论文部分内容阅读
Purpose:Head motion in PET/CT leads to artifacts due to the mismatch of PET and CT images.This work aims at reducing motion artifacts in PET image reconstruction.Methods:The correction method for head motion is based on the facts that the motion is rigid,irregular and within a short time.Our method contains three important steps.Firstly,the acquired data is cut into short time frames(such as 10s per frame)and reconstructed with a coarse image matrix.Through dynamic reconstruction,we could find out the time intervals during which motion happens.The listmode data acquired in these time intervals would not be used.Secondly,one of the frames is selected as the reference.To simplify the problem,the three-dimensional motion is decomposed into three two-dimensional motions.That is,the motion in tangential,sagittal and coronal directions.The two-dimensional registration is achieved by maximization of the mutual information of two reconstructed frames.Additionally,to ensure accuracy of the estimated correction parameters,the registration is repeatedly performed at different image slices from the same frame and their weighted average is used as the final result.Thirdly,since each LOR is uniquely determined by four parameters(s,φ,z,θ),we generate four lookup tables,each of which corrects one parameter during the image reconstruction process.Results:To demonstrate the feasibility and the accuracy,the performance of our method was verified by two clinical datasets both of which had obvious head motion.Results showed that the proposed method could well detect the time intervals when the motion happened and correct the motion blurring.Conclusions:The correction method could detect the time intervals when the motion happened and correct the motion blurring.It reduces the motion artifacts in PET image reconstruction.