Research On Medical Image Registration Theory And Application

With development of modern medical imaging systems, medical images become more and more important in medical applications. It brings diagnostics, computer aided surgery, focus monitoring with support of high quality information. However, there are new problems for us to solve.Medical image registration is a way of eliminating the incon- stancy. The term of'image registration'comes from electronic engineering, which was defined as: image registration is a course of processing two or more images obtained from different sensors, different time, different angle or so no to align them with best possible match. Fundamentally, image registration need to analysis the geometrical detorsion in both images and uniform them in a single coordinate system by imposing a geometrical transformation. Image registration is the most important step in all the image analysis methods, which is based on the fact that only by combining image in a right way that we can obtain the most information we need.Medical images are always acquired in different times, with different poising positions, from different equipments.Dimini- shing the inconstancy between them is a problem to be solves. Medical images features small dynamic range, low contrast, obscure stigmas and high noise, which made medical image processing a separate research field from general image processing.By defining the term of medical image registration, it is a problem of finding the best mapping from one image space to another image space. This problem can be subtract into several sub- problems, like transform, metric, interpolating, optimizing. Medical image registration can be cataloged into single mode registration and multi-mode registration by the way images are acquired; 2D registration and 3D registration by image dimensions; or by the type of sub-problems, like mutual information regis- tration and so on. In this article, medical image registration problem is considered as a process of searching for best mapping method from one image space to another.ITK is the most popular one of medical image registration. ITK is short for Insight medical image segmentation and registration Toolkit, which is a open-source project. ITK is a powerful tool for medical image registration programming. But its complicated structure makes it difficult to develop. ITK is based on generic programming and data streaming structure. With this idea, this structure can be evaluated as a pipeline. Data are inputted at one end, processed by several parts which are connected by pipes, and outputted at the other end of pipeline.Despite the rapid development of medical image registration research, there are still some problems within:1, medical image registration concepts and theories are mass without organization. Medical image registration combines medical imaging technique, image preprocessing, geometry transformation, image interpolating, optimization searching technique, similarity metric and so on. There are a lot of articles on these subjects. However, there is still a lack of uniformed collection solely on medical image registration. In addition, most articles are written in English, which is a abyss in front of Chinese researchers.2, ITK image registration system is very tricky. The compli- cation lies not only in it's software structure, but also lies in its applications. Searching for best parameter setting is an important problem in ITK applications.3, non-rigid registration needs a lot of running time, parameter setting is very tricky, and in a great chance the output may be proved to be no use. This is not acceptable for processing a great deal of images.Basing on problems above, this article includes chapters:In chapter 1, basic concepts of medical image registration and ITK are introduced.In chapter 2, details of medical image registration theory are explained. Medical image registration include sub-problems as transforming, image interpolating, metric and optimizing. Trans- forming is for mapping one image space to another image space. Give the fact that computer based images are represented in grids, we need interpolators to give intensity of a point when it is mapped outside of a grid point. Metric is the most important part of all. It is used to quantitively evaluated how much an image matches another one. Optimizing is the process of find route from origin to best solution in space formed by transform parameters. These theory are very complicated. They are organized with regard of ITK system, in convenience for readers to compare with the next chapter, so that these abstract concept can be comprehended with real-world applications.In chapter 3, there are detailed explanations in ITK components and their relationship.In chapter 4, 2 experiments on medical image registration are given. Medical image registration is closely related to real-world applications. Experiments are best way to understand the theories. The first experiment focuses on parameter setting problem. The second one starts from an assumption and discusses if a non-rigid registration problem can be solved by a rigid registration method. With these 2 experiments, some problems have been discovered, and some are solved by suggesting possible solutions. This article is from a view of non-medical major researcher. Theories a introduced with close relation to real world appli- cations. Along the way, many problem are aroused and possible solutions are given. The author hopes that this article is of help to those who are dedicated to medical image registration research. We may wish a better further to medical image registration.