Utilizing k-means clustering to extract bone tumor in CT scan and MRI images

Segmentation is one of the most significant parts of medical image processing. In image segmentation, the digital image is part of multiple sets of pixels. Magnetic Resonance Imaging, MRI and CT scanning is very important imaging techniques to explore the inner physiological constructions of the body noninvasively. A bone tumor is one of more life-threatening diseases, so exact detaching of the tumor regions is a pressing need. In this work, the K-means algorithm is employed on six MRI and CT scan images with different numbers of clusters. As well as many morphological operations like opening and dilation were applied after extract the fine tumor areas effectively. The results and the calculated surface areas of the separated tumor regions were compared to the radiologist delineation and the percent relative differences were found ranged from (0.63-1.75) % for MRI images and (0.34-1.51) % for CT scan images. This result indicates the high-quality performance of the adopted segmentation clustering-based method.

K-Means is from the many of the difficult clustering methods, used clustering algorithm to partition data into a determined number of clusters. k-Means classify the pixels into clusters being based on their intensity values. Every cluster is usually characterized by constant intensity. k-means clustering is often fitting for medical image segmentation for the reason that a variety of clusters (k) is typically regarded for photos of precise areas of human anatomy. in biomedical programs, the spatially various depth change of a biomedical structure is typically caused by inhomogeneity inside the method of image acquisition, including the inhomogeneous distribution of the comparison agent in CT imaging or inhomogeneous distribution of the magnetic discipline gradient in MRI imaging. a preliminary segmentation is obtained through the simple k-mean algorithm, the ordinary probability function is maximized on a point-point foundation, with the mean and the variance of every cluster is up to date after every iteration [10]. When the changes in the cluster seeds from one level to the next are close to zero or smaller than a pre-specified value algorithm stops. Each object is assigned to just one cluster [11]. K-means means algorithm aims to reduce the target function that represents the function of squared error which obtained by [12]. Where: 'c' is the number of cluster centers while 'ci' is the number of data points in the ith cluster and '||xi -vj||' is the Euclidean distance among xi and Vj. Many researcher were adopted this algorithm to extract abnormal regions in the medical images for difference body organs like [13][14][15][16][17][18][19][20][21].

3.
Morphological Operations Morphological operators are used in image processing due to their robust performance in preserving the shape of a signal ,while suppressing the noise [22]. The basic idea in mathematical morphology is to convolve an image with a given mask (known as the structuring element) [23].
There are two basic morphological operators: erosion and dilation, opening and closing are two the image X and B_x is nonempty. This is often referred to as B_x hitting X and, therefore, the name "hit " is often applied to this operation. It may be expressed as [26]:

4.
Experiments and methodologies The method was executed by using the Mat lab program; the proposed procedure to achieve the extraction of the tumour regions is summarized in Figure (1).   (291 x 273). The adopted images were acquired from different resources: Hila teaching hospital and websites. Figure 2 presents the experimental images.

6.
Results and discussion in this study The results of implementing the proposed work are as follows:

K-Means Algorithm K-Means clustering algorithm was implemented with different number of clusters for the six input images,
It is found that the proper number of clusters were (3, 5, 6, 6, 3 and 5) for the images (bon1, bon 2, bon 3, bon 4, bon 5 and bon 6) respectively. Figure (3) represent the results of implementing this method.

6.3
Comparison By comparison the results of tumour region in grey that obtained from the manual delineation the abnormal region and the tumour region in grey that obtained from the applying the K-Means algorithm.
(a) (b) Figure 5. Represent the comparison between the original images in the first row, the second row extracted abnormal regions in grey by K-Means algorithm and the third row represent the extracted abnormal regions in grey by manual delineation.

Conclusion
In this work, clustering based segmentation method K-Means, was implemented to extract the bone tumour in MRI and CT scan images with different number of clusters. From the result. It was cleared that the proper number of clusters were (3, 5, 6, 6, 5, 3) for MRI and CT images respectively. As well as the results insure the high quality performance of the adopted segmentation method since the present relative differences with the radiologist delineation were ranged from (0.63 to 1.75) % for MRI images and for CT scan images were ranged from (0.34 to 1.51) %.