PET parametric imaging based on MR frequency-domain texture information

An integrated positron emission tomography (PET)/magnetic resonance (MR) system combines the unique advantages of the two separate systems to achieve dual modality imaging in the same spatiotemporal region. With this advantage, PET parametric imaging guided by coregistered MRI can effectively improve the spatial resolution of an image. The existing guided imaging methods generally use only the neighborhood intensity information of MR and ignore the texture features in images, but these features can be used as a priori information. Therefore, we propose a novel method of MR-guided kernel reconstruction in this paper. First, a new kernel matrix is constructed by using Gabor texture and voxel intensity information from MR, including diverse image information in the frequency and spatial domains. Then, a new kernel matrix is applied for iterative parametric imaging to obtain improved images. Simulation experiments with PET/MR digital phantom and clinical patient images demonstrate that the additional a priori texture features extracted from the images can further improve the quality of parametric imaging. Compared with the maximum likelihood expectation maximum (MLEM) algorithm and the original kernel algorithm, the proposed method effectively improves tumor recovery and noise reduction in parametric imaging.