Paper Title (use style: paper title)

3 downloads 0 Views 1MB Size Report
Now a day, brain tumor segmentation for MRI is difficult task for ..... saves the diagnosis time of the medical experts. REFERENCES. [1] R.C. Gonzalez and R.E. ...
International Journal of Applied Engineering Research, ISSN 0973-4562 Vol. 10 No.76 (2015) © Research India Publications; http://www.ripublication.com/ijaer.htm

Brain Tumor Segmentation of MRI Brain Images through FCM clustering and Seeded Region Growing Technique Kalaiselvi T

Nagaraja P

Image Processing Lab, Department of Computer Science and Applications, Gandhigram Rural Institute – Deemed University, Gandhigram, Dindigul, Tamil Nadu, India [email protected]

Image Processing Lab, Department of Computer Science and Applications, Gandhigram Rural Institute – Deemed University, Gandhigram, Dindigul, Tamil Nadu, India [email protected] detection of any abnormal changes in tissues such as computed tomography (CT), magnetic resonance image (MRI), functional-MRI (fMRI), positron emission tomography (PET) and single photon emission computed tomography (SPECT). Radiologists use these images for the visualization of the internal structure of the body [4].

Abstract—This proposed work is aimed to develop a semi automatic method to detect the brain tumor from T2-weighted MRI brain images using the fuzzy c-means (FCM) clustering and seeded region growing techniques. Initially applied FCM clustering is used to separate image into four regions: background, gray matter, white matter, cerebrospinal fluid (CSF). In CSF portion tumor region appears good and produces mask of brain tissue portion from CSF region. Then applied seeded region growing method is used to detecting the tumor region from the segmented portion. The quantitative measures of results were compared with using the metrics are predictive accuracy (PA) and dice coefficient (DC). The PA and DC values of the proposed method attained maximum value while compared to conventional FCM clustering technique. This proposed method is more efficient and faster for detecting the tumor region from T2-weighted MRI brain images.

In recent years, many approaches have been developed to the brain tumor detection, segmentation and analysis. A.Z.Arifin and A.Asano proposed an image thresholding method by using clusters organization from the histogram of the image. The similarity measure is based on inter-class variance of clusters to merge and the intra class variance of the newly merged cluster [5]. Emblem K.E. et al. proposed knowledge-based fuzzy c-means (FCM) clustering on multiple classes of MR image for glioma detection [6]. Wafa M. and Zagrouba E. proposed multi-featured FCM and evidence theory on multimodal MRI for brain tumor segmentation [7]. Fletcher-Heath L.M. et al. proposed two-stage FCM combined with KB procedure on non-contrasted MR images for tumor segmentation [8].

Keywords—Clusterin; Fuzzy c-means; Segmentation; Tumor; Seeded Region Growing;

I. INTRODUCTION Image segmentation is partitioning of an image into a set of disjoint regions that are visually different, homogeneous and meaningful with respect to some characteristics or computed property such as grey level, texture or colour to enable easy image analysis. It is most important study of image analysis, which is used to obtain the essential information from the images. It plays an important role in medical image analysis [1] [2]. Now a day, brain tumor segmentation for MRI is difficult task for clinical applications. In recent years, magnetic resonance imaging (MRI) has become an important modality for neurological image diagnosis. MRI brain images are oriented to dissimilar from two goals: classifying tissues, anatomical structures. The MRI brain image comprised of different tissue classes contains four regions, namely gray matter (GM), white matter (WM), cerebrospinal fluid (CSF) and background [3].

Anam Mustaqeem et al. proposed an efficient brain tumor detection algorithm by using watershed segmentation, thresholding and morphological operators. This method provided satisfied results in MRI brain images and thus locating the tumor in images [9]. Sayana Sivanand proposed a method for image segmentation by using the adaptive local thresholding and kernel FCM clustering. This method overcomes the problem of conventional FCM by using kernel FCM, which is used for images having unequal sized clusters [10]. The commonly used segmentation techniques can be classified into two categories: (1) region-based techniques that look for the regions satisfying a given homogeneity criteria and (2) edge-based segmentation techniques that look for edges between regions with different characteristics. For the region-based segmentation category, adaptive thresholding, clustering, region growing, watershed and split and merge are the well known methods for segmentation. Clustering and region growing techniques are the most popular techniques for

A brain tumor is any mass that results from abnormal and uncontrolled cells growing in the brain. There are three common types of tumors, namely benign (non-cancer), premalignant (pre cancerous stage) and malignant tumor (cancer). Medical imaging techniques can be performed for the

427

International Journal of Applied Engineering Research, ISSN 0973-4562 Vol. 10 No.76 (2015) © Research India Publications; http://www.ripublication.com/ijaer.htm

d-dimensional measure data, cj is the d-dimensional enter of the cluster and dij is the Euclidean distance between ith data point (xi ) and jth centroids (cj).

segmentation of medical images due to its simplicity and good performance. Seeded region growing technique is group’s pixels or regions that have similar properties based on predefined criteria. It starts with a set of initial seed points that represent the criteria, and grow the region. Seeds can be automatically or manually selected [11].

The updated membership functions are defined as follows, uij 

Edema is surround the tumor region and infiltrating mostly white matter or gray matter is based image modalities was most often not considered as important for tumor segmentation [12] [13]. This present paper proposed a combination of fuzzy c-means (FCM) and seeded region growing (SRG) methods for fast extraction of the tumor region from T2-weighted images. It relies on the information provided in the T2-weighted image channel for identifying tumor portion in proposed method. The proposed work is a semi automatic method detects the presence of brain tumor in T2-weighted images. Initially applied FCM clustering is used to separate image into four regions: background, gray matter (GM), white matter (WM), cerebrospinal fluid (CSF). In CSF portion tumor region appears good and produces mask of brain tissue portion from CSF region. Finally, SRG is used to detecting the tumor region from the segmented portion.

(2)

n

cj 

u i 1 n

m ij

u i 1

xi

m ij

(3)\

This condition will stop if the improvement of the objective function over the previous iteration is below critical value, ε∈ [1, 0]. This algorithm is iteratively updating the centers and membership grades for each data point. FCM iteratively moves the cluster centers to the right location within a data set. The detailed FCM algorithm proposed by Bezdek [16] is: Step 1: Randomly initialize the membership matrix U=[Uij], U(0) that has a constraint equation given by,

II. METHOD

c

u

The proposed method focused on segmenting the tumor portion from the high and low grade T2-weighted glioma MRI brain images. In the CSF region, tumor area appeared in good and as the largest connected region. The framework of the proposed work is given in Figure 1. Initially applied FCM clustering is used to divide image into four regions. In CSF portion generates the mask corresponding to major portions of brain by removing the most of the non-brain details like background, bone, fat and muscles. This mask is used to extract the brain substructure for further analysis. Then SRG technique is used to separate brain tumors from the extracted substructures. SRG section, user can select the seed points as manually. The following sections are explaining the implementation details of FCM clustering and SRG process.

i 1

ij

 1, j  1,2,...n,

(4) Step 2: At k-step, calculate the centroids and objective function using equations (3) and (1) respectively, Step 3: Update the membership U(k), U(k+1) by using equation(2). Step 4: If stopping criteria is reaches stop; otherwise return to Step 2. B. Seeded region growing process Seeded region growing is a procedure that groups pixels or sub regions into larger regions based on predefined criteria. Seeded region growing requires seeds as additional input. The basic approach is to start with a set of seed points and grow the regions by appending to each seed’s neighboring pixels that have similar properties to the seed.

A. Fuzzy c-means clustering This algorithm divides the image space into smaller regions or units called clusters and by definition such regions are to be disjoined [14] [15]. It is based on fuzzy partioning is that makes the data point belongs to all groups with different membership grades between 0 and 1. The aim of FCM clustering is to find the cluster centers that minimize dissimilarity (objective) function.

The seeded region growing algorithm applied in this study is summarized as follows: Input : CSF region Input image Output : The segmented Tumor image

The objective function is,

1.

Read the Input Image I(i,j)

2.

Select the seed point (User) i.

c

J m   uijm d ij i 1 j 1

2

 d ij  m1    k 1  d ik  c

This paper is organized as follows. The proposed method is explained in section 2, the results and discussion are given in section 3 and conclusion is given in section 4.

n

1

(1)

where m∈[1,∞] is a weighting exponent, uij∈[1,0] is the degree of membership xi in the cluster j, xi is the ith element of

3.

T=graythresh(I);

4.

For entire image I (i,j) do i.

428

seed=I(x1,y1);

if Y(i,j)==1

International Journal of Applied Engineering Research, ISSN 0973-4562 Vol. 10 No.76 (2015) © Research India Publications; http://www.ripublication.com/ijaer.htm

MRI images

FCM Clustering

Background

Gray Matter

White Matter

Cerebrospinal Fluid (CSF)

CSF Region

Seeded Region Growing

Users were selecting the seed point, and then neighboring pixels will join together.

Abnormal Detection

Figure 1. Framework of proposed method

429

International Journal of Applied Engineering Research, ISSN 0973-4562 Vol. 10 No.76 (2015) © Research India Publications; http://www.ripublication.com/ijaer.htm

Figure 2. Sample MRI brain images are in column1, the corresponding ground truth images are in column 2, the results of conventional FCM proposed method are in column 3 and the results of proposed method are in column 4.

430

International Journal of Applied Engineering Research, ISSN 0973-4562 Vol. 10 No.76 (2015) © Research India Publications; http://www.ripublication.com/ijaer.htm

ii.

if (i-1)>0 & (i+1)0 & (j+1)