Research Of Algorithm Of Medical Image Registration And Segmentation

Medicine is related to the applied science of human health. The level of development and application of medical clinical research plays an important role on human health.This thesis mainly discusses about image segmentation and image registration algorithms. The main work is as follows:1. This thesis mainly introduces a parametric method for bias field estimation and segmentation of MR images. But for this medical segmentation method, it is not robustness to initialization. For some initialization, it would calculate a wrong result. Because of this disadvantage of the parametric method, combine the method of level set evolution without re-initialization. A new energy function for parametric method was proposed. It also demonstrates the effectiveness of the new energy function.2. With the development of medical instruments, the picture always extended to 3D or 4D. But the most registration algorithms in our country are applied on 2D images. This thesis does rigid registration between two 3D images, and also compared different similarity metric in registration process between two modality images.3. Due to the complexity of the medical image, simple algorithms cannot segment concrete physical structure in images. Atlas-based segmentation algorithm is a well-established technique for this segmentation. It combined the image segmentation and registration algorithms. But because of the individual differences and limitations of registration, it often failed to get good results. In this thesis, a new variation method for performing label fusion was proposed and that takes into account the image information and regularity of the region of interest. The results show that the new method works better for medical image segmentation.