A NOVEL APPROACH FOR FEATURE EXTRACTION AND SELECTION ON MRI IMAGES FOR BRAIN TUMOR CLASSIFICATION

David C. Wyld, et al. (Eds): CCSEA, SEA, CLOUD, DKMP, CS & IT 05, pp. 225–234, 2012.
© CS & IT-CSCP 2012 DOI : 10.5121/csit.2012.2224
A NOVEL APPROACH FOR FEATURE EXTRACTION AND
SELECTION ON MRI IMAGES FOR BRAIN TUMOR
CLASSIFICATION
V.P.Gladis Pushpa Rathi1
and Dr.S.Palani2
1
Department of Computer Science and Engineering, Sudharsan Engineering
College Sathiyamangalam, Pudukkottai , India
gladispushparathi@gmail.com
2
Department of Electronics and Communication Engineering , Sudharsan
Engineering College Sathiyamangalam, Pudukkottai , India
palani_keeranur@yahoo.co.in
ABSTRACT
Feature extraction is a method of capturing visual content of an image. The feature extraction is
the process to represent raw image in its reduced form to facilitate decision making such as
pattern classification. The objective of this paper is to present a novel method of feature
selection and extraction. This approach combines the Intensity, Texture, shape based features
and classifies the tumor as white matter, Gray matter, CSF, abnormal and normal area. The
experiment is performed on 140 tumor contained brain MR images from the Internet Brain
Segmentation Repository. PCA and Linear Discriminant Analysis (LDA) were applied on the
training sets. The Support Vector Machine (SVM) classifier served as a comparison of
nonlinear techniques Vs linear ones. PCA and LDA methods are used to reduce the number of
features used. The feature selection using the proposed technique is more beneficial as it
analyses the data according to grouping class variable and gives reduced feature set with high
classification accuracy.
KEYWORDS
Linear Discriminant Analysis, BrainTumor, Shape, Intensity, Texture, PCA, SVM, MRI
1. INTRODUCTION
Early and accurate diagnosis of brain tumor is the key for implementing successful therapy and
treatment planning. However the Diagnosis is a very challenging task due to the large variance
and complexity of tumor characterization in images, such as size, shape, location and intensities
and can only be performed by professional neuroradiologists. In the recent past several research
works have been done for the diagnosis and treatment of brain tumor. MRI is the viable option
now for the study of tumor in soft tissues. The method clearly finds tumor types, size and
location. MRI is a magnetic field which builds up a picture and has no known side effects related
to radiation exposure. It has much higher details in soft tissues. Researcher had proposed various
features for classifying tumor in MRI. The statistical, Intensity, Symmetry, Texture features etc,

226 Computer Science & Information Technology (CS & IT)
which utilize 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.
In image processing, feature extraction is a special form of dimensionality reduction. When the
input data to an algorithm is too large to be processed and it is suspected to be notoriously
redundant (much data, but not much information) then the input data will be transformed into a
reduced representation set of features (also named features vector). Transforming the input data
into the set of features is called feature extraction. If the features extracted are carefully chosen it
is expected that the features set will extract the relevant information from the input data in order
to perform the desired task using this reduced representation instead of the full size input.
Feature extraction involves simplifying the amount of resources required to describe a large set
of data accurately. When performing analysis of complex data, one of the major problems stems
from the number of variables is involved. Analysis with a large number of variables generally
requires a large amount of memory and computation power or a classification algorithm which
over fits the training sample and generalizes poorly to new samples. Feature extraction is a
general term for methods of constructing combinations of the variables to get around these
problems while still describing the data with sufficient accuracy.
Feature selection is the technique of selecting a subset of relevant features for building robust
learning models By removing most irrelevant and redundant features from the data, feature
selection helps improve the performance of learning models by:
• Alleviating the effect of the curse of dimensionality.
• Enhancing generalization capability.
• Speeding up learning process.
• Improving model interpretability.
•
Feature selection also helps people to acquire better understanding about their data by telling
them which are the important features and how they are related with each other.
This Paper is organized as follows. In section 2 we describe normalization, and feature extraction
and selection. In section 3 comparative analysis of PCA and LDA is described. In section 4 tumor
classification and experimental results are discussed. The conclusions are given in section 5.
2. PROPOSED METHOD
The architecture of our system is illustrated in Figure 1.The major components of our system are
Brain tumor Database, Normalisation, Feature selection, Feature extraction and Classification.

Computer Science & Information Technology (CS & IT) 227
Figure 1. Architecture of proposed method
2.1. Data Description
Experiments are conducted on MR images collected from 20 different patients with gliomas.
Each patient has 3 sequences of MR images T1, T2 and FLAIR. Each volume contains 24 slices
in axial plain with 5 mm slice thickness. MR imaging was performed on 3.0T siemens devices.
The imaging conditions of different protocols are; T1 weighted, T2 weighted, and Flair weighted.
The MRI image data description of the proposed method is shown in table 1.Each set of features
are individually normalized to the range of 0 to 255.
Table 1. Data Description
Attribute Description Value
Age Age in Years 17 to 83
Sex Sex Men -46, Women -52
Matrix
size
Size of the
matrix
192*256*192
Voxel size Size of the
voxel
0.98*0.98*1mm
Sequences MRI image
sequences
Axial 3D T1 weighted , Sagittal 3D T2
weighted , Fluid Attenuated Inversion
Recovery (FLAIR)

2.2. Normalization
Initially, these MRI images are normalized to gray level values from 0 to 1 and the features are
extracted from the normalized images. Since normalization reduces the dynamic range of the
intensity values, feature extraction is made much simpler.
2.3. Feature Extraction
Features, the characteristics of the objects of interest, if selected carefully are representative of the
maximum relevant information that the image has to offer for a complete characterization of a
lesion. Feature extraction methodologies analyse objects and images to extract the most
prominent features that are representative of the various classes of objects. Features are used as
inputs to classifiers that assign them to the class that they represent. The purpose of feature
extraction is to reduce the original data by measuring certain properties, or features, that
distinguish one input pattern from another pattern. The extracted feature should provide the
characteristics of the input type to the classifier by considering the description of the relevant
properties of the image into feature vectors. In this proposed method we extract the following
features.
Shape Features - circularity, irregularity, Area, Perimeter, Shape Index
Intensity features – Mean , Variance, Standard Variance, Median Intensity, Skewness,
and Kurtosis
Texture features –Contrast, Correlation, Entropy, Energy, Homogeneity, cluster shade,
sum of square variance.
Accordingly, 3 kinds of features are extracted, which describe the structure information of
intensity, shape, and texture. These features certainly have some redundancy, but the purpose of
this step is to find the potential by useful features. In the next step the feature selection will be
performed to reduce the redundancy.
2.4. Feature Selection
Feature selection (also known as subset selection) is a process commonly used in machine
learning, wherein a subset of the features available from the data is selected for application of a
learning algorithm. The best subset contains the least number of dimensions that contributes to
high accuracy; we discard the remaining, unimportant dimensions.
2.4.1. Forward Selection
Start with no variables and add them one by one, at each step adding the one that decreases the
error the most, until any further addition does not significantly decrease the error. We use a
simple ranking based feature selection criterion, a two –tailed t-test, which measures the
significance of a difference of means between two distributions, and therefore evaluates the
discriminative power of each individual feature in separating two classes. The features are
assumed to come from normal distributions with unknown, but equal variances. Since the
correlation among features has been completely ignored in this feature ranking method, redundant
features can be inevitably selected, which ultimately affects the classification results. Therefore,
we use this feature ranking method to select the more discriminative feature, e.g.by applying a
cut-off ratio (p value<0.1), and then apply a feature subset selection method on the reduced
feature space, as detailed below

2.4.2. Backward Selection
Start with all the variables and remove them one by one, at each step removing the one that
decreases the error the most (or increases it only slightly), until any further removal increases the
error significantly. To reduce over fitting, the error referred to above is the error on a validation
set that is distinct from the training set. The support vector machine recursive feature elimination
algorithm is applied to find a subset of features that optimizes the performance of the classifier.
This algorithm determines the ranking of the features based on a backward sequential selection
method that remove one feature at a time. At each time, the removed feature makes the variation
of SVM based leave-one-out error bound smallest, compared to removing other features.
2.5. Classification
There are many possible techniques for classification of data. Principal Component Analysis
(PCA) and Linear Discriminant Analysis (LDA) are the two commonly used techniques for data
classification and dimensionality reduction. Linear Discriminant Analysis easily handles the case
where the within-class frequencies are unequal and their performance has been examined on
randomly generated test data. This method maximizes the ratio of between-class variance to the
within-class variance in any particular data set thereby guaranteeing maximal separability. The
use of Linear Discriminant Analysis for data classification is applied to classification problem in
speech recognition We decided to implement an algorithm for LDA in hopes of providing better
classification compared to Principal Components Analysis. The prime difference between LDA
and PCA is that PCA does more of feature classification and LDA does data classification. In
PCA, the shape and location of the original data sets change when transformed to a different
space whereas LDA doesn’t change the location but only tries to provide more class separability
and draw a decision region between the given classes. The classification process is divided into
the training phase and the testing phase. In the training phase known data are given. In the testing
phase, unknown data are given and the classification is performed using the classifier after
training. The accuracy of the classification depends on the efficiency of the training.
2.5.1. Linear Discriminant Analysis
LDA methods are used in statistics, pattern recognition, and machine learning to find a linear
combination of features. LDA attempts to express one dependent variable as a linear combination
of other features or measurements. LDA is also closely related to PCA and factor analysis in that
they both look for linear combination of variables which best explain the data. LDA explicitly
attempts to model the difference between the classes of data. PCA on the other hand does not take
into account of any difference in class, and factor analysis builds the feature. Combination is
based on differences rather than similarities. LDA searches for those vectors in the underlying
space that best discriminable among classes. More formally given a number of independent
features relative to which the data is described, LDA creates a linear combination of those which
yields the largest mean differences between the desired classes. We define two measures: 1) one
is called within- class scatter matrix as given by
Sw= ( )( )Tj
i
c
j
Nj
i
j
i xx j
1 1
j µµ −−∑∑= =
(1)
where xi
j
is the ith
sample of class j, µj is the mean of class j, c is the number of classes, and µj is
the number of samples in class j and 2)between class scatter matrix

Sb= ( )( )T
c
j
µµµµ j
1
j −−∑=
(2)
where µ represents the mean of all classes.
2.5.2. Support Vector Machine
Support vector machines are a state of the art pattern recognition technique grown up from
statistical learning theory. The basic idea of applying SVMs for solving classification problems
can be stated briefly as follows: a) Transform the input space to higher dimension feature space
through a non-linear mapping function and b) Construct the separating hyperplane with maximum
distance from the closest points of the training set.
In the case of linear separable data, the SVM tries to find among all hyper planes that minimize
the training error, the one that separates the training data with maximum distance from their
closest points
0=+• bxw (3)
with w and b weight and bias parameters respectively.
In order to define the maximal margin hyperplane (MMH) the following constrains must be
fulfilled:
Minimize ( ) 1||||
2
1 2
≥+• bxwwithyw ii (4)
This is a classic nonlinear optimization problem with inequality constraints. It can be solved by
the karush-kuhn-Tucker (KKT) theorem by introducing Lagrange multipliers
maximize j
T
ijij
l
ji
i
l
i
i xxaayya ∑∑ ==
−
1,1 2
1
(5)
subject to 00
1
≥=∑=
ii
l
i
i andaya (6)
The solution of w is:
w= ii
l
i
i xya∑=1
(7)
The only nonzero solutions define those training data (usually a small percentage of the initial
data set) that are necessary to form the MMH and are called support vectors. The optimal hyper
plane theory is generalized for non-linear overlapping data by the transformation of the input
vectors into a higher dimensional feature space through a mapping function
( ) [ ] fT
nn
n
i RxaxaxaxzRx ∈ΦΦΦ=→∈ )(),.....,(),( 2211

The KKT conditions transform to
Maximize )(
2
1
1,1
jijij
l
ji
i
l
i
i xxKaayya ∑∑ ==
− (8)
subject to 00
1
≥=∑=
i
l
i
ii andaya (9)
The optimization problem is solved using the MATLAB optimization toolbox
2.6. Experiment Results
In all the selected 60 features, there are 22 Intensity based features, 5 Shape based features, 33
texture based features . It is found that there are 3 kinds of features extracted in our work and are
all useful for the classification. Besides, the distribution of T1, T2, and FLAIR are 10, 20,30
respectively. It means FLAIR provides the most information for tumor segmentation, T2 provides
less and T1provides the least. This result is in accordance with the conclusion in Medical Imaging
that FLAIR and T2 are more sensitive in pathological discrimination than T1. The distribution of
selected features is shown in table 3.
Table 3: Distribution of Selected Features
Features T1 T2 FLAIR TOTAL
Intensity 6 5 11 22
Shape 1 1 3 5
Texture 8 5 20 33
Total 10 20 30 60
Efficiency or accuracy of the classifiers for each texture analysis method is analysed based on the
error rate. This error rate can be described by the terms true and false positive and true and false
negative as follows:
True Positive (TP) : The test result is positive in the presence of the clinical abnormality.
True Negative (TN): The test result is negative in the absence of the clinical abnormality.
False Positive (FP): The test result is positive in the absence of the clinical abnormality.
False Negative (FN): The test result is negative in the presence of the clinical abnormality
FP= false positive pixels number /tumor size (10)
FN=false negative pixel number / tumor size (11)
Correct rate=FP+FN (12)

The average correct rate by the presented method is 97.82% with FP of 1.0% and FN of 2.50%.
All the features produce classification accuracy of 98.87% using LDA. The extracted four PCA
components are classified using LDA and SVM classification and the accuracy achieved is 96%. .
The overall accuracy percentage details are shown in fig 1.
Fig: 1 overall accuracy performance of LDA, PCA and the proposed method
The comparative analysis of the proposed method and the existing algorithms are shown in
table4.
Table 4.Comparative analysis
Classification
accuracy
FP FN Correct rate With feature
selection
Without
feature
selection
Proposed
method
1.00% 2.50% 97.82% 98.87% 98.77%
KNN 2.75% 7.51% 93.50% 98.48% 95.47%
Fuzzy
connectedness
2.95% 5.02% 92.04% 98.35% 97.47%
AdaBoost 3.15% 6.07% 90.05% 98.74% 98.55%

Fig: 2 Comparative analysis of existing algorithms and the proposed method
3. CONCLUSIONS
The method mainly includes the steps, Image collection, Normalization, Intensity, shape
and Texture feature extraction, feature selection and classification. The proposed method
performs better than the existing algorithms. The shape, Intensity and Texture features
are extracted and used for classification. Vital features are selected using LDA. The
results are compared with dimension reduction techniques like PCA. The number of
features selected or features extracted by PCA and The classification accuracy by SVM is
98.87%. It is expected that the information of new imaging technique fMRI and the
Image MOMENTS when added into the scheme will give more accurate results which
will be taken for future work.
ACKNOWLEDGEMENTS
The work done by V.P.Gladis Pushpa Rathi, Dr. S.Palani is supported by Sudharsan Engineering
College Sathiyamangalam. Pudukkottai India
REFERENCES
[1] K.M. Iftekharuddin, On techniques in fractal analysis and their applications in brian MRI, in: T.L.
Cornelius (Ed.), Medical imaging systems: technology and applications, Analysis and Computational
Methods, vol. 1, World Scientific Publications, 2005, ISBN 981-256-993-6.
[2] L.P. Clarke, R.P. Velthuizen, M.A. Camacho, J.J. Heine, M. Vaidyanathan, L.O. Hall, R.W.
Thatcher, M.L. Silbiger, MRI segmentation: methods and applications, Magn. Reson. Imaging 13 (3)
(1995) 343-368.
[3] J.C. Bezdek, L.O. Hall, L.P. Clarke, Review of MR image segmentation techniques using pattern
recognition, Med. Phys. 20 (4)(1993)1033-1048.

[4] H.S. Zadech, H.S. Windham, Optimal linear transformation for MRI feature extraction, IEEE Trans.
Med. Imaging 15 (1996) 749-767.
[5] H.S. Zadech, J.P. Windham, A comparative analysis of several transformations for enhancement and
segmentation of magnetic resonance image scene sequences, IEEE Trans. Med. Imaging 11 (N3) (1992)
302-318.
[6] D. Wang, D.M. Doddrell, A segmentation-based partial-volume-compensated method for an accurate
measurement of lateral ventricular volumes on T1-weighted magnetic resonance images, Magn. Reson.
Imaging 19 (2001) 267-272.
[7] X. Zeng, L.H. Staib, R.T. Schultz, J.S. Duncan, Segmentation and measurement of the cortex from 3-D
MR images using coupled-surfaces propagation, IEEE Trans. Med. Imaging 18 (10) (1999) 927-937.
[8] M. Just, M. Thelen, Tissue characterization with T1, T2 and proton density values: results in 160
patients with brain tumors,Radiology 169 (1988) 779-785.
[9] M.W. Vannier, R.L. Butterfield, D. Jordan, W.A. Murphy, R.G. Levitt, M. Gado, Multispectral analysis
of magnetic resonanceimages, Radiology 154 (1) (1985) 221-224.
[10] M.W. Vannier, T.K. Pilgram, C.M. Speidel, L.R. Neumann, D.L. Rickman, L.D. Schertz, Validation of
magnetic resonance imaging(MRI) multispectral tissue classification, Comput. Med. Imaging Graph. 15
(4) (1991) 217-223.
Dr.S Palani graduated from P.S.G. College of Technology, Coimbatore, did his post
graduate studies at IIT Kharagpur and the doctoral degree from Regional Engineering
College Trichy. He is a faculty member of the department of Electrical and Electronics
Engineeri ng, Sudharsan Engineering College, Pudukkottai, India. He has more than 40
years of teaching experience. His field of Interest includes Control Systems, Electrical
Engineering and Digital signal Processing.
V.P.GladisPushpaRathi graduated from Cape Institute of Technology, Tirunelveli, did her
post graduate studies at M.S.University Tirunelveli and is now doing her Ph.D in
AnnaUnivers ity,Trichy. She is a faculty member of the department of Computer Science
and Engineering, Sudharsan Engineering College. She has 5 years of teaching experience.
Her field of Interest includes Digital Image Processing, Soft computing, and Datamining.

A NOVEL APPROACH FOR FEATURE EXTRACTION AND SELECTION ON MRI IMAGES FOR BRAIN TUMOR CLASSIFICATION

More Related Content

What's hot (19)

Similar to A NOVEL APPROACH FOR FEATURE EXTRACTION AND SELECTION ON MRI IMAGES FOR BRAIN TUMOR CLASSIFICATION (20)

More from cscpconf (20)

Recently uploaded (20)

A NOVEL APPROACH FOR FEATURE EXTRACTION AND SELECTION ON MRI IMAGES FOR BRAIN TUMOR CLASSIFICATION