Improvement of PET image reconstruction using high-resolution anatomic images

Summary form only given. The authors previously reported a Bayesian approach for PET (positron emission tomography) image reconstruction that incorporates prior information derived from spatially correlated, high-resolution CT (computed tomography) or MR (magnetic resonance) images. A potential problem in this approach is that discrepancies exist between functional and structural boundaries due to either the residue error in correlating the PET and CT/MR images, or the natural differences in the functional distribution and the anatomic map. The authors developed strategies for reducing the errors and artifacts resulting from this problem. For the residue mismatch between the PET and anatomic images, a refining registration scheme is performed on each slice by cross-validating the boundaries. Weighting functions and/or modified line site designs can be incorporated in the Gibbs random field model for enhancing the functional boundaries when there is an exact match with the anatomic boundaries, and for eliminating the effect of the anatomic boundaries when a mismatch exists. These strategies appear to be effective in reducing the artifacts and improving the image quality.<>