of clusters.
The next table shows the classification accuracy using CBIR and k-Means segmentation techniques.
TABLE 1 THE ACCURACY OF THE CLASSIFICATION TECHNIQUES USING K-MEANS AND CBIR SEGMENTATION ALGORITHMS
Segmentation
DA
NN
NB
SVM
DT
KNN
TAN
Brain MRI using
(K-Means)
75.93
91.44
76.08
92.59
87.04
82.3
99.4
Brain MRI using
(CBIR)
90.12
96.10
93.52
92.59
96.19
93.7
99.8
Fig.10. The accuracy of classification techniques using CBIR and K-mean Segmentation techniques.
Fig.11. Classification results in terms of specificity and sensitivity.

Ali et al., International Journal of Advanced Research in Computer Science and Software Engineering 5(11),
November- 2015, pp. 90-97
© 2015, IJARCSSE All Rights Reserved Page | 96
The next table states a comparison between the classification techniques according to the classification time see table 2.
TABLE 2 THE CLASSIFICATION TECHNIQUES CLASSIFICATION TIME.
Classification
Technique
Classification
Time (seconds)
DA
760.8
NN
754.3
NB
11.6
SVM
735.7
DT
10.06
KNN
0.03
TAN
1.6
Fig.12. A comparison between classification techniques according to classification time.
From the results of experiments, the tree augmented naïve Bayes algorithm gave the best detection rate. It, achieving a
classification rate of 99.8 %. But when considering computational performance, however, K-Nearest Neighbor algorithm
proved to have a faster build time (Time it takes to build a model on network training data) at 0.03 seconds while having
a detection rate of 93.7% as shown in table 1. Tree augmented Naïve Bayes had the second best build time at 1.6 seconds
and a detection rate of 99.8%. Computational performance is particularly important when considering the real-time
classification of potentially thousands of simultaneous networks traffic. From experiments, Tree augmented Naïve Bayes
appears to be the best suited for real-time classification tasks due to its relatively fast classification speed and high
detection rate.
V. CONCLUSION
This paper presents a survey on various image mining techniques that was proposed earlier by researchers for the
better development in the field of content-based image retrieval. The purpose of the mining is to produce all considerable
patterns without prior knowledge of the patterns. Important information can be hidden in images, conversely, few
research talks about data mining on them. Image segmentation is the primary phase in image mining. In other words,
image mining is simply an expansion of data mining in the field of image processing. Image mining handles with the
hidden knowledge extraction, image data association and additional patterns which are not clearly accumulated in the
images. Also, this paper provides a marginal overview for future research and improvements. Certain possible future
investigations that are discussed may be done in the area of image mining which included the experimentation's on other
image elements such as textures, shape, etc.
In future, this program can be done more advanced so that hydrocephalus growth can be analyzed by plotting the
graph which can be obtained by studying sequential images of hydrocephalus affected patient.
The future research work may include the implementation of the Bayesian networks for relevance feedback and more
extensive tests with other examples of image forensic work. It is also envisaged that subjective testing will be performed
with input from forensic experts.
Some possible future studies that may be conducted in the area of image mining include the experimentation's on other
image elements such as textures, shape, and so forth. It will also be interesting to investigate hidden relationships among
images. For example, intensive and extensive exploratory pattern analysis involved in the existing systems in the
database can be very useful.
ACKNOWLEDGMENT
I am very grateful and would like to thank my guides Prof. Mohammed Haggag and Prof. Ahmed Farag for their
advice and continued support. Without them, it would not have been possible for me to complete this paper. I would like
to thank my husband for the thoughtful and mind stimulating discussion we had, which prompted us to think beyond the
obvious.

Ali et al., International Journal of Advanced Research in Computer Science and Software Engineering 5(11),
November- 2015, pp. 90-97
© 2015, IJARCSSE All Rights Reserved Page | 97
REFERENCES
[1] K. Kharat, P.Kulkarni, M.B.Nagori," Brain Tumor Classification Using Neural Network Based Methods ",
International Journal of Computer Science and Informatics ISSN (PRINT): 2231 –5292, Vol-1, Iss-4, 2012.
[2] Brain MRI dataset, www.CEwebsource.com, 2014.
[3] V. Gladis, P. Rathi, S. Palani, " BRAIN TUMOR MRI IMAGE CLASSIFICATION WITH FEATURE
SELECTION AND EXTRACTION USING LINEAR DISCRIMINANT ANALYSIS ", Sudharsan Engineering
College Sathiyamangalam, Pudukkottai , India,2014.
[4] J. Hussain, P. V. Sree, "Detection of Hydrocephalus Lateral Ventricles Quantitatively in Brain MRI images of
Infants", International Journal of Computer Applications, Vol. 83, No. 9, PP. 12, December 2013.
[5] M. Ivkovic, M. Zimmermann, "MRI assessment of the actazolamide and External lumber drainage in idiopathic
Normal Pressure", University of Miami School od Medicine , USA,2015.
[6] Sweetman B., Xenos M., Zitella L., Linninger A.A.: Three-dimensional computational prediction of
cerebrospinal fluid flow in the human brain, Computers in Biology and Medicine, Vol. 41, pp. 67-75, 2011.
[7] A. Gale, S. Salankar," A Review On Advance Methods Of Feature Extraction In Iris Recognition System " ,
International Conference on Advances in Engineering & Technology – 2014 (ICAET-2014), PP 65-70,2014.
[8] M. Nagori, S. Mutkule , "Detection of Brain Tumor by Mining fMRI Images", International Journal of
Advanced Research in Computer and Communication Engineering , Vol. 2, Issue 4, January 2013.
[9] A. Jose, S.Ravi, M.Sambath , "Brain Tumor Segmentation Using K-Means Clustering And Fuzzy C-Means
Algorithms And Its Area Calculation" , International Journal of Innovative Research in Computer and
Communication Engineering, Vol. 2, Issue 3, March 2014.
[10] S. Anwar , R. Ibrahim, "A Framework for Medical Images Classification Using Soft Set", The 4th International
Conference on Electrical Engineering and Informatics (ICEEI 2013), Malaysia.,2013.
[11] P. Kamavisdar, S. Saluja, S. Agrawal, "A Survey on Image Classification Approaches and Techniques",
International Journal of Advanced Research in Computer and Communication Engineering, Vol. 2, Issue 1,
January 2013.
[12] H. Aslam, T. Ramashri, M. Ahsan, "A New Approach to Image Segmentation for Brain Tumor detection using
Pillar K-means Algorithm" , International Journal of Advanced Research in Computer and Communication
Engineering, Vol. 2, Issue 3, March 2013.
[13] K.Kothavari, R.Keerthana, M.Mariselvam,S.Kaveya, L.Mekala, "A Hybrid approach for PNN-Based MRI Brain
Tumor Classification and Patient Detail Authentication Using Separable Reversible Hiding" , International
Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering, Vol. 2, Issue 4, April
2013.
[14] V. Rekha, S. Anantharajan, "C.A.D.S., for Classification of MRI Brain Tumor Using Decision Tree",
International Journal of Innovative Research in Science, Engineering and Technology, Vol. 3, Issue 3, March
2014.
[15] M. Suganya, M. Menaka, "Various Segmentation Techniques in Image Processing: A Survey", International
Journal of Innovative Research in Computer and Communication Engineering, Vol.2, Special Issue 1, March
2014.
[16] C. P. Loizou, M. Pantziaris, C. S. Pattichis, I. Seimenis, "Brain MR image normalization in texture analysis of
multiple sclerosis", Journal of Biomedical Graphics and Computing, Vol. 3, No. 1, 2013.
[17] P. Loizoua, C. Kyriacou, "Brain white matter lesion classification in multiple sclerosis subjects for the prognosis
of future disability", Intelligent Decision Technologies, Pp. 3–10, 2013.
[18] Dr S. McKay, Dr R. Hadfield, "Current Knowledge in Brain Cancer Research", Developed for Cure Brain
Cancer Foundation, Pp. 1-33, May 2014.
[19] A. Morgado, L. Caldeira, N. Silva, "Dynamic analysis of MR-PET data on brain tumors", 3rd Conference in
PET/MR and SPECT/MR , Kos Island, Greece. 19-21 May 2014.
[20] B. Wei, Z. Yu, G. Yang, "A New Multistage Medical Segmentation Method Based on Super pixel and Fuzzy
Clustering" , Hindawi Publishing Corporation Computational and Mathematical Methods in Medicine, Volume
2014, Article ID 747549, 13 pages, 2014 .
[21] Ms. P. Jayashri, Mrs. D.Gandhimathi, "COMPARATIVE STUDY ON DIFFERENT BRAIN TUMOR
SEGMENTATION IN MRI IMAGES", International Journal of Advanced Technology in Engineering and
Science, Volume No.02, Issue No. 09, September 2014.