An Investigation of Block-Sequential Algorithms in Statistical PET Image Reconstruction
Statistical iterative reconstructions have been widely applied for clinical PET imaging. However, performance of reconstruction methods greatly depends on the choice of objective functions and the corresponding reconstruction algorithms. In this research, we investigate two important PET image estimations: maximum likelihood (ML) and maximum a posteriori (MAP). And the corresponding image estimation problems are solved by ordered subset expectation maximization (OSEM) and block sequential regularized expectation maximization (BSREM), respectively. These two fast iterative algorithms are categorized as the block-sequential algorithms. Using the contrast recovery coefficient (CRC) as a figure of merit, we compare how image estimation methods affect reconstruction accuracy. For a small 0.2㏄lesion, the CRCs of MAP-BSREM are 20-50 % higher than the CRCs of ML-OSEM under matched background variations. The results demonstrate that the MAP-BSREM estimation can improve the detectability of small hot lesions markedly.