Retrospective Medical Image Registration

Medical image registration and fusion is a crossing research topic of information science, computer image technology and modern medicine. Retrospective medical image registration is a key technology to make full use of the modern multimodality medical images integratedly. It is very important to the automatic analysis of serial images and multimodality images, and widely used in the areas of clinical diagnoses, therapy, quality assurance and evaluation.This thesis is devoted to a general study of retrospective medical image registration. Related concepts, methods and the up-to-the-date situation are briefly reviewed at first. Then, thorough research is conducted to feature based rigid registration, image content based rigid image registration, and nonrigid image registration respectively.Much research has been made on feature based image registration. It is widely used in clinic practice. We introduced a method based on wavelet transform, which extracted the modul maxima in the wavelet coefficient domain of one of the images to be registered according to the wavelet coefficients of the other image, then inversely transformed the extracted data to the image domain to measure the difference between the original and recovered image. With this method, we have completed successful registration between CT and MRI images.The main research interests focued on the whole-image-content-based registration methods, operating directly on image gray levels. Those methods can make full use of the information available in the images to determine the transformation parameters. Global registration based on the criteria of maximization of mutual information seems now the principal method in multimodality medical image registration, owing to its superior properties of accuracy, robustness and fully automated procedure. However, the objective function suffers from the artifacts of local maxima induced by interpolation while the translation distance coincides with integer times of the pixel dimensions. The optimizing procedure may end at the local maxima and misregistration happens. The reasons why the artifacts occur are analyzed inthe thesis, and two methods are proposed to solve the problem. With the new method, the objective function becomes more smooth and the probability of successful registration is guaranteed without significantly increase the computational complexity.Elastic image registration based on thin plate spline interpolation is studied. With the corresponding feature points extracted from the images, elastic registration of the out contours is achieved.A novel method called "segmentation-counting method" used for serial image registration is proposed in this thesis, which is based on the joint histogram of the two images to be registered. Usually, it is not difficult, by using a properly selected threshold, to separate human tissue from the background in medical images. With the thresholds of the two image sets to be registered, the joint histogram is divided into four separated regions. The configuration of the joint histogram bears a greatly different appearance between the non-registered and registered images. This different appearance reflects how good of the registration. The objective function is defined as counting the number of points in a specific region or several regions of the joint histogram. The new method simplified the computational complexity greatly and speeded up the registration process significantly. With the fine property of the criterion function, Powell's direction set method in multi-dimensions is chosen to carry out the optimizing process to calculate the registration parameters. The comparison of the results from both mutual information based method and our method shows that the new method based on thresholding and counting is a fast, simple, efficient and accurate registration method.It is an important morphological research topic to reconstruct the 3D shape from serial section tissue images. An investigation of three dimensional microstructure registration of seria...