Using K-mean Clustering to Classify the Kidney Images

Abstract

      This study has applied digital image processing on three-dimensional C.T. images to detect and diagnose kidney diseases.  Medical images of different cases of kidney diseases were compared with those of   healthy cases. Four different kidneys disorders, such as stones, tumors (cancer), cysts, and renal fibrosis were considered in additional to healthy tissues. This method helps in differentiating between the healthy and diseased kidney tissues. It can detect tumors in its very early stages, before they grow large enough to be seen by the human eye. The method used for segmentation and texture analysis was the k-means with co-occurrence matrix. The k-means separates the healthy classes and the tumor classes, and the affected parts were isolated from the healthy parts. To isolate the kidney from the other anatomical parts in a CT image, a mask must be generated, which is a binary image (0s or 1s). This mask was also utilized to remove undesired characteristics from the images. Density slicing was utilized to color the image based on its texture density. A slice is considered a band of neighboring gray levels in a gray scale image seen through monocular color. The gray scale band of (0-255) is transformed into a variety of color slices; it is the conversion of a gray scale image to a colored image that efficiently displays symmetric and diverse regions. Density slicing is a property process for segmentation. The unsupervised classification process, the K-Mean clustering, is used the application of K-mean on C.T. images to detect and classify the type of tumor in the kidney. The K-mean clustering separates each class depending on the texture properties and the distance from each class and color. This method of segmentation was used to separate the affected part from the healthy part of the tissue; the K-mean with Co-occurrence matrices gives statistical properties such as energy, homogeneity, contrast, and correlation. These give an indication of the nature of the tissues that are different in density. The standard deviation for the cancer was higher than the stone, so was the mean, the contrast and the correlation. This means that the texture of the cancer was brighter and has a none of grey level more than the stone and this can be seen from the energy value; the texture of the cancer was highly correlated. This method proved to be a good method for the early diagnosis.