Improved Brain Tumor Detection

Abstract

brain tumor can be detected by using computer based image processing algorithm. MRI scan has been done to find the brain tumor. MRI images are not enough to thoroughly diagnose the tumor. Fuzzy c mean algorithm is very popular image segmentation. Fuzzy c mean algorithm output also contains some unwanted part. In our proposed work, these unwanted parts can be removed by using median filter. In the proposed work, DWT with SVM are used to identify the types of tumor, whether it is Benign and Malignant type. Filtered image by median filter also helps in better detection by SVM classifier.

Keywords

FCM (Fuzzy C Mean), MRI (Magnetic Resonance Imaging), DWT (Discrete Wavelet Transform), SVM (Support Vector Machines), image segmentation, Grayscale image, MRI (Magnetic resonance imaging), computerized tomography (CT) scan, image pre-processing, image filtering.

Introduction

Brain tumor can be detected by various brain scanning techniques. CT scan provides the detailed picture of brain and MRI test where the computer is linked with a strong magnetic field which provides the clear 2D picture of brain. MRI (Magnetic Resonance Imaging) discards radiation unlike CT scan [2,4]. The MRI image gives the complete view of the brain and a proper inspection has to be done by an expert to find the tumor which makes the process slower and costlier. To solve this, computer based segmentation algorithms has been created. These algorithms provide the tumor as the output image. The most popularly used segmentation is FCM (Fuzzy C Mean) segmentation algorithm. FCM algorithm provides accurate results for data set which is overlapped and it is much efficient than k-means algorithm [1].

Brain tumors can be classified as Benign and Malignant. A benign tumor is one that does not grow abruptly. It never affects its neighboring tissues and not at all expands to other parts. Malignant tumor is one which worsens with the passing time and ultimately proves to be fatal. We can say malignancy is a tumor in descriptive or advance stage from where it is quite impossible to return back [4]. To extract features out of MRI brain image, Wavelet transform is effective since it allows image analysis at different levels of motion suitable to its multi-resolution diagnostic property [1]. In order to differentiate type of brain tumor, an SVM (Support Vector Machines) classifier is popularly used. SVM model represents points in space which are mapped so that the examples of separate categories are divided by clear gap that is as broad as possible.

In our proposed work, FCM algorithm is used for segmentation of the brain MRI image. The segmented image is further enhanced by using median filter. Here median filter removes the unwanted segmented part by considering them as a noise. The segmentation output is then fed to DWT and SVM classifier to correctly identify the type of tumor.

Ii. Fcm (Fuzzy C Mean) Segmentation

Fuzzy c-mean can be called as a sub optimum segmentation method that surrenders global optimality for enhanced statistical performance and adaptability from the segmentation process. Computational valuation on FCM is determined by the amount of image points that need to be highly processed every iteration [5].

FCM is a technique of clustering which allow one piece of information which belongs to two or more clusters [6]. The main aspect of this algorithm works by assigning membership values to each data point consequent to each cluster center on the basis of distances between the cluster and the data point, Higher the membership value then more the data near to the cluster center. Clearly, summation of membership of each data point should be equal to one [10].

FCM algorithm is a method of iterative clustering that produces an optimal c partition by minimizing weight within group sum of the squared error objective function (JFCM) [8].

(1)

Where,

X = {x1, x2 , ..., xn} ≤ R,

n = number of data items,

c = number of clusters with 2 ≤ c < n,

uik = degree of membership of xk in the ith cluster,

q = weighting exponent on each fuzzy membership,

vi = prototype of the centre of cluster i,

d2(xk,vi) is a distance measure between object xk and cluster centre vi.

A solution of object function (JFCM) can be calculated by a iterative process, which is as follows:

• • First set the values for q, c, & e,
  • Second, fuzzy partition matrix need to be initialized,
  • Third, need to set the loop counter such that b = 0,
  • Calculate c cluster centers {vi(b)} with U(b)

(2)

• Calculate the membership U(b+1), For k = 1 to n, calculate the following:

Ik={i|1<=i<=c

dik=||xk-vi||=0},

~Ik={1,2,……c}-Ik, for the kth column of the matrix, compute new membership values, and if Ik=Ø , then

(3)

else uik(b+1) = 0 for all iє~Ik and ƩiєIk uik(b+1) =1, next k [9],

if ||Ub-U(b+1)|| < Ɛ , stop; otherwise set b=b+1 and go to step 4.

For the medical images segmentation, suitable clustering type is fuzzy based clustering. Fuzzy c-means (FCM) can be considered as the fuzzified version of the k-means algorithm. It is a kind of clustering algorithm which enables data item to have a degree of belonging to each and every cluster by degree of membership [6].

Iii. Dwt (Discrete Wavelet Transform)

The wavelet gives idea of different frequencies of an image using different scales. DWT provides wavelet coefficient out of brain MR images. Two dimensional DWT gives four sub-bands, that are LL(low–low), HL(high–low),LH(low– high), HH(high–high) with the two-level wavelet decomposition of Region of Interest (ROI). The wavelets approximations at first and second level are represented by LL1, LL2, respectively; which is representing the low-frequency part. The high-frequency part of the images are represented by LH1, HL1, HH1, LH2, HL2 and HH2 which gives the details of horizontal, vertical and diagonal directions at first and second level, respectively as shown in the fig. 1 below [2].

Iv. Median Filter

Median filter is very popular in image filtering. It behaves like low pass filter which blocks all high frequency component of the images like noise and edges, thus blurs the image [11]. For the filtering of high density corrupted image need large window size so that the sufficient number of noise free pixels will present in the window. So the size of the sliding window in the median filter is varying according to the noise density. The window size 3×3, 5×5, 7×7, and 9×9 median filter are mainly applicable. Output of the median filter is given by

y(i,j)=median{x(i-s,j-t),x(i,j)/(s,t)∈W,(s,t)≠(0,0)} (4)

where {x} is the noisy image and y(i,j) is the recovered image with preserve edges.

V. Svm (Support Vector Machine) Classifier

SVM classifier is applied in our work to determine the type of tumor, whether it is benign and malign tumors. It is very effective learning method used in classification problems. SVM uses kernel functions in separating classes with large data. SVM provides better results in applications with less data with bigger dimensionality [7]. SVM is a popular discriminative classifier which is formally defined by a separating hyperplane. It can also be defined as a given labeled training data that is supervised learning, this algorithm outputs an optimal hyperplane which discriminate new examples. In the 2D space, this hyperplane is a line which is dividing a plane in two different parts where each class has taken space in either side.

VI. Proposed Algorithm

The flowchart of proposed algorithm is shown in fig.3. the process starts with reading the image into MATLAB. After that FCM algorithm is applied for segmentation of the image. The segmentation output still contains some unwanted part as a noise therefore median filter is applied to remove them. Then DWT followed by SVM classifier is applied to identify the type of brain tumor.

Vii. Simulation Results

Table 1: Simulation results of previous work and proposed work with classifier output

S.No. Original image Previous work Proposed work Classifier output

Viii. Conclusion

Brain tumor can be detected and classified by using image processing algorithms. FCM is very effective algorithm for segmentation of image. But still the output of FCM contains unwanted parts therefore median filter is introduced in our work to filter out unwanted part. Then in our proposed DWT and SVM is used to identify the type of brain tumor. The segmentation output of proposed work is better than previous work as shown in results. Proposed algorithm is better in terms of both quality as well helps in providing better segmented image to classifier for better classification.