neural network modelling for brain tumor detection

Neural Network Modelling For Brain Tumor Detection Using Texture Features

NEURAL NETWORK MODELLING FOR BRAIN TUMOR DETECTION USING TEXTURE FEATURES 1

BHAVIN TANDEL, 2SAURABH KATARIA

1,2

Bachelor of Engineering (E&T/C) from MIT E&T/C, Dept, Pune University, Pune, India

Abstract—An abnormal and uncontrolled cell division in the brain leads to brain tumor. Any brain tumor is inherently serious and life threatening. Its threat level depends on factors like type of tumor, its location, its size and its state of development. This paper provides an efficient technique for its detection which in turn gives accurate values of these factors. The basis of this technique is use of feature extraction; co-occurrence and gray level run length matrices, in combination with artificial neural networks (ANNs). The new model clearly distinguishes between the points of interest (tumor) and the rest of the data. In this model various features are extracted from the co-occurrence and gray level run length matrices. Following the feature extraction, ANN is employed to identify and learn correlated patterns between input data sets and target values. This paper presents a comparative result by using different learning algorithms, Support Vector Machines (SVM), Radial Basis Function (RBF) and the Back-Propagation Network. Keywords— co-occurrence matrix, gray level run length matrix, back-propagation, Radial Basis Function and Support Vector Machines

I.

types of algorithms are used for dividing the brain images into three categories (a) Pixel Based (b) Region or Texture Based (c) Structural Based. Image texture, defined as a function of the spatial variation in pixel intensities (gray values), is useful in a variety of applications and has been a subject of intense study by many researchers. One immediate application of image texture is the recognition of image regions using texture properties. Based on textural properties, we can identify a various regions of the brain through the MRI scans and identify tumor affected and unaffected areas. Texture is the most important visual cue in identifying these types of homogeneous regions. This is called texture classification. The goal of texture classification then is to produce a classification map of the input image where each uniform textured region is identified with the texture class it belongs to. Texture classification assigns a given texture to some texture classes. Two main classification methods are supervised and unsupervised classification. Supervised classification is provided examples of each texture class as a training set. A supervised classifier is trained using the set to learn a characterization for each texture class. Unsupervised classification does not require prior knowledge, which is able to automatically discover different classes from input textures. The majority of classification methods involve a twostage process. The first stage is feature extraction, which yields a characterization of each texture class in terms of feature measures. It is important to identify and select distinguishing features that are invariant to irrelevant transformation of the image, such as translation, rotation, and scaling. Such features can be obtained through some texture feature extraction methods like co-occurrence matrices or gray level run length matrices.

INTRODUCTION

The brain is a soft, delicate, non-replaceable and spongy mass of tissue. A tumor is a mass of tissue that grows out of control of the normal forces that regulates growth. Brain tumor is a group of abnormal cells that grows inside or around the brain. It can also indirectly damage healthy cells by crowding other parts of the brain and causing inflammation, brain swelling and pressure within the skull. [2] The structure and function of the brain can be studied noninvasively by doctors and researchers using Magnetic Resonance Imaging (MRI). [3]Magnetic Resonance Imaging (MRI) is the state-of-the-art medical imaging technology which allows crosssectional view of the body with unprecedented tissue contrast. MRI has rapidly evolved into an accepted modality for medical imaging of disease processes in the musculoskeletal system, especially the foot and brain. MRI provides a digital representation of tissue characteristic that can be obtained in any tissue plane. MRI has the added advantage of being able to produce images which slice through the brain in both horizontal and vertical planes. The aim of this work is to design an automated tool for brain tumor quantification using MRI image data sets by segmenting tumor in the brain. Segmentation is an important process to extract information from complex medical images. By segmenting tumor in the brain, surgeon will be able to see the tumor and then ease the treatment. The digital image processing community has developed several segmentation methods. Four of the most common methods are: 1) amplitude thresholding, 2) texture segmentation 3) template matching, and 4) regiongrowing segmentation. It is very important for detecting tumors, edema and necrotic tissues. These

Proceedings of IRF International Conference, 5th & 6th February 2014, Pune India. ISBN: 978-93-82702-56-6 148


The second stage is classification, in which classifiers are trained to determine the classification for each input texture based on obtained measures of selected features. In this case, a classifier is a function which takes the selected features as inputs and texture classes as outputs.

Texture classification can sort image data into more readily interpretable information, which is used in a wide range of applications such as industrial inspection, image retrieval, medical imaging and remote sensing. Texture segmentation partitions an image into a set of disjoint regions based on texture properties, so that each region is homogeneous with respect to certain texture characteristics. Segmentation of texture involves extracting features and deriving metrics to segregate textures.

An artificial neural network (ANN) is a mathematical model or computational model that is inspired by the structure and/or functional aspects of biological neural networks. It consists of an interconnected group of artificial neurons and processes information using a connectionist approach to computation. In most cases an ANN is an adaptive system that changes its structure based on external or internal information that flows through the network during the learning phase. They are usually used to model complex relationships between inputs and outputs or to find patterns in data. The proper ANN is obtained by taking into consideration the requirements of the specific application, as each ANN topology does not yield satisfactory results in all practical cases. II.

III.

CO-OCCURRENCE MATRICES

[4]Spatial gray level co-occurrence estimates image properties related to second-order statistics. Cooccurrence matrices give relative distance among the pixels and their relative orientation. The 'value' of the image is originally referred to the grayscale value of the specified pixel. A 32-bit color will yield a 232 x 232 co-occurrence matrix. Their main applicability has been in the measuring of texture in images, so the typical definition assumes that the matrix is in fact an image. The co-occurrence matrix for a pair (d, φ)where d is the relative distance between pixels and φ is the orientation; is defined as the Ng x Ng (no. of pixel values) matrix where η(I1,I2) is the number of pixel pairs, at relative position (d, φ ), which have graylevel values (I1 , I2), respectively. R is the total number of possible pixel pairs.

TEXTURE ANALYSIS

Texture analysis is an important and useful area of study in machine vision. Most natural surfaces exhibit texture and a successful vision system must be able to deal with the textured world surrounding it. Major goals of texture research in computer vision are to understand, model and process texture, and ultimately to simulate human visual learning process using computer technologies. Texture analysis might be applied to various stages of the process. At the preprocessing stage, images could be segmented into contiguous regions based on texture properties of each region. At the feature extraction and the classification stages, texture features could provide cues for classifying patterns or identifying objects.

(0,0) (0,1) (0,2) (0,3) (1,0) (1,1) (1,2) (1,3) = 1/ (2,0) (2,1) (2,2) (2,3) (3,0) (3,1) (3,2) (3,3) There are four orientations used to construct cooccurrence matrices i.e. 00, 450, 900 and 1350, making the features invariant of rotation. For example, 0 = 1 3 3

input Image Acquisition

4 1/24 1 1 0

Preprocessing

0 1 2 2

1 2 0 0

2 0 3 2

1 0 6 3

2 0 3 2

( = 1) = 0 0 3 2

Features derived from co-occurrence matrices are:

Feature Extraction

1.

Inverse difference moment

2.

Contrast

Post processing

3. 4.

Entropy Angular second moment

decision

5.

Correlation

6.

Variance

Classification

Fig.1 Texture Analysis process



7.

Sum/Difference average

8.

Sum/Difference variance

9.

Sum/Difference entropy

Back-propagation network is a multilayered network. It contains an input layer, hidden layer and an output layer. The back propagation algorithm selects a training example, makes a forward (feed forward) and a backward pass (back propagation) & updates weights, and then repeats until algorithm converges satisfying a pre-specified mean squared error value. It is required that the activation function used by the artificial neurons (or "nodes") is differentiable.

10. Information measure I & II IV.

GRAY LEVEL RUN LENGTH MATRICES

VI.

[1] A gray level run is a set of consecutive pixels having the same gray level value. The length of the run is the number of pixels in the run. Run length features encode textural information related to the number of times each gray level a pixel value appears in the image by itself, the number of times it appears in pairs, and so on. For each of the four directions (0°, 45°, 90°, 135°) we define the corresponding run length matrix QRL. For extracting appropriate features of an image, the gray level run length is taken as matrix with the column number representing the value and the row number representing the length of the run corresponding to the value. Like the cooccurrence matrix, gray level run length is also rotation sensitive. For example, 0 2 0 0 0 2 0 0 1 0 0 (0 ) = 0 1 0 = 1 1 1 0 1 1 1 2 1 0 0 2 1 0 Features derived from gray level run length are

V.

1. 2. 3.

Short Run Emphasis (SRE) Long Run Emphasis (LRE) Gray level non-uniformity

4. 5.

Run length non-uniformity Run percentage

RADIAL BASIS FUNCTION

[6]A Radial Basis Function (R.B.F.) is a real valued function whose value depends only on the distance from the origin or from a predefined point c, called as a center. Any function φ that satisfies the property φ(x) = φ(||x||) is a radial function. The norm is usually Euclidean distance. Distance between xi and xj is given by:

=

(

−

)

RBFs are typically used to build up function approximations of the form where the approximating function y(x) is represented as a sum of N radial basis functions, each associated with different ci, and weighted by an appropriate coefficient wi. In an RBF network, there are three types of parameters that are needed to be chosen to adapt the network for a particular task: the center vectors ci, the output weights wi, and the RBF width parameters σi. In the sequential training, weights are updated at each time step as data streams in.

0 0 0 0

VII.

SUPPORT VECTOR MACHINES

[5]SVMs are the most well-known learning systems based on kernel methods. It is as an alternative to neural networks, and that has been successfully employed to solve clustering problems, especially in biological applications. It performs classification by constructing an N dimensional hyper plane that optimally separates the data into two categories. A classification task usually involves training and testing data which consist of some data instances. Each instance in the training set contains one “target value” (class labels) and several “attributes” (features). The goal of SVM is to produce a model which predicts target value of data instances in the testing set which are given only the attributes.

BACK-PROPAGATION NETWORK

[6] Back-propagation, or propagation of error, is a common method of teaching artificial neural networks how to perform a given task. It is a supervised learning method, and is an implementation of the Delta rule. It requires a teacher that knows, or can calculate, the desired output for any given input. It is most useful for feedforward networks (networks that have no feedback, or simply, that have no connections that loop).

VIII. 1.

ALGORITHMS AND FLOWCHART

CO-OCCURRENCE MATRIX 1) Step 1 : Select distance angle 0o ,45o ,90o ,135o 2) Step 2: FOR all different pairs of pixel

Fig.2 Back-propagation network architecture



2.

values Compare all consecutive pairs of matrix elements in specified distance angle, with specified pair of pixel values from matrix element (0,0) 3) Step 3: IF there is a match increment counter for corresponding pair of pixel values ELSE continue comparing 4) Step 4: Repeat steps 2-4 starting from last matrix element 5) Step 5: Calculate corresponding features GRAY LEVEL RUN LENGTH MATRIX

4.

Acquire image

Image segmentation

Gray level Run Length features

Step 1: Select distance angle Step 2: Starting from matrix element (0, 0), calculate run of each gray level pixel value. 3) Step 3: Store pixel value as row no. and run as column no. of output matrix. Store no. of same length runs in the corresponding element of output matrix. 4) Step 4: Calculate corresponding features BACK-PROPAGATION NETWORK

Normalized feature vectors Training algorithm

If output =1

1) Step 1 : Create a feed-forward network with ni inputs, nh hidden units and no output units. 2) Step 2 : Initialize all the weights to small random values (e.g. between -0.5 to 0.5) 3) Step 3 : Until termination condition is met, Do For each training example ( ⃗, ⃗),

2.

Tumor detected

IX.

δh=oh(1–oh) ∑ 4.

hidden

∈

( )

unit

kh

h,

δk

Update each network weight wji as follows : ji ← ji +

∆

COMPARISON PARAMETERS

[5] Consider a scenario where people are tested for a disease. The test outcomes can be positive (sick) or negative (healthy), while the actual health status of the persons may be different. In that setting:  True positive: Sick people correctly diagnosed as sick.  False positive: Healthy people wrongly identified as sick.  True negative: Healthy people correctly identified as healthy.  False negative: Sick people wrongly identified as healthy. Sensitivity: It is defined as the probability that we encounter an abnormality in the medical image and it is classified correctly.

Input the instance ⃗ and compute the output ou of every unit. For each output unit k, calculate

For each calculate

No Tumor detected

Fig. 3 Process Flowchart

δk = ok (1 - ok) (tk - ok) 3.

Co-occurrence features

Feature vectors

Each training example ( ⃗, ⃗) is of the form where ⃗ is the input vector and ⃗ is the target vector. n is the learning rate (e.g. 0.5), ni, nh and no are number of input ,hidden and output nodes respectively. Input from unit i to unit j is denoted as xij and its weight is denoted as wji.

1.

η δk δki

FLOWCHART

1) 2)

3.

ji =

where ∆

Sensitivity = TP/(TP+FN) * 100 Specificity: It is defined as the probability that there

ji



is no abnormality in the medical image and the results support it.

CONCLUSION As seen from the result table, detection using both Co-occurrence and gray level run length matrices is far more superior than using just one of them. It can also be seen that SVM provides better results for brain tumor detection that RBF.

Specificity = TN/(TN+FP) * 100 X.

RESULT

Parameters Cooccurrence with RBF

Cooccurrence with SVM

Cooccurrence and Gray level Run length with BackPropagation

True Positives True Negatives False Positives False Negatives Sensitivity

1363

1970

1870

102444

102048

110067

19

18

7

19

18

6

98.63%

99.09%

99.68%

Specificity

99.98%

99.98%

99.99%

REFERENCES [1] Dong-hui xu1, Arati S.Kurani2, Jacob D. Furst3, Daniela S. Raicu4, Run Length Encoding for Volumetric Texture, Intelligent Multimedia Processing Laboratory, School of Computer Science, Telecommunication, and Information Systems, DePaul University, Chicago, Illinois, 60604 [2]T.Logeswari and M.Karnan, ‘An Improved Implementation of Brain Tumor Detection Using Segmentation Based on Hierarchical Self Organizing Map’ ,International Journal of Computer Theory and Engineering,Vol.2 2 [3]1S. Murugavalli and V.Rajamani,‟An Improved Implementation of Brain Tumor Detection Using Segmentation Based on Neuro Fuzzy Technique’1Department of Computer Science and 2 Engineering Department of Electronics and Communication Engineering, PSNA College of Engineering and Technology, Dindigul, Tamil Nadu, India [4]Sergios Theodoridis and Konstantinos Koutroumbas, ‘Pattern Recognition Second Edition’ [5]Satish Chandra1, Rajesh Bhat2, Harinder Singh3, D.S.Chauhan4, ‘ Detection of Brain Tumors from MRI using Gaussian RBF kernel based Support Vector Machine’1Department of CSE & IT, Jaypee University of IT, Solan,HP, India , 2Department of Computer Science, Indian Institute of Technology, New Delhi, India, 3 Department of Mathematics, Jaypee University of IT, Solan, HP, India, 4Department of Electronics, Uttarakhand Technical University , Dehradun, India [6] Mohan, Ranka, “Elements of Artificial Neural Networks”, Penram International

Table 1 Result table

 Fig. 4 Brain Tumor Detection Result
