Research On Some Key Techniques For Beyond Wavelet Transforms And Level Set Methods In Medical Image Fusion

Along with the popularization and development of computer technology in the field of medical imaging, all sorts of medical image systems has played a crucial role in medical diagnosis, retrieval and clinical teaching etc. However, all kinds of medical imaging equipment have different principles of imageing. They reflect the positional and pathological information of human body respectively from different angles. Therefore, by the image registration, image segmentation and image fusion technology, we can make full use of superiority of various image information, as much as possible to express various kinds of information in an image of human body. The structure functions of human body’s internal organs can be expressed, through computer to provide more intuitive, convenient, accurate and scientifical data for doctors and improve auxiliary diagnostic accuracy.This thesis is supported by Ministry of Education Culture, Sports, Science and Technology of Japan’s natural science fund project:"Research on3D Medical Image Fusion and Representation Technology", Jiangsu province "Six Talents Peak" the7th high-level talents project:"Multivariate and sub-Gaussian Distribution Model based Image Retrieval" and Japan Auto-System company cooperation projects:"X-ray Medical Imaging System Research and Development". The main research of this thesis is about CT and MRI images fusion based on beyond wavelet transforms and level set methods, and analyses the medical image registration technology, medical image level set technology and beyond wavelet transforms image fusion techniques. The main contributions of this thesis are described as follows:(1) Proposed the neighborhood energy based image fusion method.This paper firstly proposed the neighborhood energy based beyond wavelet transforms in medical image fusion. Firstly, respectively input CT and MRI images, after the beyond wavelet transforms; we get high-frequency coefficients and low-frequency coefficients of these two images. For low-frequency coefficients, we use neighborhood energy method to select the suitable fusion coefficients. For high frequency coefficients, the maximum absolute value method or the other ways are used to select fusion coefficient. Through a lot of experiments in this thesis, we found that it gets a very good fusion result.(2) Proposed an improved Chan-Vese level set segmentation algorithm.In order to further reduce the image data quantity, raise the image processing speed, we adopted a new improved Vese Chan level set image segmentation algorithm. The new algorithm can obviously improve the segmentation speed; effectively segment the medical image lesions area. Experiments show that this algorithm has some robustness. Compare with the traditional algorithm, it enhances the speed more than30times.