Abstract
In the recent years, statistical PET image reconstruction is gaining attention. The statistical image reconstruction can describe the relationship between the source and the projection data by an accurate probability model which includes the parameters such as geometric, attenuation factor and detector efficiency. The size of the matrix is very huge, and it increases the burden for storing and computing. In order to overcome this problem, we presented a new probability matrix by exploiting its sparseness and the high degree of symmetry. We reduced the size to 0.006% of the full size for an image with size of 128 128. OSEM method groups projection data into an ordered sequence of subsets. We effectively improve the slow convergence and long computation time of MLEM methods. One problem in employing the method is the fact that, after a certain stage in the iterative process, the noise properties of the reconstructed images start to deteriorate. The noise in the reconstructed image increased with the convergent rate. To overcome this problem, we presented a modified OSEM method which exploits the multi-resolution approach of MREM method to the ordered subsets of OSEM methods. The modified OSEM method not only has the advantage of fast convergence, but also slows down the deterioration of the reconstructed images after certain stage.