Parallel Computing For Multi-pinhole SPECT Reconstruction Algorithm

Nuclear medical imaging systems such as SPECT and PET play an important rolein modern medical imaging.Implementation of a SPECT system.includes software andhardware designing.The most important consideration in the design of software system is the choice ofimage reconstruction algorithm. Commonly used ECT reconstruction algorithms areiterative and algorithms. Analytical reconstruction algorithm is characterized by itssimple model and fast calculation speed; however it has poor noise property. Recentlyordered subset expectation maximization (OSEM) algorithm is widely used in SPECTimage system, OSEM has the advantage of accurate system modeling, inhibition of thenoise in the reconstruction process. At the same time, he iterative algorithm leads tosubstantial computational burden and the accurate and complex system model leads tohuge memory usage for storing system transfer matrix in order to solve this problem, weneed to parallelize the algorithm and optimize the system transfer matrix storage.This thesis realizes the parallel OSEM algorithm using C++programminglanguage and MPICH API library.Parallel OSEM algorithm program runs in a two-node,quad-core computing cluster.This program is based on two parallelization strategies:image division and projection division. The basic idea is realizing a OSEM with thecomputing tasks are divided into different sections, assigned to run on differentcomputing nodes. Calculated by analyzing the parallel speedup achieved in the1to8parallel processes we achieved speedup factors of1.00,2.07,2.96,3.87,4.64,5.60,6.27,and7.32respectively. Due to the relatively small number of iterations of the serialprogram in the two level structure parallel systems the advantage of parallel OpenMPthreads is not obvious and therefore significant acceleration is not observed.