Research On Non-Rigid Registration Algorithm Of Medical Images

In image processing one is often interested not only in analyzing one image but in comparing or combining the information given by different images. For this reason, image registration is one of the fundamental tasks within image processing. The task of image registration is to find an optimal geometric transformation between corresponding image data. In practice, the concrete type of the geometric transformation as well as the notions of optimal and corresponding depend on the specifie: application. Image registration has been successfully applied in many practical aspects, for example, astro-and geophysics, computer vision, and medicine. In the last three decades, image registration has played an increasingly important role particularly in medical imaging. Registered images are now used routinely in a multitude of different applications such as the treatment verification of pre-and post-intervention images and time evolution of an injected agent subject to pa-tient motion. Image registration is also useful to take full advantage of the complementary information coming from multi-modality image.In processing of medical images, anatomy structures of various parts of body are complicated, thus simple rigid transformation can not adequately simulate the deforma-tion of complicated structures and local detail information. However, the deformation of detail features play an important role in diagnosis and treatment. Consequently, non-rigid registration of medical images is more accordant with practical circumstances. At present, the study of the rigid registration has been well studied, Problems in Non-rigid registration have remained widely open. It is worth to further be discussed that how to establish an appropriate transformation, how to select an interpolation algorithm, how to improve the speed and the precision, and how to evaluate the effect of the non-rigid method.In non-rigid registration technique of medical images, current researches mainly focus on the method based on the feature and the non-parameter model. On the basis of thor-ough research on these both methods, this dissertation discusses the similarity measures, the bright correction field and the constraint feature, and proposes some new methods in establishing the energy models, optimization methods and implementation details. The main contributions can be summarized as follows:1. We propose an non-rigid registration method of medical images based on Renyi entropic graph.The registration algorithm of feature-matched is one of the most well-known meth-ods in non-rigid registration of medical images. In feature points matching, the selection of feature points is of critical importance, when the structure of the registered image is unclear, most commonly used feature extracting methods often fails to find a suffi-cient quantity of matching points. Furthermore, the mismatching of features points is inevitable. Thus, the selecting of feature points need to be further improved. The normal non-rigid transformation has not adjustable parameters. Based on the above analysis, we adopt multi-resolution strategy, and propose a registration algorithm of pyramid struc-ture with non-uniform layers and blocks. This algorithm uses Renyi entropic minimum spanning graph as similarity measures of sub-block image. Compared with mutual in-formation, this similarity measure is simple and time-saving. It is a method particularly suited for non-rigid registration of multi-modal images. Furthermore the method adopts the multi-quadric transformation to simulate the non-rigid deformation of the registered images. The advantage of this transformation lies in controlling the degree of deformation. Experiments show that this method can establish non-rigid registration of multi-modality medical images. Moreover this method runs quickly and achieves high accuracy results.2. We propose an improved non-rigid registration of medical images based on a non-parameter model.In group of similarity measure functions, the different intensity between the two reg-istered images is very simple and easy to carry out. However, this similarity measure has an assumption:the intensity of same location between the registered images keeps an constant. In many practical applications, this assumption is not satisfied, it is af-fected by a lot of actual situation. Thus, we need take account on the change of gray value. We present a improved algorithm based on non-parameter model. Considering the change of the intensity value, the model takes a new regularization filed of intensity as the data term, which overcomes the drawback of gray constant conjecture and is simple in terms of computational complexity and implementation. Furthermore, the model adopts a structure-adaptive regularization, which can estimate the discontinuation displacemen-t of object. In terms of implementation, we propose a more effective multi-resolution scheme, which integrates the twice downsampling strategy with a support-weight median filter. Numerous experiments show that our method is more effective and produces more accurate results for image registration.3. We propose a new non-rigid registration of medical images preserving details.At present, research about non-rigid registration of medical images focuses on non-parameter model. In processing applications of medical images, the deformation of body's different part is very complicated, the displacement estimation of some features is also in-significant for diagnosing and treating. The class non-rigid registration of medical images based on non-parameter model can recover the displacement field for most complicat-ed structures, because the displacement field is estimated for each pixel. However, this method cannot precisely estimate the displacement of some features. Based on the above analysis, we present a novel non-parameter energy model. The model not only contains the classic data term and regularization term, but also the feature term. The feature term has fully taken the displacement of some features into consideration, and constrains the displacement of the whole image. Thus, the proposed model can estimate the dis-placement of some features. The experimental results show this method is an accurate non-rigid registration of medical images with preserving details.