Texture Features Based Medical Image Retrieval

With the appearance of medical imaging equipment in the medical field, medical imaging has become a basic tool in modern medicine. It plays an irreplaceable role in clinical diagnosis and treatment, medical teaching and research. Nowadays, its application results in large medical image data, and generate urgent demand of effective management and retrieval for them. How to efficiently manage large medical image database has become an urgent problem to be solved. The technology of content-based image retrieval (CBIR) is introduced into medical field. Research on content-based medical image retrieval (CBMIR) is a very significant work and becomes a hot topic in recent years.Content-based medical image retrieval has the following basic steps. Firstly, establish the medical image database. All images in database should be preprocessed and analyzed. Then, extract the features which can effectively describe the characteristics of images and establish feature database. Feature database and medical image database are associated by a particular identity. In the retrieval process, features of query medical image are extracted, and the compared with features of medical images in database. System retrieves the similar images from database according to the comparison result. The retrieval result would be evaluated by users. After learning the return information, the system repays a better retrieval result to users.This paper focuses on feature extraction of medical images, which is the key technology in retrieval system. After studying the existing feature extraction method, through combining medical imaging mechanism, a texture feature extraction method is presented based on amplitude-frequency modulation model. Use Euclidean distance to measure the similarity of features of different images. Based on this texture feature extraction method, a small content-based retrieval system of medical images is designed. This thesis includes the following work: (1) Systematically introduce the basic knowledge of content-based image retrieval and some key technologies of it, including:CBIR architecture, image representation and feature extraction methods, similarity matching, and relevance feedback and retrieval performance evaluation criteria.(2) Expatiate on nonlinear signal amplitude-frequency modulation model. Deeply study the basic principle of AM-FM models and energy separation algorithm and analyze the AM-FM model in the application of two-dimensional image signal. Finally, propose an improved ESA algorithm against discrete error arising in energy separation algorithm.(3) Grounded on the characteristics of medical images and MRI principle, first images are pre-treated, and then the improved energy separation algorithm (Gabor-ESA algorithm) is applied to extract AM-FM texture features of medical images. Use the normalized Euclidean distance to measure the degree of similarity between the images. Build image database and feature database offline, and then retrieval experiments are analyzed and discussed.(4) Implement a content-based medical image retrieval system. The system function modules are realized according to the actual needs. The function modules include: storing of images, pre-treatment for images, extraction of texture features, retrieval setting and retrieval stage, etc. Additional, some table structures in the database are interpreted.