| Summary: | 碩士 === 國立清華大學 === 原子科學系 === 90 === Parallel computing performs more than one independent computation for single event at the same time. Therefore, parallel computing must be executed on multiprocessor system. By partitioning task onto multiple processors, we could increase the computing rate. In medical image processing, the data structure of images is the same as that of matrices, and we usually process the image data as matrices. There is some problems about parallel computing: data independence, task partition, and data synchronization etc.. We''ll try to solve them by parallelizing the matrix multiplication problem, and find out an appropriate way to parallelize medical image processing. Iterative image reconstruction method, MLEM (Maximum Likelihood Expectation Maximization), could get better image quality than FBP, but iterative image reconstruction needs to implement forward and backward projection for many times. If we can parallelize the forward and backward projector efficiently, we will greatly decrease the execution time in MLEM. This faster iterative image reconstruction method will help us dealing with the clinical data. A suitable way for parallel computing will decrease the execution time of a process. If we chose an unsuitable way for parallel computing, it may cost more time than non-parallel computing program.
|
|---|
