The Applications Of Wavelets And Normalized Cuts In Medical MR Image Segmentation

Medical image segmentation is a significantly typical application of image segmentation in medical fields. The computer-aided diagnostic system, which is promoted by medical image segmentation, plays a more and more important role in modern medical development. More importantly, the development of medical image segmentation technology not only deeply influences other related techniques (such as visualization, 3-D reconstruction, operation simulation, graph-guided operation and so on), but also plays a crucial role in biomedical image analyses. Furthermore, since brain MR image (especially the one of hippocampus) segmentation is the key and foundation of volume measurement and 3-D reconstruction of brain tissue, the volume measurement of hippocampus provides an important high-tech method for some medical image diagnoses. And it also has significant application values in the computer-aided diagnoses of such illnesses as temporal lobe epilepsy, Alzheime's disease and schizophrenia. These years more and more research focused on medical image segmentation methods and achieved great accomplishment. However, the existent semi-automatic and automatic brain image segmentation methods, as well as atlas-based registration methods, have many limitations, such as the limitation of local features of image, poor ability for anti-interference and too low computation speed. These aspects all confine their future application in medical diagnoses.In order to solve the limitation of such existent methods, combined with the requirement of theme and the special characteristics of medical image processing, this dissertation mainly focuses on the new thoughts and new algorithms for brain MR images processing, hippocampus segmentation and other medical image segmentation. Such new algorithms include wavelet and watershed-based medical image segmentation, wavelet-based medical image boundary tracing method and improved normalized cuts based medical image segmentation. The detailed research content and results are shown as below: 1) Since watershed based image segmentation has the drawback of...