Based On The Improved Spectral Clustering Algorithm In Medical Image Application

Image segmentation is one of the key research subjects in the field of image processing and computer vision. Effective image segmentation can provide the basis for the smooth development of the follow-up research work. Especially in recent years, the medical image segmentation has been widely attracted by domestic and international scholars. An effective medical image segmentation can provide a reliable basis for clinical diagnosis, doing treatment plan and pathology study. Medical image segmentation consists of three types:human organ segmentation, organ lesion and other tissue segmentation. Currently, there are still some difficulties in dividing the following pixel-based medical image segmentation:Firstly, the information of pixel space near distribution makes segmentation when it is difficult to catch a local sub problem; Secondly, the Euclidean distance similarity measure does not adequately reflect the complexity of the sample distribution space, resulting in local optimal solution.Spectral clustering is one of the popular methods of cluster analysis. In this paper,spectral clustering is used as the research object, deeply analyzes the similarity measurement based on Euclidean distance and the application of the improved spectral clustering algorithm in medical image segmentation. In order to solve the problem of medical images based on pixel is difficult to carry out effective segmentation problem.This paper made innovation the following two aspects: on the one hand, the global problem is divided into sub-problems with strong correlation to extract local features to improve the accuracy of image segmentation; on the other hand, proposed manifold learning structure similarity matrix to get the global consistency of the segmentation image.The main research work of this paper is as follows:(1) To clarify the research background and significance of medical image segmentation.This paper briefly introduces several common methods for segmentation of medical image.The basic theory of spectral clustering is described in detail, including the K-means clustering algorithm, the spectral graph partitioning theory, Laplace matrix properties, as well as spectral clustering algorithm.(2) Different medical tissue structure(such as the anatomical structure and pathologicalchanges), defined local sub-problem is difficult to be incorporated into the global model, so the global model is difficult to solve problems. Since the local sub-problem consists of a large number of sub-graphs which are not reflected by the whole appearance image, in order to solve the local sub-problem, this paper proposes to divide the global problem domain into a targeted local sub-problem domain.(3) Based on the Euclidean distance similarity measure, we propose a new measure of spectral clustering algorithm based on manifold learning. Using manifold learning structure similarity matrix, can enhance the inner class similarity, and weaken the inter class similarity.In order to verify the validity of the proposed method, respectively used the algorithm proposed in this paper and existing improved spectral clustering algorithm on the brain magnetic resonance imaging(MRI) images. The experimental data results show that proposed approach obtains better effect of segmentation.