Glomerulus Extraction And Nuclei Statistical Analysis

With the rapid development of computer science and digital image technology, medical images diagnosis plays more and more important role in modern medical treatment. The information provided by medical image has been indispensable for modern medicine domain, and it also becomes significant in clinic diagnosis, scientific research and teaching aspects.The aim of this paper is to study the recognition techniques of medical images according to the glomerulus segmentation from kidney-tissue image and statistical analysis of nuclei. Three glomerulus segmentation methods are proposed in this paper: (1) From sample images, it can be seen that the information of cavum is much stronger than glomerulus edge, so cavum is defined as boundary which is enhanced with a nonlinear threshold surface constructed by neural network under typical user-defined feature template. After being syncretized with traditional boundary, the complete glomerulus boundary can be obtained and object will be extracted successfully. (2) To reduce the difficulty of boundary enhancement, LOG filter is applied to enhance glomerulus edge. From the point of view of boundary search, genetic algorithm is used to search boundary in the images containing noises andaccordingly glomerulus will be extracted. (3) To preserve the integrality of real boundary, watershed algorithm is applied. The barycenter of glomerulus found by genetic algorithm is used as the seed for region growing, and then the object will be obtained.As for nuclei in the glomerulus, variable thresholds and eigenvalue feedback strategy are used to get binary image firstly. A nonlinear threshold surface is constructed by Gaussian function. Then through adjusting this surface to optimize user-defined target function, the optimal segmentation results will be got. In terms of statistics, average value and variance can be calculated to present pathological data.In this paper, image processing methods and intelligent algorithm are used to segment glomerulus and nuclei, and an automatic analysis system of the kidney image is realized. The experimental result indicates the good performance of this system.