Automatic High Precision Biomedical Image Registration And Implementation

With the development of the computer technology, technology of digital biomedical image, applied to clinical diagnosis and treating extensively, has great influence on biomedical research and clinical diagnosis。 There are many kinds of formation modalities of medical image, for the principle of the formation of image is different from equipment。 We divide into two general categories: structural and functional。 One is the structural modalities that characterized by clear, definitive boundaries between adjacent regions; and the other is the functional modalities that generally less effective in representing anatomic structure , instead , provide information on physiological processes 。 And different medical images usually reflect different or complementary information, even the information overlapped partly。 For instance, the image of X Ray, CT, Magnetic Resonance (MR), etc, which mainly describe human body's physiological information, and provide an excellent physical representation of the scanned structure and are extremely useful in identifying regions of interest; The Positron Emission Topography (PET) , Single Photon Emission Computed Topography (SPECT ),etc, which mainly describe human metabolism information, such as the uptake of glucose, and are useful in detecting areas of low or high metabolic activity which may indicate dysfunction or tumourous growth。 These medical images are all essential for accurate diagnosis, treatment and operation to the patient。 However, because of limited resolution and noise issues, functional images lack the inherent accuracy of structural modalities。The benefit of registering functional images to structural images becomes clear: we can obtain a single representation, which possesses metabolic information with sufficient accuracy to pinpoint regions of interest.This paper introduces an automatic medical registering system on the basis of existing research results, which contains two main parts: one is the Maximum of Mutual Information (MI) from information theory, including correctness the localization. which is used for the whole image registration, for there is few assumptions are made about the nature of the image registration: the other is a landmark-based method. Thin Plate Spline (IPS), including the non-linear image deformation and the automatic landmark localization, and TPS is used to partial elastic registration to improve the accuracy of the registration result.This paper also introduces a new automatic landmark selected method, which is for the partial elastic registrational part of system.