Research On Rigid And Non-Rigid Registration Of Medical Images

Medical image registration can be defined as the process of aligning two or more medical images which aims at finding the optimal transformation to best aligns the input images. Medical image registration plays an important role in the clinical area.It is a crucial step for image analysis and help combine valuable informations from more than one image; i.e.images captured at different moment,from different viewpoints or by different equipments.Image regiatration help providing more reference resources thus raise the incuracy of diagnosis.However,some organs from human beings stay stable at a period of time,others such as lung make autonomic movement,such organs have different sizes,different shapes at different times.According to this feature, the thesis mainly focuses on developing rigid medical image registration method and non-rigid medical image regiatration method to adapt this two kinds of images and optimize the process of registration, finally achieving accurate registration of two images.The main work of this paper is as follows.1. For the images of none autonomic movement organs,i.e.images of humans’ skull,this paper adopt rigid registration method based on the normalized mutual information.Firstly, we use Gauss low-pass filtering mothod to preprocess the reference image and the floating image.Then we use PV interpolation method to count the joint histogram of two images.according to the histogram,we can calculate the mutual information of two imges.Furthermore,Powell optimal search algorithm help finding the registration parameters base on the theory of maximum mutual information.2. Powell optimal search algorithm is based on the one dimension search algorithm, thus,different one dimension search algorithms induce different deviations between the reference image and the floating image.In this paper,Powell optimal search algorithm uinte Fibonacci method and Brent method respectively to compare the regiatration results.3. For the images of autonomic movement, i.e. images of humans’ lung,this paper adopt non-rigid medical image registration method.This method is based on the local control characteristics of B-spline.Firstly,this method define a standard grid to change the pixel location and gray value of the floating image and use the error gradient decent method to transform the control grid,thus the control mesh can smoothly fit the image after deformation.For the inaccuracies existing in a single B-spline registration,we adopt multilevel B-spline to solve this problem.The control grid points keep increasing and each layer superimposed on each orther to achieve the final registration effects.4. Since the computation of the original B-spline deformation registration method is considerable large,so it cannot meet the requirements of real-time registration of the clinical.In this paper,we first use moments spindle method to get coarse registration between two images,then adopt a local update strategy based on hierarchical B-spline fast registration method to achieve the final registration.