| Diffusion magnetic resonance imaging is a modern technology which is based on magnetic resonance imaging to investigate into tissue architecture at the microscopic level in vivo. Diffusion weighted imaging adds the diffusion sensitive gradient magnetic field to acquire the diffusion information, and further diffusion tensor can be calculated by using multiple volumes of DWI. Diffusion tensor images are then constructed, and tractography technique can be used to construct pathways in deep white matter regions, which provides a qualitative visualization of fiber tracts. Relevant Research shows that tractography and fibers clustering analysis is valuable in the anatomical and connectivity researches of brain nerve fiber, as well as the applications in clinical diagnosis and treatment.The pipeline of our study is:Firstly the eddy-current distortion of origin DWI images is corrected by a systematic method, and the whole brain fibers can be estimated from the corrected diffusion tensor of DTI, then a novel visualization method is proposed to visualize the fiber tracts. Finally, many fiber clustering methods are proposed and compared to improve the fiber visualization. The core research works in the study includes three parts. The first part is an improvement of the systematic correction algorithm. The second part is DTI tensor based fiber tracking and line illuminating visualization of fibers. And the third part is fiber clustering.The main contributes of our study are presented as follows:1. We analysis the cause of eddy current distortion in DWI images, and summarize the mainly three classes of methods to solve the EC problem, as well as the post-process image registration based distortion correction methods. Further we propose a systematic method, which firstly calculates the distortion parameters of every slice, and then smooth the parameters incorporate the parameters of the neighbor slice to warp the images. Its improvement is that it considers the problem of high b value DWI image which is not appropriate for ICC, and meanwhile it incorporates information of neighbor slice to obtain the optimized parameters.2. We summarize the streamline fiber tracking, including FACT and Tensorline algorithms, and analysis the comparison between those. We propose a fiber tracts visualization using a line illuminating model instead of streamtubes, and introduce two local lighting models, i.e. maximum reflection principle and cylinder averaging. The experiment shows that our performance is better to overcome the delay problem in interactive manipulation.3. We analysis the characteristic of nerve fibers, according that introduce some popular used similarity measure of fibers, such as nearest neighbor distance, Hausdorff distance, and also summarize the unsupervised clustering algorithms. The aim to cluster fibers is to overcome the clutter observation and to help identifying fiber clusters. At last we show the comparison between the result of fiber clustering and thee manual classification of brain fiber clusters.The cooperation of the domestic and foreign relevant research is growing fast, and the demand of the application of dMRI is increasing, hence our study is not only significance in theoretical research in DTI, but also valuable for the medical diagnosis and treatment. |
