Multi-modality image reconstruction with a runtime segmented anatomical prior

Abstract

Multimodality imaging methods that integrate positron emission tomography (PET) with computed tomography (CT) or magnetic resonance imaging (MRI) has gained great popularity in clinical use. The advance of hardware allows for both anatomical and functional images to be acquired during one scan. These two sets of images can be registered readily to help identify tissue boundaries in PET images and thereby yielding images with better contrast recovery. In this study, we used an order-subset expectation maximization (OSEM) reconstruction method with a label mean prior (LMP) or median root prior (MRP). In order to avoid the artifacts caused by these errors, we proposed a runtime segmentation scheme, which re-computes the region labels alongside the iteration process. The priors can reduce noise contamination without blurring the tissue interface. In this work, we also took into consideration the inconsistencies between functional and anatomical images. Consequently, the tissue boundaries were estimated after each subset of iteration. Computer simulation studies were carried out to investigate the usefulness of the proposed algorithm. The performance of the proposed method will be evaluated in terms of image quality, and the effectiveness in compensating signal mismatches.