Regularization of PET reconstruction using multi-scale adaptive thresholding

Abstract

A multi-scale adaptive thresholding scheme is presented in this study. It was evaluated as a regularization process to filtered back-projection (FBP) for reconstructing clinical PET brain data. Adaptive selection of thresholding operators for each multi-scale sub-band enabled a unified process for noise removal and feature enhancement. A cross-scale regularization process was utilized as an effective signal recovering operator. Together with non-linear thresholding and enhancement operators, they offered remarkable postprocessing to FBP reconstructed data. In addition, such effectiveness was formulated as a regularization process to optimize FBP reconstruction. A comparison study with multiscale regularized FBP (MFBP), standard FBP with clinical settings and iterative reconstruction (OSEM) was reported. The proposed regularization process has shown competitive improvement in the image quality of PET reconstructions when compared to the current state-of-the-art method used in clinical commercial systems (OSEM).