PET image reconstruction via an iterative correction scheme

The maximum-likelihood and expectation-maximization (ML-EM) algorithm as well as its accelerated variant ordered subset expectation-maximization (OS-EM) have been extensively used in positron emission tomography (PET) image reconstruction. The principal drawback of the ML-EM/OS-EM algorithm is the “checkerboard effect” with increasing iteration number. One approach to alleviate it is to filter reconstructed images during or after reconstruction process. This study investigated the applications of three advanced image-filtering algorithms, namely, total variation, bilateral filtering, and nonlocal means via an iterative correction (IC) scheme for addressing the drawback of ML-EM/OS-EM. Numerical simulated and physically acquired PET datasets were used to test the proposed IC scheme. Our results demonstrate that the IC with image-filtering scheme can achieve significant gains than the conventional ML-EM/OS-EM algorithm in reducing noise, preserving edges, and reducing the “checkerboard effect”.