Density Function Based Medical Image Clustering Analysis and Research

Abstract

Image clustering analysis is one of the core techniques for image indexing, classification, identification and segmentation for image processing. On the basis of investigation on the laws of gray distribution of medical image, density function construction based medical image clustering analysis method is designed. For the special kind of data object -medical images, firstly its density function is constructed; secondly, a method of density function construction based on medical image clustering analysis and its implementation algorithm, that is, hill climbing, is offered; At last, abdomen medical images are used to do experiments according to those two algorithms. Experiment results show that medical image clustering on the ground of density function construction achieves good effects and can clearly express the content and semanteme of medical images. . INTRODUCTION Clustering analysis has wide applications in the fields of pattern recognition and image processing, especially in image identification and segmentation. Some of those researches have achieved great successes. In 1979 Coleman and Andrews used clustering analysis method to segment image. Stewart uses fuzzy clustering analysis to recognize and classify radar objects. Current applications of clustering analysis in this field are focused on improving traditional clustering analysis algorithms and methods for better quality and processing speed of segmentation and classification. Density based clustering method is to cluster data objects according to the concepts of density. It produces clusters by density of data object or some kind of density function. DBSCAN, DBCLASD, DENCLUE and so on, are typical density based clustering methods. From the end of 50s * Supported by the National Natural Science Foundation of P.R.China under Grant No. 60572112 and the Social Development Foundation of Zhenjiang, Jiangsu province of P.R.China under Grant No. SH2003014. to the beginning of 60s in last century, Rosenblatt and Parzen offered famous density estimation method, which doesn’t need prior knowledge of data distribution and any presupposition but can accurately express laws of data distribution. Data distribution mode of density estimation provides direct mean for definition and discovery of clustering. The clinical value of medical images is to show the projections of different tissues and organs of body on medical images and distinguish normal projections from abnormal ones. Expressive forms of projections are pixels gray of digital image and law of gray distribution. Investigation on the distribution law of different gray density has special clinical value. Currently, making use of clustering analysis method to identify and segment medical images is on germinal stages. On the ground of deep investigation on gray distribution for medical images, we present a new clustering analysis method for medical image processing -density function construction based medical image clustering analysis method. Researches show that this method can express medical image content very well and has obvious clustering effect; therefore this method is fit for medical image clustering analysis. This novel K. Elleithy et al. (eds.), Advances in Computer, Information, and Systems Sciences, and Engineering, 149–155. © 2006 Springer.