Segmentation and Classification of MRI Brain Tumor

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 07 | July -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 683
Segmentation and Classification of MRI Brain Tumor
Mukambika P. S.1, Uma Rani K.2
M.Tech (2nd year) Student, Biomedical Signal Processing and Instrumentation, Dept. of IT, SJCE, Mysuru,
Karnataka, India
Associate Professor, Dept. of IT, SJCE, Mysuru, Karnataka, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Bio-medical image processing is the most
challenging and emerging field in medical diagnosis.
Processing of MRI images is one of the difficult parts of this
field. The present work presents the comparison study of
two techniques used for tumor detection of MRI images. One
is based on the Level set method that uses the non
parametric deformable models with active contour to
segment the brain tumor from the MRI brain images. The
other one is the K-means segmentation algorithm. After the
segmentation decision making is performed in two stages:
Feature extraction using Discrete Wavelet Transform and
Gray Level Co-occurrence Matrix, and classification using
the Support Vector Machine. It is observed that the results of
segmentation accuracies from the proposed methods are
comparatively high with the existing method.
Key Words: Magnetic Resonance Imagining (MRI),
Brain Tumor, Level set, K-means, Performance
Evaluation, Discrete Wavelet Transform (DWT), Gray
Level Co-occurrence Matrix (GLCM), Support Vector
Machine (SVM).
1. INTRODUCTION
Automated and accurate classification of Magnetic
Resonance (MR) brain images is extremely important for
medical analysis and interpretation. The classification of
MR brain images is becoming increasingly important in
the medical field since it is crucial for treatment planning
and diagnosing of abnormalities, measuring tissue volume
to examine tumor growth and studying the anatomical
structure and patient‟s following procedure [1]. Medical
image analysis is an important biomedical application
which is highly computational in nature and requires the
aid of the automated systems [2]. These image analysis
techniques are often used to detect the abnormalities in
the human bodies through scan images.
Magnetic Resonance imaging scan of head is the most
common test taken to confirm the presence of a brain
tumor and for identifying its location for selecting
treatment options. Compared to all the other imaging
techniques, MRI is efficient in the application of brain
tumor detection and identification, due to the high
contrast of soft tissues, high spatial resolution, and
absence of any harmful radiation [3]. Manual classification
of magnetic resonance brain tumor images is a challenging
and time-consuming task. Manual classification is highly
prone to error due to inter observer variability and human
error. As a result, the classification results are highly
inferior leading to fatal results.
2. DATASET
Dataset of MRI brain tumor images includes 17 benign and
24 malignant tumor images of different patients which are
DICOM in nature. These images T2 weighted sequence
images obtained from Philips , 3T machine, and are
provided by JSS hospital, Mysuru. These images are
labeled by the expert radiologist.
3. METHODOLOGY
Detection and classification of MRI brain tumor images
involve four modules: Pre processing, Segmentation,
Feature Extraction and Classification. The steps followed
for the implementation are depicted in Fig-1. Skull
removal is used for preprocessing, and the preprocessed
image is given as input for the segmentation process. Two
types of segmentation algorithms are used to extract the
tumor regions: Level set method and K-means algorithm.
Statistical textural features are obtained by using DWT
and GLCM. The obtained features are classified using SVM
as benign or malignant tumors.
Fig -1: Block diagram

3.1 Preprocessing
MRI brain images cannot be fed directly as the input for
the segmentation technique. Therefore skull removal is
performed. Skull is the outer part of the brain. Morphology
technique using double thresholding is applied to remove
the skull out of the MRI brain images
(SonaliPatil&Udupi2012). The binary image b (x, y) of the
original image I(x, y) using double threshold is obtained by
the equation 1.
( ) {
( )
(1)
3.2 Segmentation
Segmentation is the partition of a digital image into similar
regions to simplify the image representation into
something more meaningful and easier to analyze.
3.2.1 Level set method
Level set methods use non parametric deformable models
with active contour energy minimization techniques which
solve computation of geodesics or minimal distance
curves. Level set methods are governed by curvature
defining speeds of moving curves or fronts. There are
large numbers of level set methods developed for
segmentation of medical images and all most all these
methods follow some common generic steps. First
placement of an initial contour arbitrarily, outside or
inside the region of interest, level set φ = signed Euclidean
distance function of the contour and Function φ allowed to
evolving according to first or second derivative partial
differential equation (PDE), then it is reinitialized after a
number of iterations and go to second statement until the
function φ converges or = 0.
In the level set function as shown in Fig-2, a 2D curve r is
represented by a 3D iso-surface φ. Let r(t=0) be a smooth
closed initial curve in the Euclidian plane R2 and let r(t) be
a family of curves generated by moving r(t=0)along its
normal vector field with speed F(k).
Fig-2: Contour and Level Set Evolution from Distance
Function
The propagation front r(t) is embedded as the zero
levelset of φ(t=0)of the higher dimensional function φ.
Φ(x,t=0), where x € R2 are the plots in the image space
defined by,
( ) (2)
where d is the distance from point x to r(t=0) and sign
depends on the position of the point x considering the
initial surface r(t=0). If x is outside (inside) the initial
surface the sign is plus (minus).
3.2.2 K-means method
K-Means algorithm is one of the unsupervised clustering
algorithm due to its efficiency and simplicity in clustering
large data sets [4]. In traditional K-Means algorithm the
numbers of clusters is defined and are initialized apriori.
Next, k cluster centers are randomly chosen. Euclidean
distance is generally considered to determine the distance
between data objects and the cluster center. The distance
between the each data point to each cluster centers are
calculated. The data point is moved to particular cluster
which has the shortest distance. The centroid is then
reestimated. Again each data point is compared with all
the centroids. The process continuous and at the end of
loop, k-centroids change their point step by step until the
centroids do not move any more. This algorithm works on
basis of minimizing squared error function. The K-Means
algorithm always converges to a local minimum. Local
minimum found depends on the initial cluster centroids.
K-means clustering algorithm is summarized as,
1. Randomly initialize the Nc cluster centroid vectors.
2. Repeat
a) For each data vector, assign the vector to the class with
the closest centroid vector, where the distance to the
centroid is determined using,
( ) √∑ ( ) (3)
where denotes the Pth data vector, denotes the centroid
vector of cluster j and k subscripts the number of features
of each centroid vector.
b) Recalculate the cluster centroid vectors, using
= ∑ ( ) (4)
where denotes the number of data vectors in cluster j,
stands for the centroid vector and cj is the subset of data
vectors that form cluster j ,until a stopping criterion is
satisfied.
The K-means clustering process can be stopped when any
one of the following criteria are satisfied: when the
maximum number of iterations has been exceeded, when

there is little change in the centroid vectors over a number
of iterations or when there are no cluster membership
changes.
3.3 Feature Extraction
Features are said to be properties that describe the whole
image. It can also be referred as an important piece of
information relevant for solving the computational task
related to specific application. From the segmented image,
features are computed to encode the useful diagnostic
information.
3.4.2 Gray Level Co-Occurrence Matrix
A statistical method of examining texture that considers
the spatial relationship of pixels is the GLCM, also known
as the gray-level spatial dependence matrix. The GLCM
functions characterize the texture of an image by
calculating how often pairs of pixel with specific values
and in a specified spatial relationship occur in an image,
creating a GLCM, and then extracting the statistical
measures from this matrix. GLCM calculates the co-
occurrence matrix of an image by computing how often a
pixel with a certain intensity i‟ occurs in relation with
other pixel, j‟ at a certain distance d‟ and orientation. The
gray-level co-occurrence matrix can reveal certain
properties about the spatial distribution of the gray levels
in the texture image. In order to extract information
concerning spatial distribution and its orientation, four co-
occurrence submatrices can be calculated across four
scanning directions θ (00 , 450, 900,1350). The features
extracted using GLCM method are:
Contrast: Contrast returns a measure of the intensity
contrast between a pixel and its neighborhood. The
contrast feature is a measure of the amount of local
variations present in an image.
∑ ( )( ) (5)
Correlation: Correlation measures how correlated a pixel
is to its neighborhood. It could also be described as a
measure of linear dependencies among neighboring pixels
in an image.
∑ ( )( )( ) (6)
Energy: Energy can be defined as the quantifiable amount
of the extent of pixel pair repetitions. Energy is a
parameter to measure the similarity of an image. It is also
referred to as angular second moment, and it is defined as.
∑ ( ) (7)
Homogeneity: Homogeneity measures the similarity of
image pixels. It is inversely proportional to the contrast.
∑ ( ) ( | |) (8)
3.4.1 Discrete Wavelet Transform
The input MRI images are decomposed into four sub-
bands as smooth, horizontal, vertical and diagonal images
in order to separate low and high frequencies by applying
discrete wavelet transform (DWT). Decomposition may be
either one level or two level decompositions [5]. The
wavelet decomposition can be obtained by performing the
filtering operations consecutively along horizontal and
vertical directions. The other features extracted are:
Mean: The mean of an image is calculated by adding all
the pixel values of an image divided by the total number of
pixels in an image.
( ) ∑ ( ) (9)
Standard Deviation: The standard deviation is the second
central moment describing probability distribution of an
observed population and can serve as a measure of
inhomogeneity. A higher value indicates better intensity
level and high contrast of edges of an image.
( ) ∑ (
( ( ) )
) ⁄
(10)
Variance: Variance is a measure that is similar to the first
order statistical variables called standard deviation.
Variance is the square of standard deviation.
( ) (11)
Entropy: Entropy is a measure of randomness of intensity
image.
∑ ( ) ( ( )) (12)
Inverse Different Moment: Inverse Difference Moment
(IDM) is a measure of image texture.
∑ ( ) | | (13)
Kurtosis: The shape of a random variable’s probability
distribution is described by the parameter called Kurtosis.
For the random variable𝑋, the Kurtosis is defined as.
∑ (( ) ( )) ( ) (14)
Skewness: Skewness is a measure of the asymmetry of the
data around the sample mean.
∑ (( ) ( )) ( ) (15)

3.4 Classification
After extracting features using DWT and GLCM, they are
directly given to the support vector machine (SVM) for the
classification. The process involves two phases: Training
phase and testing phase. In training phase the patterns in
terms of features and class labels of benign and malignant
tumor are fed to the classifier for training. In testing phase
test pattern is fed and knowledge gained during training
phase will classify the unknown pattern. SVM performs
the robust non-linear classification with the kernel trick
[6]. It finds the separating hyper plane in some feature
space inducted by the kernel function while all the
computations are done in the original space itself. For the
given training set, the decision function is found by solving
the convex optimization.
4 RESULTS AND DISCUSSION
This section provides the experimental results of the
present work.
Preprocessing results:
(3a) (3b)
Fig-3: (3a) Input image (3b) Skull removed image
Segmentation results:
(4a) (4b)
(4c)
Fig-4: Segmentation results using level set method
(4a) Input image (4b) Skull removed image
(4c) Segmented tumor
Fig-5: Segmentation results using K-means method
Classification results:
(6a)
(6b)
Fig-6: Classification results using level set method
(6a) Benign (6b) Malignant

(7a)
(7b)
Fig-7: Classification results using K-means method
(7a)Benign (7b) Malignant
Performance Analysis:
Evaluation of the proposed MRI brain images classification
method is carried out by calculating the performance in
terms of true positive, false positive, true negative and
false negative. By using these matrices sensitivity,
specificity and accuracy can be obtained.
Fig-8: Confusion matrix
 TP-Number of benign tumor region correctly
detected in the MRI images.
 FP-Number of benign tumor region detected as
malignant in the MRI images.
 TN-Number of malignant tumor region correctly
detected in the MRI images.
 FN-Number of malignant tumor region detected
as benign tumor region by the system.
Calculation of the accuracy, sensitivity and specificity is
done according to these equations:
 SENSITIVITY = 𝑇𝑃/(𝑇𝑃+𝐹 )
 SPECIFICITY = 𝑇 /(𝑇 +𝐹𝑃)
 ACCURACY = 𝑇𝑃+𝑇 /(𝑇𝑃+𝑇 +FP+𝐹 )
As the data base contains 41 MRI brain images of two
cases (benign, malignant), the dataset is split. In order to
split the feature subset into the training and testing sets,
the Holdout method is selected, which is part of the model
evaluation used to segregate training and test datasets,
where two-third (70%) of the samples from all the classes
are allocated to the training set and the remaining one-
third (30 %) of the samples from all the classes are
allocated to the testing set.
Table -1: Ratio of images used for training and testing
Class Number of
images
Training
set (70%)
Testing set
(30%)
Malignant 24 17 10
Benign 17 12 7
Total 41 29 17
Table -2: Performance Evaluation of SVM
SVM
(Level set
method)
SVM
(K-means
method)
Confusion
matrix
TP=6 FN=1 TP=6 FN=2
FP=0 TN=10 FP=1 TN=8
Accuracy 94.12% 82.35%
Sensitivity 100.00% 75.00%
Specificity 90.01% 88.89%
From the above results, it is clear that classifier
performace of SVM for Level set method is better
compared to classifier performance of SVM for K-means
method.
5. CONCLUSIONS
This work presents an effective method to segment and
classify the MRI brain images as benign or malignant. It

also presents a comparison study of two segmentation
algorithms namely, Level set method and K-means
algorithm. To differentiate malignant tumor region from
benign tumor region, GLCM, Wavelet transform based
features are extracted and are given to SVM classifier.
The experimental result shows that Level set algorithm
gives better segmentation results compared to k-means
algorithm.
ACKNOWLEDGEMENT
The authors are grateful to Dr. Rajesh, Radiology
Department, J.S.S., Hospital, Mysuru, for helping us to
collect the MRI data of patients.
REFERENCES
[1] H. Zhang, J. E. Fritts, S. A. Goldman,“Image
Segmentation Evaluation: A Survey Of Unsupervised
Methods”, Computer Vision And Image Understanding, Pp.
260-280, 2008.
[2] Mert R. Sabuncu, B. T. Thomas Yeo, Koen Van Leemput,
Bruce Fischl, and Polina Golland. “A Generative Model for
Image Segmentation Based on Label Fusion”, IEEE
Transactions on Medical Imaging, vol. 29, no. 10, pp. 1714-
1729. 2010.
[3] Farjam, R., Parmar, H. A., Noll, D. C., Tsien, C. I., and
Cao, Y. “An approach for computer-aided detection of
brain metastases” in post-Gd T1-W MRI. Magnetic
resonance imaging, vol. 30, no. 6, pp. 824–836. 2012.
[4] Kshitij Bhagwat, Dhanshri More, Sayali Shinde, Akshay
Daga, Assistant Prof. Rupali Tornekar, “ Comparative
Study Of Brain Tumor Detection Using K Means ,Fuzzy C
Means And Hierarchical Clustering Algorithms ”
International Journal Of Scientific & Engineering Research
, vol. 2,no. 6,pp. 80-86, 2013.
[5] Sindhumol S, Anil Kumar, Kannan Balakrishnan,
“Spectral clustering independent component analysis for
tissue classification from brain MRI”, Biomedical Signal
Processing and Control, vol. 8, no.6, pp. 667-674, 2013
[6] K. Uma Rani, Mallikarjun S Holi, “A comparative study
of Neural networks and Support Vector Machines for
Neurological disorder voice classification” in IJERT, vol. 3,
no. 4, pp. 652-658, 2014.

Segmentation and Classification of MRI Brain Tumor

More Related Content

What's hot (19)

Similar to Segmentation and Classification of MRI Brain Tumor (20)

More from IRJET Journal (20)

Recently uploaded (20)

Segmentation and Classification of MRI Brain Tumor