Materials and Methods 1 Data Set Description

Different type of Tumor patient dataset was collected by a SIEMENS SOMATOM EMOTION SPIRAL CT scanner located at Multi Speciality Hospital, Coimbatore. Besides a normal scan performed at a routine clinical dosage (130 mA), an additional scan from the same patient was acquired at a much lower tube current, i.e. 20 mA.

V. V. Gomathi, S. Karthikeyan

The 3D image data consisted of DICOM (Digital Imaging and Communications in Medicine) consecutive slices, each slice being of size 512 by 512 and having 16- bit grey level resolution. Each of the organs of interest in this research was manually contoured by the expert for the comparison of auto segmented output with manual contoured image.

III.2 Methodology

This paper proposes a new Hybrid segmentation algorithm based on Medoid shift, K-Means and Signature quadratic form distance segmentation method for computer tomography images. Medoid shift algorithm is also a nonparametric clustering approach. It is a mode seeking method that computes shifts towards areas of greater data density using local weighted medoids. The use of medoids to discover structure in data is natural since, locally, the medoid can be considered a good representative of its neighborhood. Unlike means, medoids do not need an explicit feature space and require only a valid distance measure [17]. The medoid shift algorithms also automatically calculate the number of clusters during execution like mean shift [18].

The most popular method for image segmentation is k-means clustering [19] [20]. The procedure follows a simple and easy way to classify a given data set through a certain number of clusters fixed a priori. The clustering results of the K-means algorithm greatly depend on its initialization. The number of clusters must be known in advance.

The distance measure plays an important role in acquiring exact clusters. It is used to discover the similarity and dissimilarity between the pair of objects in the clustering techniques. Clustering techniques are based on measuring similarity and dissimilarity between data objects by calculating the distance between each pair. The choice of distance measure between clusters has a large effect on the shape of the resulting clusters [21].

Signature Quadratic form distance is a generalization of the Quadratic form distance. It (SQFD) [22] is an adaptive distance-based similarity measure. Signature Quadratic Form Distance measure which allows efficient similarity computations based on flexible feature representations. This approach bridges the gap between the well-known concept of Quadratic Form Distances and feature signatures. The Signature Quadratic Form Distance (SQFD) is a recently introduced distance measure for content-based similarity. It makes use of feature signatures, a flexible way to summarize the features of a multimedia object. The SQFD is a way to measure the similarity between two objects [21].

An efficient Hybrid Segmentation algorithm is proposed to avoid the pitfalls present in the HMSK algorithm and SQFD segmentation algorithm. One of common drawback was over fragmentation produced by the above said two algorithms. Over fragmentation produces many connected components.

Classifying many connected components is a tedious process. This will also cause wrong segmentation result and also leads to wrong decision making by the radiologist.

The above said algorithm is also not well suited to separate the joined organs. Some organs are joined together. For example heart and liver is joined together and also heart and spleen is joined together. In these circumstances, the proposed Hybrid segmentation algorithm is well suited for segmenting the joined organs efficiently. The Proposed Hybrid Segmentation is as follows.

Hybrid Segmentation Algorithm

Step 1: Consider the Single Dicom image or slices of Dicom images

Step 2: Apply the ECFT (Enhanced Curvelet Filter Technique) algorithm to get a noiseless image

Step 3: Obtain the Histogram of the input image Step 4: Initialize the control parameter

Step 5: Find the gray level cluster values based on an initialized control parameter

Step 6: Find no of pixel values present between each range of all gray level cluster values.

Step 7: Cluster the pixels which lies between the ranges to the respective gray level cluster value

Step 8: Each cluster are considered as data points Step 9: Find the distance between each cluster to all the data points

Step 10: Make the data point allocation by using













Step 11: Find the new cluster center by using







 

i _j i

/ C x





_

Step 12: Obtain the Clustered Image Step 13: Initialize the cluster step value

Step 14: Generate the cluster centers based on the cluster step value.

Step 15: Calculate the similarity matrix A using cluster centers P and input image pixel values Q.

Step 16: Compute the Signature Quadratic Form Distance (SQFD) value using the following formula

SQFDA (Q, P) =

_

_

_

_

T

Q | P * A* Q | P 

where

A -Similarity Matrix

P - Intensity Vector Pixel Values Q - Input Image Pixel

T – Transpose Matrix

Step 17: Find the minimum distance value

Step 18: Find the cluster center value based on minimum distance value and assign that cluster center value to the respective pixel position in the image

Step 19: Repeat the step 15, 16, 17 and 18 until convergence is attained (i.e. no pixels change clusters).

V. V. Gomathi, S. Karthikeyan

Copyright © 2014 Praise Worthy Prize S.r.l. - All rights reserved International Review on Computers and Software, Vol. 9, N. 9 In the proposed hybrid segmentation algorithm

consists of medoidshift with K-means and signature quadratic form distance method. Initially an ECFT (Enhanced Curvelet Filtering Technique) has been applied for removal of noise in the CT images.

These noiseless images are the input images. Histogram is found for the input images. Based upon the histogram of the input image, the control parameter is initialized. The random cluster values have taken based on the control parameter. The Closest cluster value is found based on the occurrences of cluster values.

Here each cluster is considered as data points. The distance between data points and cluster points (Closest cluster) has been calculated and found the new cluster.

Finally the similar cluster image has been obtained. The initial cluster step value has chosen either by manually or randomly. Then find the cluster centers based on the initialized cluster step value. The number of clusters in the image is equal to the number of cluster centers. Then the distance measure has been calculated between every pixel and the cluster centers. Signature Quadratic form distance is used to find the distance. For this process the two vectors P and Q is formed. By using P and Q the similarity matrix is generated.

Then the SQFD similarity between the P and Q is identified. The position with minimum SQFD value is identified. The cluster center consists of the cluster value.

The minimum value in the cluster center position is replaced with the original pixel value in the same position. Repeat the same process till the convergence attained (ie there is no change in the pixel value of an image). Finally we obtained the segmented image.