THESIS
2008
xii, 92 leaves : ill. ; 30 cm
Abstract
Medical cerebral images can be classified to two categories: structural and functional. The former embodies anatomical structures in brains and the latter reflects the metabolic and physical characteristics of brains. The goal of segmentation of structural brain images is to divide the brain into meaningful subregions, such as tissues or structures. In this thesis, we target the problem of cerebral image segmentation and present two automatic methods on cerebral tissues and structures, respectively....[
Read more ]
Medical cerebral images can be classified to two categories: structural and functional. The former embodies anatomical structures in brains and the latter reflects the metabolic and physical characteristics of brains. The goal of segmentation of structural brain images is to divide the brain into meaningful subregions, such as tissues or structures. In this thesis, we target the problem of cerebral image segmentation and present two automatic methods on cerebral tissues and structures, respectively.
The proposed method on tissue segmentation combines a label MRF with a boundary MRF and admits sophisticated prior information about boundary patterns. It considers all possible configurations of labels and edge positions in the case where a single edge goes through a local window. The prior information is general and training for different data is not needed.
The proposed method on structure segmentation fuses the information of edge features, region statistics and inter-structure constraints for detecting and locating all target brain structures such that manual initialization is unnecessary. The multi-object template is organized in the form of a hierarchical Markov dependence tree, which makes the matching of multiple objects computationally efficient. The final segmentation is obtained through refinement by a B-spline based non-rigid registration between the exemplar image and the target image. Our approach needs only one example as training data and alleviates the demand of a large training set.
A variety of related works on the same topics are briefly introduced and theoretical comparisons are made. Experiments have been performed on real medical data sets and the accuracy, efficiency and robustness of the proposed methods are shown in this thesis. The two proposed methods have some advantages over related works and their weaknesses are also discussed in the thesis. In the field of neuroimaging, we foresee that the research interests will be moving from low-level processing to high-level analysis and from structural brain study to functional brain research. The work in this thesis may be part of the foundation for higher-level analysis and functional work on human brains.
Post a Comment