Docstoc

LEVEL BASED NORMAL- ABNORMAL CLASSIFICATION OF MRI BRAIN IMAGES

Document Sample
LEVEL BASED NORMAL- ABNORMAL CLASSIFICATION OF MRI BRAIN IMAGES Powered By Docstoc
					  International Journal of JOURNAL OF COMPUTER (IJCET), ISSN 0976-
 INTERNATIONALComputer Engineering and Technology ENGINEERING
  6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME
                           & TECHNOLOGY (IJCET)

ISSN 0976 – 6367(Print)
ISSN 0976 – 6375(Online)                                                   IJCET
Volume 4, Issue 2, March – April (2013), pp. 403-409
© IAEME: www.iaeme.com/ijcet.asp
Journal Impact Factor (2013): 6.1302 (Calculated by GISI)
                                                                        ©IAEME
www.jifactor.com




       LEVEL BASED NORMAL- ABNORMAL CLASSIFICATION OF MRI
                         BRAIN IMAGES

         Sumesh M. S.1, GopakumarC.2, RejiRajan Varghese3, Abraham Varghese4
          1
            (Computer Engineering, College of Engineering, Chengannur, Kerala, India,)
   2
     (Dept. of Electronics & Communication Engineering, College of Engineering, Chengannur,
                                               India)
            3
              (Senior Resident, Radio diagnosis, Cochin Medical College, Cochin, India)
  4
    (Dept. of Computer science and Engg, Adi- SankaraInsitute of Engineering and Technology,
                                           Kalady, India)



  ABSTRACT

          This work proposes a new concept for the normal- abnormal classification of MRI
  brain images, a level based approach, and compare the result with the existing methods. The
  existing works does not consider the anatomical structure of the brain slices for the
  classification of MRI brain images. In the aspect of image processing, the anatomically
  similarity of the brain slices can be treated as the similarity of brain slices in the viewing
  aspect along with the actual anatomical structure. This work aimed to prove that the
  consideration of the anatomical structure for the normal– abnormal classification will
  improve the result of the classification.
          The existing work shows that the feature vector, statistical features along with gray
  level co-occurrence matrix (GLCM) features with support vector machine (SVM) classifier
  produce better results than other methods. It uses statistical features along with GLCM
  features as feature vector and SVM classifier. Related works in current literatures for the
  normal/abnormal classification of MRI images does not consider the anatomical structure of
  the brain slices. Because of the dissimilarity in the anatomical structure, it may produce
  undesirable results. In this proposed work, the anatomical structure of the brain slices is
  considered for the classification. To accompany this level based concept is introduced here.
  In the level based concept, the brain slices are classified into four levels depending on the
  similarity in the anatomical structure to implement the normal/abnormal classification at that
  particular level.

                                               403
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

Keywords: Brain tumour, Level based classification, Magnetic resonance imaging, Medical
imaging, Support vector machine.

1.     INTRODUCTION

        Medical imaging is widely used for disease diagnosis and treatment evaluation.
Medical imaging techniques and analysis tools enable both doctors and radiologists to
identify and diagnose various disorders [1, 2]. The medical image data obtained from bio-
medical devices have important roles in disease diagnosis. MRI is a non-hazardous method
which detects signals emitted from normal and abnormal tissue, providing clear images of
most tumours [3, 4]. The radiologist or doctor can identify abnormal tissues by examining the
MRI slices based on the visual interpretation. The shortage of radiologists and the large
volume of MRI to be analysed make such readings laborious and cost expensive. Also the
manual classification by mere visual interpretation of the radiologists may cause bad results
due to vision problems. This leads to automated system to aid the doctors and radiologists in
the identification of abnormal brain slices.
        To develop an accurate and sensitive automated system for the normal- abnormal
classification of MRI brain slices, it has to identify a good set of feature vectors that can be
substituted instead of the original image without losing its actual meaning and a good
classifier. The related works suggests several feature vectors and classifiers which are shown
in Table 1. This works shows that the combination of statistical features and GLCM features
[6, 7] along with SVM classifier [8, 9] provides better results than the other methods.

                   Table 1: Related woks for the classification of MRI brain slices

     Pre-processing       Feature Extraction     Feature Reduction          Classification
    WAVELET
                          DWT [1, 8,14,16],
 TRANSFORM [10, 1                                                       SVM [11,12,13,16],
                           GLCM [11,12 ],
         ],                                          PCA [14,15 ],        ANN [14,15 ],
                            SLANTLET
   HISTOGRAM                                           GA [16].          K-NN [11,14,15],
                           TRANSFORM
  EQUALISATION                                                              MLP [11].
                               [13].
       [15].


        The proposed method also use the combination of statistical features along with
GLCM features as feature vector and is used as the input to the SVM classifier. The related
works for the normal/abnormal classification of MRI images does not consider the
anatomical structure of the brain slices. Because of the dissimilarity in the anatomical
structure, it may produce undesirable results. So in this proposed work, the anatomical
structure of the brain slices is considered for the classification. In the aspect of image
processing, the anatomical similarity of the brain slices can be treated as the similarity of
brain slices in the viewing aspect along with the actual anatomical structure. To accompany
this level based concept is introduced here. In the level based concept, the brain slices are
classified into four levels depending on the similarity in the anatomical structure of the brain
slices [17]. That is, classify the brain slices into one of the four levels and implement the
normal/abnormal classification at that particular level.

                                               404
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

2.     METHODOLOGY

        The proposed methodology for the normal- abnormal classification of MRI brain
slices has 2 steps, feature extraction and classification. Significant difference in tissue types,
observed in variety of texture measures of MRI, is used for this classification. The classifier
has 2 phases, training and testing phases. In training phase the statistical and GLCM texture
features of MRI brain slices along with its label and normality or abnormality details, are
given as input to the classifier. In testing phase, if the feature vector of a new slice is given as
input to the classifier, a well-trained classifier can accurately classify it according to the
parameters formed in the training phase. In the level based approach, the brain slices are
grouped into four classes according to the similarity of anatomical structure in visual aspect
of the image and the above texture extraction, training and testing processes are done at each
level independently.

2.1    FEATURE EXTRACTION

        The purpose of feature extraction is to reduce the original data set by measuring
certain properties, or features, that distinguish one input pattern from another [18]. The
extracted features provide the characteristics of the input type to the classifier by considering
the description of the relevant properties of the image into a feature space. Most of the
tumour is heterogeneous tissues and the mean values of relaxation times are not at all
sufficient to characterize the heterogeneity of the different tumour types [3]. An alternative
approach, which is being investigated within the framework of this study, is to apply texture
analysis to the T2 FLAIR images to describe quantitatively the brightness and texture of the
images. Texture analysis covers a wide range of techniques based on first- and second order
image texture parameters. In the present study the statistical features based on image intensity
like mean & variance and features from gray level co-occurrence matrices (GLCMs) such as
entropy, contrast, energy, inverse difference moment and correlation [ 6,7 ,11] are used to
investigate the adequacy for the discrimination of normal and abnormal patient.
        The gray level co- occurrence matrix (GLCM) calculates how often a pixel with gray
level value occurs either horizontally, vertically, or diagonally to adjacent pixels with the
value j, where i and j are the gray level values in the image. Haralick features [6, 7] based on
GLCM is a proven technique to analyse the object with irregular outlines [6, 7]. Haralick
introduced fourteen textural features from the GLCM and out of these fourteen features five
of the textural features are considered to be the most relevant. Those textural features are
Energy, Entropy, Contrast, Correlation and Inverse Difference Moment. Energy is also called
Angular Second Moment (ASM) where it measures textural uniformity [19]. If an image is
completely homogeneous, its energy will be maximum. Entropy is a measure, which is
inversely correlated to energy. It measures the disorder or randomness of an image [19].
Next, contrast is a measure of local gray level variation of an image. This parameter takes
low value for a smooth image and high value for a coarse image. On the other hand, inverse
difference moment is a measure that takes a high value for a low contrast image. Thus, the
parameter is more sensitive to the presence of the GLCM elements, which are nearer to the
symmetry line x (i, i) [19]. The last feature, correlation, measures the linear dependency
among neighbouring pixels. It gives a measure of abrupt pixel transitions in the image [20].



                                                405
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

2.2.1 FEATURES USED

 Statistical Features

                                                Variance


 GLCM Features
 Entropy                                        Energy
                                                Correlation
 Contrast
                                                                          µ µ
                                                                    σ σ


 Inverse Difference Moment

        Where P(i, j) is the GLCM Matrix, R is the total number of pixel pairs used for the
calculation of GLCM and , ,        and      are the mean and standard deviation values of
GLCM values accumulated in the x and y directions respectively.

2.2     CLASSIFICATION
        The aim of classification is to group items that have similar feature values into
groups. Classifier achieves this by making a classification decision based on the value of the
linear combination of the features. SVM is a binary classification method that takes as input
labelled data from two classes and outputs a model file for classifying new unlabelled or
labelled data into one of two classes [1, 9,11].

2.3     SUPPORT VECTOR MACHINE
        Support Vector Machine (SVM) is a binary classifier based supervised learning
theory, a recent advances in statistical learning theory. SVMs deliver state-of-the-art
performance in real-world applications such as text categorisation, hand-written character
recognition, image classification, bio sequences analysis, etc. The basis of this approach is
the projection of the low-dimensional training data in a higher dimensional feature space,
because in this higher dimensional feature space it is easier to separate the input data. This
projection is achieved by using kernel functions. So kernel functions provides the bridge
between non-linear to linear. Thus kernel function is used to map the low dimensional data
into the high dimensional feature space where data points are linearly separable. There are
many types of kernels are available for SVM and this work uses the following kernels:
Linear, Polynomial and radial basis function (RBF) [1, 9, 11].

3.     RESULTS AND DISCUSSIONS

       In the proposed work, T2 FLAIR weighted axial MRI Brain images as input data set.
Here two types of databases are used
       1. Simulated Brain Database.
       2. Brain Database of a Hospital


                                             406
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

        The input data involved 100 patients (50 abnormal and 50 normal). At first the
normal- abnormal classification is done without considering the anatomical structure. In this
stage of the work a set of 320 brain slices, 160 normal and 160 abnormal, are used. Out of
these two hundred images, 160 slices, 80 normal and 80 abnormal, are used for training and
remaining hundred are used for testing.
        For level based normal- abnormal classification, the whole normal and abnormal
images are divided into 4 levels according to the similarity of the brain slices based on a
viewing aspect of the images. Thus each level contains a total of 160 images with 80 normal
and 80 abnormal. Out of these 160 images 80, 40 normal and 40 abnormal are used for
training phase and remaining 80 are used for testing phase. Results are summarised in Tables
3, 4 and 5.

                      Table 3: Classification using Polynomial Kernel

                       Level 1        Level 2       Level 3       Level 4      All Levels

TP                        40            39            39             40            78
FN                         0             1             1             0              2
TN                        40            40            39             40            78
FP                         0             0             1             0              2
Sensitivity (TPR)          1          0.975          0.975           1            0.975
             (FPR)         0          0.025          0.025           0            0.025
Specificity (TNR)          1             1           0.975           1            0.975
             (FNR)         0             0           0.025           0            0.025
         Accuracy          1          0.9875         0.975           1            0.975


                         Table 4: Classification using RBF Kernel

                       Level 1       Level 2        Level 3       Level 4     All Levels

TP                       40            39             39            40             73
FN                        0             1              1             0             7
TN                       40            39             38            40             72
FP                        0             1              2             0             8
Sensitivity (TPR)         1           0.975          0.975           1          0.9125
             (FPR)        0           0.025          0.025           0          0.0875
Specificity (TNR)         1           0.975          0.95            1            0.9
             (FNR)        0           0.025          0.05            0            0.1
         Accuracy         1           0.975         0.9625           1          0.90625




                                              407
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

                           Table 5: Classification using Linear Kernel

                        Level 1       Level 2         Level 3       Level 4      All Levels
TP                        40             40              39            39            72
FN                         0             0               1             1              8
TN                        40             38              39            40            73
FP                         0              2              1             0              7
Sensitivity (TPR)          1              1            0.975         0.975           0.9
             (FPR)         0              0            0.025         0.025           0.1
Specificity (TNR)          1            0.95           0.975           1           0.9125
             (FNR)         0            0.05           0.025           0           0.0875
         Accuracy          1           0.975           0.975        0.9875        0.90625

        The results shows that level based normal-abnormal classification got better result than
non-level based classification. Also it shows that SVM with Polynomial kernel got better result
than those with RBF and Linear kernels.

4.     CONCLUSION

        This work is intended to prove that, the consideration of anatomical structure of the MRI
Brain slices, for the normal/abnormal classification, will help to get more accurate result. Level
based normal abnormal classification got better results than non- level based classification. Here
support vector machine with polynomial kernel of degree 3 shows better results than those with
linear or RBF kernel.
        This work will surely help the radiologists and doctors in the identification of abnormal
brain slices. Magnetic Resonance Images are examined by radiologists based on visual
interpretation of the films to identify the presence of tumour abnormal tissue. The shortage of
radiologists and the large volume of MRI to be analysed make such readings labour intensive,
cost expensive and often inaccurate. The sensitivity of the human eye in interpreting large
numbers of images decreases with increasing number of cases, particularly when only a small
number of slices are affected. Hence this automated systems for analysis and classification of
such medical image will surely become an aid for both radiologists and doctors in tumour
analysis and detection. Also it will be the key step for the automated tumour detection system
development.

REFERENCES

[1]       Abdullah, N, Ngah, U.K.; Aziz, S.A., “Image classification of brain MRI using support
vector machine” Imaging Systems and Techniques (IST), 2011 IEEE International Conference on
17-18 May 2011.
[2]      T Kesavamurthy, S SubhaRani, ``Pattern Classification using imaging techniques for
Infarct and Hemorrhage Identification in the Human Brain"Calicut Medical Journal 2006.
[3]      http://www.braintumor.org/TumorTypes
[4]      http://www.bio-medicine.org/Biology
[5]       D. Selvathi, R.S. Ram Prakash, Dr.S.ThamaraiSelvi, “Performance Evaluation of Kernel
Based Techniques for Brain MRI Data Classification"International Conference on Computational
Intelligence and Multimedia Applications 2007.


                                                408
International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME

[6]     R. M Haarlick, “Statistical and structural approaches to texture", Proc. IEEE, vol. 67, pp.
786-804, 1979.
[7]     R. M. Haralick, K. Shanmugam, and I. Dinstein, “Textural Features for Image
Classification", IEEE transaction on systems, man, and cybernetics, Vol.3, No 6, pp. 610-621,
1973.
[8]      S. Chaplot, L.M. Patnaik, N.R. Jaganathan, “Classification of magnetic resonance brain
images using wavelet as input to support vector machine and neural network" Biomed. Signal
Process. Control (2006).
[9]      V.N. Vapnik, “Statistical Learning Theory", Wiley, New York, 1998.
[10] VerlagDr. Mueller, “Image De-noising Using Wavelet Transforms" April 2008.
[11] H. Selvaraj, S. ThamaraiSelvi, D. Selvathi, L. Gewali, “Brain MRI Slices Classification
Using Least Squares Support Vector Machine”, IC-MED, Vol. 1, No. 1, Issue 1, Page 21 of 33
[12] Ching-Tsorng Tsai, Hsian Min Chen, Jyh-Wen Chai, Chen, Chein-I Chang,
“Classification of Magnetic Resonance Brain Images by Using Weighted Radial Basis Function
kernels",IEEE International Conference of Electrical and Control Engineering (ICECE), 2011,
Page(s): 5784 - 5787.
[13] MadhubantiMaitra, AmitavaChatterjee and FumitoshiMatsuno, “A Novel Scheme for
Feature Extraction and Classification of Magnetic Resonance Brain Images Based on Slantlet
Transform and Support Vector Machine",IEEE SICE Annual Conference 2008, Page(s): 1130 -
1134.
[14]     N. HemaRajini, R.Bhavani, “ Classification of MRI Brain Images using K-Nearest
Neighbour and Artificial Neural Network",IEEE-International Conference on Recent Trends in
Information Technology, ICRTIT 2011, Page(s): 563-568.
[15] ShahlaNajafi, Mehdi ChehelAmirani and Zahra Sedgh, “A New Approach to MRI Brain
Images Classification",IEEE 19th Iranian Conference on Electrical Engineering (ICEE), 2011,
Page(s): 1.
[16]     Ahmed Kharrat, Karim Gasmi n, Mohamed Ben Messaoud, NacéraBenamrane and
Mohamed Abid, “Automated Classification of Magnetic Resonance Brain Images Using Wavelet
Genetic Algorithm and Support Vector Machine",IEEE International Conference on Cognitive
Informatics (ICCI 10), 2010, Page(s): 369-374 .
[17]     Abraham Varghese, RejiRajan Varghese, KannanBalakrishnan, J. S. Paul, “Axial T2
Weighted MR Brain Image Retrieval Using Moment Features",Springer- Advances in Intelligent
Systems and Computing Volume 177, 2013, pp 355-363.
[18]     Alberto Martin and SobriTosunoglu, “Image Processing Techniques for Machine
Vision", Florida University, 2000
[19] A. Baraldi, F. Parmiggiani, “An Investigation Of The Textural Characteristics Associated
With GLCM Matrix Statistical Parameters”,IEEE Trans. on Geos. and Rem. Sens., vol. 33(2), pp.
293-304, 1995.
[20] A. Ukovich, G. Impoco, G. Ramponi, “A tool based on the GLCM to measure the
performance of dynamic range reduction algorithms”,IEEE Int. Workshop on Imaging Sys. &
Techniques, pp. 36-41, 2005.
[21] Selvaraj.D and Dhanasekaran.R, “MRI Brain Tumour Detection by Histogram and
Segmentation by Modified Gvf Model”, International Journal of Electronics and Communication
Engineering & Technology (IJECET), Volume 4, Issue 1, 2013, pp. 55 - 68,
ISSN Print: 0976- 6464, ISSN Online: 0976 –6472.
[22] B.Venkateswara Reddy, Dr.P.Satish Kumar, Dr.P.Bhaskar Reddy and B.Naresh Kumar
Reddy, “Identifying Brain Tumour from MRI Image using Modified FCM and Support Vector
Machine”, International journal of Computer Engineering & Technology (IJCET), Volume 4,
Issue 1, 2013, pp. 244 - 262, ISSN Print: 0976 – 6367, ISSN Online: 0976 – 6375.

                                               409

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:1
posted:5/6/2013
language:
pages:7