| 英文摘要 | Diffusion tensor magnetic resonance imaging (DT-MRI) is an important component of functional magnetic resonance (MR) imaging. With DT-MRI, diffusion anisotropy effects can be fully extracted, characterized, and exploited, providing even more exquisite details on tissue microstructure. DT-MRI has been used to demonstrate subtle abnormalities in a variety of diseases (including stroke, multiple sclerosis, schizophrenia, et al.) and is currently becoming part of many routine clinical protocols. The most advanced application is certainly that of fiber tracking in the brain, which in combination with BOLD-FMRI (Blood Oxygenation Level Dependant functional MRI), might open a window on the important issue of brain connectivity. DT-MR image processing and analysis is multi-disciplinary, touching on many techniques in mathematics, physics, medicine, and computer image processing. This dissertation mainly concentrates on some key techniques in DT-MR image processing and analysis, such as segmentation of ischemic lesion region, brain white matter tractography, and diffusion tensor regularization. The contribution of this dissertation is as follows: 1. An unsupervised method for segmentation of cerebral ischemic lesion from DT-MR images is proposed. It is very important to employ the techniques of DT-MRI in qualitative and quantitatively cerebral ischemia diagnosis, including accurately detecting the location and size of ischemic lesion volume, and quantitatively analyzing the changes in water diffusion anisotropy of ischemic lesion. In original DT-MR images, cerebral ischemic lesion shows up with high signal intensity while the intensity of nerve tracts is also high due to diffusion anisotropy. It is difficult to distinguish ischemic lesion from nerve tracts. In clinical diagnosis, mean diffusion weighted image is acquired to reduce the diffusion anisotropy for discerning the ischemic lesion region, but some useful information will be also reduced. So the best way is to do some analysis on original DT-MR images, including detecting the location and size of ischemic lesion volume. Automatic segmentation of ischemic lesions in DT-MR images is still a difficult issue mainly because of the disturbance of diffusion anisotropy, and manual region tracing methods are usually used. Accounting for noise, intensity overlapping, intensity inhomogeneity, and partial volume effect (PVE), a new adaptive unsupervised method for segmenting brain ischemic lesion from DT-MR images of stroke patients is developed based on multi-scale statistical classification (MSSC) and partial volume voxel reclassification (PVVR). By incorporting diffusion anisotropy into MSSC model, we can successfully distinguish ischemic lesion from nerve tracts. Accounts for intensity inhomogeneities in PVVR, partial volume voxels are reclassified based upon means and standard deviations of local region other than ofthe |
修改评论