Medical Image Registration And Pattern Recognition Technology

Medical imaging has been an important part of modern medical sciences. Medical imaging includes two separate parts: medical imaging system and medical image processing. Medical imaging system lays stress on the image acquiring process. Medical image processing is focused on image post-processing in order to improve image quality, extract image features or fuse different model of images.Medical image registration and fusion are research focuses recently. Medical image registration is the key procedure of image information integration for clinical diagnosis of cancer, cardiac and cerebral diseases. Although there have been some commercial image registration and fusion software, further researches are still been doing in registration theory, modeling and implement technique. This thesis emphasizes on solving the difficulties in implement and application of image registration considering both theory and practice. Besides, image pattern recognition and interpolation have also been investigated, which are useful for image registration.In chapter 1 of this thesis, first, registration methods based on conventional mutual information are investigated. The influence of factors such as optimization, multi-resolution to the matching accuracy is also discussed. Second, the application of higher-order mutual information, which applies the spatial information to avoid misregistration, is extended to registration of PET and MR images. Third, a novel registration algorithm based on morphological tools is proposed. The matching process is accelerated and spatial information is taken into account using this new technique. Finally, as a primary study, two elastic registration methods based on two sets of corresponding point landmarks for human eyeground and brain atlas images are presented.Since the technique of image pattern recognition is vital in medicine and has vase application in medicine, it has been investigated from both theory and application property in chapter 2. Two image features based on moment invariant have been proposed. The image description properties of them are also investigated. Then the abilities of pattern recognition are discussed using radial harmonic Fourier moments and 26 English alphabets. The practical application of pattern recognition in medicine is also investigated.In chapter 3, several scene-based interpolation methods are compared in three main areas typically encountered in medical applications. In each case, the efficiency and accuracy of a particular interpolation technique are evaluated with multi parameters. Some special suggestions have been made.