You are on page 1of 7

2014 International Conference on Control, Instrumentation, Communication and Computational Technologies (ICCICCT)

A Survey on Detection of Brain Tumor

from MRI Brain Images


Research Scholar. Associate Professor, Associate Professor, Department of
Department of Computer Science, Department of Electronics and Electronics And Instrumentation
Noorul Islam Univesity, India. Instrumentation , NI University, India . NI University, India.

Abstract - Brain tumor detection and segmentation is one of morphological operation, feature selection and extraction
the most challenging and time consuming task in medical and classification
image processing. MRI (Magnetic Resonance Imaging) is a
medical technique, mainly used by the radiologist for II. MATERIALS AND METHODS
visualization of internal structure of the human body without 1. Xiao et al [1] proposed an approach to estimate features
any surgery. MRI provides plentiful information about the
from the correlation between brain lateral ventricular (LaV)
human soft tissue, which helps in the diagnosis of brain
tumour. Accurate segmentation of MRI image is important for deformation and tumor and the extracted features are
the diagnosis of brain tumor by computer aided clinical tool. applied for tumor segmentation of MR images. Proposed
After appropriate segmentation of brain MR images, tumor is technique mainly consists of four stages: pre-processing,
classified to malignant and benign, which is a difficult task due feature extraction, segmentation and classification. In the
to complexity and variation in tumor tissue characteristics like first stage, the issue of non standardization of intensity,
its shape ,size ,gray level intensities and location. Taking in to geometric non uniformity and redundant data in the
account the aforesaid challenges, this research is focussed background image and skull are addressed. Lateral
towards highlighting the strength and limitations of earlier ventricular deformation is used for feature extraction. In the
proposed classification techniques discussed in the
segmentation part, unsupervised segmentation methods are
contemporary literature. Besides summarizing the literature,
the paper also provides a critical evaluation of the surveyed used to for the evaluation of LaV deformation feature on
literature which reveals new facets of research. the brain tumor segmentation. In this paper the most
frequently used methods are K nearest neighbours (KNN)
Keywords: Brain tumor Detection, Image and conventional Fuzzy connected C-mean (FCM).
Segmentation, Magnetic Resonance Imaging, Pre-processing,
Feature extraction etc.
The experimental results shows the relevancy between LaV
deformation and tumor location. Comparative experiment
I. INTRODUCTION study on tumor segmentation suggest that, tumor
segmentation accuracy improves when the extracted
Cancer in a body occurs when the cell in the body grows features are accurate. In the proposed system the Specificity
and divides in an uncontrollable manner. If this happens in and sensitivity obtained is 100%.The proposed brain tissue
brain then it is called as brain tumor. A brain tumor is a segmentation has a disadvantage of wrongly assigning a
mass of unnecessary and abnormal cell growing in the brain non CSF pixel to the cluster CSF. To remove this undesired
or it can be defined as an intracranial lesion which occupies pixel, a global mask is applied, there by leaving the region
space within the skull and tends to cause a rise in as extracted.
intracranial pressure. Brain tumors are mainly classified in
to two ie Benign and Malignant. Benign tumors are non- Future scope of this paper is that, by incorporating the LaV
cancerous and they seldom grows back where as malignant deformation as an additional features can be obtained for
tumors are cancerous and they rapidly grows and invade to pattern recognition segmentation, thereby improving brain
the surrounding healthy brain tissue. The location of tumor tissue segmentation .
in brain helps the individual to determine how the brain Nandagopal and Rajamony [2], in their paper they
tumor effects an individual normal functioning. Brain presented a combination of wavelet statistical
tumor can be diagnosed by taking personal and family features(WST) and wavelet co-occurrence texture
medical history and also by physical examination, brain feature(WCT) obtained from two level discrete wavelet
CT/MRI scan, brain angiogram, spinal tap biopsy etc. transform is used for the classification of abnormal brain
diagnosis of brain tumor can be delayed because its tissues in to benign and malignant. The proposed system
symptom is similar to symptom of other condition. There consists of four phases: segmentation of region of interest,
are several image processing technique such as histogram discrete wavelet decomposition, feature extraction and
equalisation, image segmentation, image enhancement, feature selection and classification and evaluation. The

978-1-4799-4190-2/14/$31.00 ©2014 IEEE 871

2014 International Conference on Control, Instrumentation, Communication and Computational Technologies (ICCICCT)

support vector machine is employed for brain tumor Low threshold values can improve the classification
segmentation. A combination of WST and WCT is used for accuracy of local features, but it may lead to over
feature extraction of tumor region extracted from two level clustering, which adversely affect the normal tissue
discrete wavelet transform. Genetic algorithm is used to analysis. Feature extraction cos t due to clustering in SC-
select the optimal texture features from the set of extracted ICA is another issue
features. The probabilistic neural network is used to classify
Navarro et al[5] , in their paper presented a new method for
abnormal brain tissue in to benign and malignant and the
feature selection of dimensionality reduction and several off
performance evaluation is done by comparing the
the shelf classifiers on various HMRS modalities ie, long
classification result of PNN with other neural network
and short echo times and an adhoc combination of both. In
classifier. The classification accuracy of the proposed
feature selection they are having entropy selection
system is 97.5%.
algorithm, which is a fast method to generate a relevant
However the main limitation of this proposed system is subset of spectral frequency. Feature selection is done in
that, it need new training for Gaussian SVM classifier the classifier in an independent way in the boot strap
whenever there is change in image data set and this method samples. Then a set of classifier is developed on the boot
is applied only to CT images only. In future the work can strap samples using previously selected set of features and
be extended to other types of imaging such as liver CT the outcome is the selection of a specific classifier for each
imaging ,MRI imaging, ultrasound imaging etc. data type. A final model is obtained using boot strap sample
using iterative procedure. The accuracy obtained for the
Kalbhani et al [3] in their paper, they introduced a method
proposed system is 95%.
for the classification of MR images in to normal and
abnormal one. At first two level, two dimensional discrete Therefore the future research extends the use of the
wavelet transform (2D DWT) of the input image is proposed methodology to other brain tumor classification
calculated and wavelet coefficients of details sub band are problems involving different pathologies and pathological
modelled by Generalized Auto Regressive Conditional grouping.
Hetroscedasticity (GARCH) statistical model. After feature
Saritha et al [6] in their paper proposes approach by
vector normalization, principal component analysis (PCA)
integrating wavelet entropy based spider web plots and
and Linear Discriminant Analysis (LDA) are used to extract
probabilistic neural network for the classification of Brain
the proper feature and to reduce the redundancy from the
MRI. The proposed technique uses two steps for
primary feature vectors. Finally the extracted feature are
classification ie wavelet entropy based spider web plot for
applied to the K nearest neighbour (KNN) and support
feature extraction and probabilistic neural network for
vector machine (SVM) classifier separately to determine
classification. The obtained brain MRI, the feature
the normal or abnormal type images. The results shows that
extraction is done by wavelet transform and its entropy
the proposed algorithm can achieve high classification rate
value is calculated and spider web plot area calculation is
and it needs only less number of features for classification.
done. With the help of entropy value classification using
In the first scenario the classification accuracy achieved for
probabilistic neural network is computed. Probabilistic
KNN and SVM classifier are about 97.62% and 98.21%
neural network provides a general solution for pattern
respectively and in the second scenario both classifier
classification problem and its classification accuracy is
achieve about 100% accuracy
about 100%.
Sindhu mol et al[4], proposed a method to improve the
The limitation of this technique is that whenever there
classification of brain tumor from Magnetic Resonance
is a change in image database, fresh training is required.
Image based on spectral angle based feature extraction and
spectral clustering independent component analysis (SC- Sumitra and Saxena [7]. In this paper they present a neural
ICA). First the MR image is divided in to different clusters network technique for the classification of magnetic
by spectral distance based clustering. Independent resonance brain images. The proposed techniques consist of
component analysis (ICA) is done on the clustered data namely 3 stages: feature extraction, dimensionality
along with support vector machine (SVM). Here reduction and classification. The feature extraction is done
T1weighted, T2 weighted and proton density fluid using PCA from MR images and essential features such as
inversion recovery images were used for evaluation. mean, median, variance, correlation values of maximum
Comparative analysis is done with ICA based SVM and and minimum intensity are extracted. In the classification
other conventional classifiers to established the stability and stage, the classifier based on back propagation, neural
efficiency of SC-ICA based classification. The accuracy network have been developed. This classifier classifies the
achieved by the analysis of ICA based SVM results in 98% image as normal, benign and malignant. The result shows
and 96.1% for reproduced lesion. that the BPN classifier gives fast and accurate classification
Future work of the proposed method can be refined with an than any other neural network classifier. The classification
adaptive threshold selection scheme and the expansion of accuracy of testing data set of brain image is 73%. Its future
multi spectral data with more informative MRI sequence, works includes that the performance can be increased by
can be used for the analysis of several brain tissues. increasing the database.
Jayachandran and Dhanasekharan [8] proposed a paper
The experimental results shows that the classification
based on hybrid algorithm for detection of brain tumor in
performance varies based on the threshold values selected.

978-1-4799-4190-2/14/$31.00 ©2014 IEEE 872

2014 International Conference on Control, Instrumentation, Communication and Computational Technologies (ICCICCT)

magnetic resonance imaging using statistical and support Mustara & Suchalatha [11] in their paper they proposed a
vector machine classifier. The proposed technique consists system for brain cancer detection and classification. The
of 4 stages namely noise reduction, feature extraction, image processing technique like histogram equalization,
feature reduction and classification. In the first stage image segmentation, image enhancement, morphological
anisotropic filter is applied for noise reduction and to make operation and feature extraction have been developed for
the image suitable for extracting features. In the second the detection of brain MRI images. The texture feature
stage texture features are extracted using gray level co- extracted in the detected tumor has been achieved by using
occurrence matrix. In the third stage extracted feature is gray level co-occurrence matrix (GLCM). These features
reduced using principal component analysis (PCA). In the are compared with the features stored in the knowledge
last stage the FSVM classifier is used to classify the subject base. Finally a neuro fuzzy classifiers has been developed
as normal or abnormal image. The classification accuracy is for the recognition of different type of brain cancer. The
about 95.80%. whole system was tested in two phase firstly
learning/training phase and secondly recognition/testing
This paper is having the limitations of using principal
phase by using ANN classifiers .
component analysis, which reduces the lower
dimensionality of the texture feature. Jafari and Shafaghi [12]. In their paper, they presented a
hybrid approach for the detection of brain tumor tissue in
Nantha Gopal and Sukanesh [9] in their paper presented a
magnetic resonance image based on genetic algorithm and
computerized software system designed for segmentation
support vector machine. Proposed system consists of 4
and classification of benign and malignant tumor. In this,
stages. In the first stage –pre-processing: noise removal and
the author proposed a method for selecting both dominant
contrast enhancing is done. The second stage is
run length and co-occurrence texture feature of wavelet
segmentation. Skull stripping is done with the help of
approximation tumor region by a support vector machine
morphological operations. The third stage is feature
(SVM).Two Dimensional Discrete Wavelet Decomposition
selection and extraction. Feature selection is done based on
is performed on the tumor region having noise. Of th ese
4 categories- static features, Fourier and wavelet transforms
Seventeen features extracted , six features are selected
histogram and the combination of prior set. Feature
using students t-test. The selected Features are fed to the
selection is done by means of genetic algorithm. In the
classifiers for classification .Here SVM and probabilistic
fourth stage, the selected features are fed as input to the
neural network classifiers are used. The classification
support vector machine classifier to detect normal and
accuracy is evaluated using k fold cross validation method.
abnormal brain with an accuracy of 83.22%.
In this the classification accuracy obtained is about 96.4%.
The limitation of this work is that wavelet transform require
The experimental results show that by using SVM classifier
large storage and its computational cost is high.
high segmentation and classification accuracy can be
obtained. However the paper has its own limitation of Rathi and Palani[13] in their paper they proposed a novel
having a new training set, whenever there is change in the method of feature selection and extraction. This approach
data set. This method is applicable only to CT images. mainly combines the intensity, texture, shape based features
More over the work can be extended to other type of and classifies the tumor region as white matter(WM), gray
imaging such as liver CT images, MRI imaging and ultra matter(GM), CSF, abnormal and normal area. The support
sound imaging. vector machine (SVM) classifier is used for the comparison
of nonlinear technique with linear ones. Principal
Deepa and Devi [10] in their paper they proposed an
component analysis (PCA) and linear discriminant analysis
automatic method that make use of the capability of back
(LDA) methods are used to reduce the number of features.
propagation (BP) and Radial Basis Function (RBF) neural
The feature selection using the proposed technique is more
network function to classify brain image in to cancerous or
beneficial than the existing system, as it analyse the data
non-cancerous one. The proposed system consists of two
according to the grouping class variable and gives reduced
phases. First phase consist of texture feature extraction
feature set with high class accuracy. The classification
from brain MRI images using statistical features. In the
accuracy obtained is 98.87%.
second phase, classification of brain images on the basis of
these texture feature using BPN and RBFN classifier is This paper is having the limitations of using principal
used for classification. After classification tumor region is component analysis, which reduces the lower
extracted from those images and are again classified as dimensionality of the texture feature.
malignant using the segmentation process.
Zarandi et al[14], in their paper presented a type II fuzzy
The main limitations of the proposed system are in expert system for diagnosing human brain tumor using T1
selecting the optimal features to distinguish between weighted MR images. The proposed system consists of four
classes. This can be overcome by incorporating spatial auto modules: pre-processing, segmentation, feature extraction
correlation by fusing at different level reduces MSE in case and approximate reasoning. A fuzzy rule base by
of RBFN.In future work can be extended in improving RBF aggregating the existing filtering method is used for pre-
using spatial decomposition technique ie like shrinking and processing step, segmentation is done by extending the
kernel catching. For better processing second level probabilistic c mean (PCM) method by using the type II
decomposition method can be implemented for image fuzzy concepts ,mahalanobis distance and kwon validity
wavelet. index. Feature extraction is done by thresholding method

978-1-4799-4190-2/14/$31.00 ©2014 IEEE 873

2014 International Conference on Control, Instrumentation, Communication and Computational Technologies (ICCICCT)

and finally a type II approximate reasoning method is feature extraction, and classifier. In the pre-processing
developed to recognize the tumor grade in brain MRI. histogram equalization and morphological operations are
Experimental result shows that the proposed system is done for enhancing and sharpening the binarize image.
superior in recognizing the brain tumor and its grade than Feature extraction is done by gray level co-occurrence
type I fuzzy expert system. matrix (GLCM). These extracted features are fed as input to
the BPN classifier for classification. There the image is
Its future works include, using parametric operator instead
classified in to normal or abnormal image. The
of standard ones makes the system more adaptive.
classification accuracy is about 96.84%.
Zhang et al [15] in their paper they presented a neural
network based method to classify a given MR brain image Future work can be done by using an improved artificial
as normal or abnormal. This method first employs wavelet neural network with improvement of feature function will
transfom to extract feature from image and then applies the help to achieve well separated data.
technique of principal component analysis (PCA) to reduce El-Dahshan et al [19] in their paper presents a hybrid
the dimensions of features. The reduced features send back technique for the classification of Magnetic Resonance
to back propagation neural network, with which scaled Images (MRI). The proposed method consists of three
conjugate radiant (SCG) is adopted to find the optimal stages: feature extraction, dimensionality reduction and
weight of neural network. The classification accuracy of classification. In the first stage feature extraction is done by
both training and test image are 100%. Discrete Wavelet Transform (DWT). In the second stage,
In future the proposed method can be employed for MR the feature of Magnetic Resonance Image has been reduced
images with other contrast mechanisms such as T1W, by using Principal Component Analysis (PCA). In the
proton density weighted. classification stage two classifiers have been developed.
The first classifier is based on feed forward back
Kharrat et al [16] in their paper they proposed a hybrid propagation artificial neural network (FP-ANN) and the
approach for the classification of brain tissue in magnetic second classifiers have been used to classify subjects as
resonance image (MRI) based on genetic algorithm and normal or abnormal MRI human image. A classification of
support vector machine (SVM). The proposed method has 3 97% and 98% has been obtained by FP-ANN and K-NN
steps:-feature extraction, feature selection and respectively. The results shows that the proposed technique
classification. In this a wavelet based texture feature is is robust and effective compared with other recent work.
derived and optimal texture features are extracted from
normal and tumor region by using spatial gray level The limitation of this work is that it requires fresh training
dependence method(SGLDM) and these features are given each time when ever, there is a change in the image data
as input to SVM classifier. The optimal features are used to base. In future the work can be extended for the developed
classify the brain tissue into normal, benign and malignant technique for processing the pathological brain tissue.
tumor and their performance is evaluated. The accuracy of Georgiadis et al[20] employed a software system for
the proposed system varies from 96.37 to 98.99%. This discrimination between metastatic and primary brain tumor
paper is having the limitations of having fresh training set on MRI. The study employed a Modified Probabilistic
whenever there is change in image database. Neural Network classifier (PNN) and incorporating a
Lashkari [17] this paper introduces a novel automatic brain nonlinear least square feature transformation (LSFT) in to
tumor detection method that uses T1,T2 weighted and PD, the PNN classifier.In this six features are extracted from the
MR images to determine any abnormality in the brain T1 weighted image. In this they accured a good
tissue. The proposed technique mainly consists of four classification accuracy of 95.24% for discriminating
stages- pre-processing, feature extraction, feature selection, between metastatic and primary tumor and 93.48% for
classification. In pre-processing stage the increase in distinguishing gliomas from meningiomas in the first level
contrast between normal and abnormal brain tissue is done and in the second level classification accuracy is
and DFT of the image is computed. In the next stage, 100%.selecting the ROI across the pixel. In the first level
feature extraction is done. Here non-statistical feature gliomas and menigiomas were grouped in to primary brain
extraction namely Gabour wavelet is used and the feature tumor with the help of third degree LSFT PNN classifier
selection is done by means of kernel-F score method. These and in the next level the primary brain tumor is again
selected features are then sent to the multilayer perception classified in to gliomas and meningiomas with the help of
neural network for further classification. However the second degree LSFT PNN.
system is having the limitation of using all the 3 modalities The added advantage of this method is the improved
such as T1, T2 weighted and PD MR Images. performance, increased class separability and
Its future works include the integration of features derived dimensionality reduction. Limitation of this work is that,
from fractural analysis which describes the local texture or external cross validation method is used to avoid over
ruggedness in terms of an estimated value called Hurst fitting condition and it has very less discriminant accuracy
Coefficient Selvaraj et al. [21] proposed an intelligent clas sification
Jain [18] in this paper he proposed a method that classifies technique for the identification of normal and abnormal
brain tumor based on artificial neural network. The brain slices. An advanced classification technique, based on
proposed system consists of three stages: pre-processing, least square support vector machine (LS-SVM) was

978-1-4799-4190-2/14/$31.00 ©2014 IEEE 874

2014 International Conference on Control, Instrumentation, Communication and Computational Technologies (ICCICCT)

proposed. The features to be classified are derived from the algorithm. b) The FLAIR images are segmented to compute
slices. This classifier using linear as well as nonlinear radial the volume of edema. c). T1 and T1E images are registered
basis function (RBF) kernels are compared with other and a different image is obtained. d). The difference image
classifiers like SVM with linear and nonlinear RBF kernels , is segmented. e) All segmented region volumes are
RBF classifiers, Multilayer perceptron‟s (MLP) classifier computed.
and KNN classifier. The classification accuracy is about
However the limitations in this method that it runs
98.64%. In future this could be used further for the
automatically except for the choice of a volume of interest
classification of images with different pathological
and seed point. A user verification step must be added to
conditions, disease status and types.
ensure the quality.
Lin et al [22] in their paper they introduce a method for
precise accurate and efficient quantification of brain tumor
via MR imaging. The purpose of this work is to build a
computerised system to evaluate its effectiveness for
routine clinical work. In this the image (FLAIR, T1 and T2)
are processed independently. The steps involved are
a) MRI image are first standardised for each protocol and
subsequent operation is done on the standardised image.
The segmentation is done through fuzzy connectedness
T able1: Comparison of Different Methods
No Author Year Methods used Limitation Accuracy
1 Xiao et al 2013 K nearest neighbours (KNN) and conventional T hey wrongly assign, a non CSF pixel to 100%
Fuzzy connected C-mean (FCM). the cluster. Undesired pixel are removed by
placing a global mask , there by leaving the
region as extracted one.
2 Nandagopal 2013 SVM is used for segmentation. A combination Whenever there is change in image data set, 97.5%
& Rajamony of WST and WCT is used for feature extraction. It need new training set for Gaussian SVM
Genetic algorithm is used to select the optimal classifier .T his method is applied only to
texture feature. PNN is used for classification. CT images.
3 Kalbhani et al 2013 2D DWT of the input image is calculated fist. they cannot model asymmetric with respect 97.62% and
T hen the features are extracted by PCA and to the sign of past values 98.21%
LDA. T he extracted feature are applied to KNN
and SVM classifier for classification.
4 Sindhu mol et al 2013 It is based on spectral angle based feature Low threshold value can lead to over 98% &
extraction and spectral clustering independent clustering. Cost of Feature extraction due to 96.1%
component analysis (SC-ICA) clustering is high.
5 Navarro et al 2013 It introduce a new method for feature selection T here are many issues involving different
and dimensionality reduction by using off the pathologies and pathological grouping.
shelf classifiers on various HMRS modalities. 95%

6 Saritha et al 2013 Approach is by integrating wavelet entropy Whenever there is an increase in image 100%
based spider web plots and probabilistic neural database fresh training is required.
network for the classification of Brain MRI.
7 Sumitra and 2013 Uses a neural network technique for the Over discriminant accuracy is less. 73%
Saxena classification of MRI images. T he feature Determination of Unique feature vector is
extraction is done by using PCA not possible
8 Jayachandran and 2013 Based on hybrid algorithm for detection of brain Principal component analysis, reduces the 95.3%
Dhanasekhara tumor MRI using statistical and SVM classifier. lower dimensionality of the texture feature.
9 Nantha Gopal and 2012 Dominant run length and co-occurrence texture Whenever there is change in the data set, it 96.4%
Sukanesh feature are selected by SVM. Features are requires a new training set and this method
extracted and selected using students t-test. is applied only to CT images.
T he Features selected are then fed to the SVM
and PNN classifiers.
10 Deepa and Devi 2012 T his methods exploit the capability of back Difficulty in selecting the optimal features 98.6%
propagation and Radial Basis Function neural to distinguish between classes.
network function to classify brain image.

11 Mustara & 2012 T he texture feature is extracted by using Computational cost is high.
Suchalatha GLCM. T hese features are compared with the
stored features in knowledge base. Finally a 97.6%
neuro fuzzy classifiers is used for classification.
12 Jafari and 2012 Hybrid approach for the detection of brain Wavelet transform require large storage 83.22%
Shafaghi tumor tissue in magnetic resonance image based and its computational cost is high.
on genetic algorithm and support vector

978-1-4799-4190-2/14/$31.00 ©2014 IEEE 875

2014 International Conference on Control, Instrumentation, Communication and Computational Technologies (ICCICCT)

13 Rathi and Palani 2012 T hey combines the intensity, texture, shape By using principal component analysis, 98.87%
based features and classifies the tumor as white which reduces the lower dimensionality of
matter, graymatter, CSF, abnormal and normal the texture feature.
area. T he classifier used is support vector
machine (SVM)
14 Zarandi et al 2011 A type II fuzzy expert system for diagnosing High computational cost Complexity and 98.5%
human brain tumor. optimization

15 Zhang et al 2011 Neural network based method to classify a It require fresh training set whenever there 100%
given MR brain image. is an increase in image data base
16 . Kharrat et al 2010 based on genetic algorithm and support vector It requires fresh training set whenever there 98.97%
machine (SVM) and feature selection is done by is change in image database.
using SGLDM
17 Lashkari 2010 In pre-processing contrast normalization is done It uses all the 3 modalities such as T 1, T 2 96.37% to
and DFT of the image is computed. feature weighted and PD MR Images as a result it 98.77%
extraction is done by kernel-F score and is time consuming.
multilayer perception neural network is used
for classification
18 Jain 2010 In pre-processing histogram equalization and When a new image is used for classification 96.84%
morphological operations are done. Feature only that selected features are extracted and
extraction is done by GLCM. T hese extracted trained classifier is used for classification.
features are fed to the BPN classifier for
19 El-Dahshan et al 2010 Feature extraction is done by DWT and the Is that it requires fresh training each time 97%
feature are reduced by PCA. Two classifiers are when ever, there is a change in the image and
used for classification ie. FP -ANN and KNN. data base. 98%
20 Georgiadis et al 2008 incorporating a nonlinear least square feature External cross validation method is used to 95.25%
transformation (LSFT ) in to a modified PNN avoid over fitting condition and it has very 93.48%
classifier less discriminant accuracy.
21 Selvaraj et al. 2007 An advanced classification technique based on More time consuming 98.64%
least square support vector machine (LS-SVM)
is used .
22 Lin et al 2005 Segmentation is done through fuzzy It runs automatically except for the choice 96.5%
connectedness algorithm.In this three images of a volume of interest and seed point. A
(FLAIR, T 1 and T 2) are processed user verification step must be added to
independently and region volumes are ensure the quality.
based on GARCH variances Series”, ELSEVIER Biomedical Signal
Processing and Control 8(2013) 909-919

III. CONCLUSION [4]. Sindhumol S, Anil Kumar, Kannan Balakrishnan “spectral clustering
independent component analysis for tissue classification from brain
MRI” ELSEVIER, Biomedical signal processing and control
In this paper we have accomplished a partial survey of (2013)667-674
various classification techniques for MRI brain image. A [5]. Navarro ,David Hackney,” Feature and model selection with
discriminatory visualization for diagnostic classification of brain
comparative study is made on various techniques. After tumor. ELSEVIER, computerized Medical Imaging and Graphics
evaluation of well-known technique it is clearly shown the 29(„05) 21-34
various methods which can detect the tumor efficiently and [6].M.Saritha ,K.Paul Joseph,Abbraham T Mathew,”Classification of MRI
brain images using combined wavelet entropy based spider web plot
provide accurate result. This work will be extended for new and probailistic neural network”ELSEVIER , pattern Recognition
algorithm for brain tumor detection which will provide letters 34 (2013)2151-215
more efficient result than the existing methods in near [7]. Sumitra , Saxena”Brain Tumor Classification Using Back Propagation
Neural Network”,I.J.Image,Graphics and Signal
future. Computational time will also be considered to Processing,2013,2,45-50
compare this technique efficiently. As the diagnosis tumor [8].Jayachandran .A & R.Dhanasekaran ,”Brain tumor Detection and
is a complicated and sensitive task, accuracy and reliability Classification of MRI Using T exture Feature and Fuzzy SVM
Classifiers”,Research Journal of Applied Sciences,Engg and T ech
are always assigned much importance .Hence an elaborate 6(12):2264-2269,2013
methodology that high lights new vistas for developing [9].A.Padma Nanthagopal & R.Sukanesh Rajamony” Classification of
more robust image segmentation technique is much sought. Benign and Malignant Brain Tumor CT images using wavelet texture
parameters and Neural Network Classifiers” Publication:J0urnal of
visulization society, Vis 16:19-28DOI 10.1007/S12650-012-0153-Y
[10]. Deepa and Devi,” Feature and model selection with discriminatory
[1]. Kai xiao,A.Lei Liang, Hai Bing Guan, Aboul Ella Hassanien visualization for diagnostic classification of brain tumor,”int journal
,”Extraction and Application of Deformation Based Feature in of computer science &engineerig IJSCET vol no4 2229-3345,2013
Medical Images”,ELSEVIER Neuro computing 2013 [11]. Mustara & Suchalatha,” Brain cancer classification using GLCM
[2].Padma Nanda Gopal &R.Sukanesh,” wavelet statistical feature based based feature extraction in artificial neural network”, ELSEVIER,
segmentation and classification of brain computed tomography computerized Medical Imaging and Graphics 29(2005) 21 -34.
images”IET Image Prosess Vol7 pp 25-32 2013
[3]. Hashem Kalbkhani, Mahrokh G Shayesteh, Behrooz Zali-vargahan “ [12]. Mehdi Jafari & Reza Shafaghi,”A Hybrid Approach for Automatic
Robust algorithm for Brain Magnetic Resonance Image Classification T umor Detection of Brain MRI Using Support Vector Machine And

978-1-4799-4190-2/14/$31.00 ©2014 IEEE 876

2014 International Conference on Control, Instrumentation, Communication and Computational Technologies (ICCICCT)

Genetic Algorithm” Global Journal of Science, Engineering and

T echnology(ISSN:2332-2441) ,2012
[13].V.P.Gladis Pushpa Rathi and S.Palani,”Linear Discriminant Analysis
For Brain T umor Classification Using Feature Selection”,Int Journel
Of communication and Engineering,vol-05-No 5.Issue March 2012
[14]. M.H Fazel Zarandi , M.Zarinbal,M.Izadi ,”Systematic Image
processing for diagnosis Brain T umor :A type II expert system
approach”,ELSEVIER applied soft computing 11(2011) 285 -294
[15]. Yudong Zhang, Zhengchao Dong, Lenan Wu “A Hybrid Method for
MRI Brain Image Classification”, ELSEVIER Expert System with
Application 38 (2011) 10049-100
[16]. Ahmed Kharrat, Karim Gasmi, Mohamed Ben ,Nacera,” A hybrid
approach for automatic classification of Brain using Genetic algorithm
and SVM ”, Leonardo Journal of Sciences Issue17 July- December
[17]. Amir Ehsan Lashari,”A Neural Network Based Method For Brain
Abnormality Detection in MR Images Using Gabour Wavelet”,Int
Journal of computer Application(0975-8887)volume 4 July 2010
[18]. Shewta Jain ,”Brain Classification Using GLCM Based Feature
Extraction in Artificial Neural Network”,Int journal of Computer
Science &Engineering T echnology(IJCSET ),ISSN:2229 -
3345,vol.4.No.07Jul 2013.
[19]. Ei-Sayed Ahmed Ei Dahshan, T amer Hosny, Abdel-Badeesh
,”Hybrid Intelligent Technique for MRI Brain Image Classification”
,ELSEVIER digital signal processing 20(2010) 433 -441
[20].Pantelis georgiadis, Dioisis cavouras,Menelaos Malamas,” Improving
Brain T umor Characterization On MRI By Probabilistic Neural
Network and Non Linear Transform of Texture Feature”, ELSEVIER
:computer methods and program in Biomedicine 2008
[21].H.Selvaraj, S.T hamaraiSelvi, D.Selvathi,L .Gewali ,”Brain MRI
Slices Classification Using Least Square Support Vector
Machine”ICMED,Vol 1.No1.Issue,Page 21
[22].Geng-Cheng Lin,Wen –June Wang,Chuin –Mu Wang, Sheng-Yih
Sun,‟Automated Classification of multi spectral MR images using
Linear Discriminant Analysis”, ELSEVIER computerized medical
Imaging and graphics 34(2010) 251-268.
[23]. C. Ramalakshmi,A.Jaya Chandran,”Automatic Brain T umor
Detection in MR Images Using Neural Network Based
Classification”,int Journal of science research and
management(IJSRM)volume Issue 2IIpages 109-113II2013
[24]. Jianguo Liu, Jayaram Kudupa, Dewey Odhner, David Hackney, “A
System for Brain T umor Volume Estimation Via MR Image and
Fuzzy Connectedness”, ELSEVIER, computerized Medical Imaging
and Graphics 29(2005) 21-34.

978-1-4799-4190-2/14/$31.00 ©2014 IEEE 877