The PET Image Reconstruction Using Anatomical Side Information As Priors

Positron Emission Tomography(abbreviated as PET)is a kind of nuclear medical imaging equipment.It could evaluate the tissue metabolism and functional activity in a non-invasive way.It is playing a more and more important role in the field of early diagnosis of tumor,postoperative evaluation,the nervous system and cardiovascular disease.PET image reconstruction is the process that transfer the projection data collected by the PET detectors to a 3D radiotracer distribution map,which makes a direct effect on the image quality.Undersampling PET may have the advantages of releasing the requirement of completeness of PET projection data.However,the reconstruction image under the undersampling circumstance suffers high noise and has low SNR.It is difficult to distinguish edges and image details,either.Those drawbacks are the key issues for the undersampling PET image reconstruction.The widely used MLEM/OSEM algorithm would also suffers high noise when the iteration number get high.With the development of the multi-modality medical image system like PET/CT or PET/MR,we can obtain the PET projection data in the meantime acquire the anatomical information of the object.This paper aims to utilize anatomical prior information and the characteristic of PET image like self-similarity to improve the PET reconstruction image quality.Based on the general algorithm framework(list mode based regularized relaxed ordered subset,abbreviated as LRMOS)developed by our team,I formulate two kinds of regularized reconstruction algorithms(named LMROS-NLM-CT and LMROS-BIL-CT)which could utilize the anatomical side information properly.And in this paper,I have made lots of simulations using the professional simulation package GATE and carried out the real experiment based on the small PET system developed by our team.The reconstruction results demonstrated that we could use the LMROS algorithm and the anatomical prior information to obtain high quality reconstruction images in different scene.Due to the huge computation burden in the 3D PET imaging,we take advantage of the GPU/CUDA to parallize the corresponding algorithms mentioned above and have achieved satisfactory acceleration results.It is expected that the work in this paper could provide some valuable references when utilizing the anatomical prior information to PET image reconstruction.