This thesis contributes to the comparison of different global optimization methods to the MEG source localization problem. Moreover, we propose some new approaches to this issue. Our methods perform over the exiting ones. The thesis is organized as follows. Chapter 1 gives the introduction of my work. The functions of different brain image techniques are introduced. Some basic idea of MEG and the formulations of its forward and inverse problem are given. Moreover, the state of the art of the MEG study is reviewed as well. Chapter 2 deals with three typical global optimization methods, i.e., the hybrid genetic algorithm, simulated annealing algorithm and Tabu search algorithm, which are proposed for MEG source localization. The details of our implementation and the experimental results and discussions of them are also given in this chapter. We first introduce a hybrid algorithm by combining genetic and local search strategies. Then, we apply the Tabu search to source localization. Finally, in order to further compare the performance of above algorithms, simulated annealing algorithm, which is popular in this problem, is also applied. The computer simulation results show that the proposed hybrid genetic algorithm is the most effective approach to dipole localization, and the Tabu search algorithm is also a very good strategy for this problem. In chapter 3 the hybrid genetic algorithm is further developed to the parallel computer system. The simulated annealing method described in chapter 2 is improved and parallelized as well. The comparison of parallel genetic algorithm (PGA) and parallel simulated annealing method (PSA) is given. The experimental results show that the parallel computation can improve the computation speed linearly, and the PGA is superior to PSA. Chapter 4 is about the new development in the study of MEG source localization in modeling, the introduction of other information, and the new advancement in algorithms; and on the basis of the previous work, the method proposed by us using PGA to get probabilistic solutions under the Bayesian framework is described and discussed.
修改评论