In the past two years, we have been working on MEG source localization and medical image processing using global optimization methods. Our major work includes the application of Genetic Algorithm, Simulated Annealing Algorithm, Taboo Search Algorithm to MEG source localization problem and the comparison of the performance of these methods. In order not to make it too long, this thesis does not include our work on medical image reconstruction. The thesis is organized as follows. In Chapter 1 we describe the necessity of our research into the optimization approaches to the MEG source analysis and medical image analysis. We also give a brief description of our major contributions and the organization of my thesis. Chapter 2 devotes to a description of the basic theory of MEG and the formulation of the MEG forward problem and inverse problem. Chapter 3 deals with the genetic algorithm, simulated annealing algorithm and taboo search algorithm. We give the details of our implementation of these algorithms on MEG source localization problem. In Chapter 4, We describe our design of the computer simulations of the algorithms described in the above chapter. We apply our genetic hybrid algorithm, simulated annealing algorithm and taboo search algorithm to this problem. Then we compare their performance on MEG source localization problem. Our results show that these algorithms all have the potential to find the dipole parameters correctly. As to the performance, our genetic hybrid algorithm performs best. Simulated annealing algorithm performs also well, but its computational cost is a little bit larger than genetic hybrid algorithm. Taboo search algorithm is not as good as expected. In some case, it performs very well, but in some difficult cases, it performs not as good as the genetic algorithm and simulated annealing do.
修改评论