【作者】 占杰；

【导师】 陈武凡；

【作者基本信息】 南方医科大学 ， 生物医学工程， 2009， 博士

【摘要】 正电子发射断层成像(Positron Emission Tomography,PET)是功能分子影像技术的杰出代表,PET借助扫描测量前注入活体内的放射性核素标记的示踪剂进行显像,够在分子水平上利用影像技术反映机体心脑代谢和功能,已经在肿瘤学,心血管疾病学和神经系统疾病学研究中,以及新医药学开发研究等领域中显示出它卓越的性能。PET成像的目的是通过放射性核素标记示踪剂分布得到感兴趣(region of interest,ROI)组织的生理参数,因而,如何在重建方法中建立生理参数与放射性示踪剂浓度动态分布之间的联系,从而在重建PET图像的同时,对相关生理参数进行估计,将是PET图像重建今后发展的方向。所以对于PET图像重建来说,有二个主要的研究方向:一方面是根据PET测量数据准确的重建示踪剂浓度分布图,即PET图像重建问题;另一方面是PET除了能定性提供组织器官的新陈代谢情况外,还可通过动态成像的方式定量分析具体的生理、生化过程,即PET参数估计问题。由于受到低计数率和一些物理噪声的影响,PET图像的重建问题在理论上是一个病态的问题。传统的滤波反投影(Filtered Back-Projection,FBP)重建方法虽然具有成像速度快的优点,但其重建图像却含有大量噪声,图像质量较差。最大似然期望最大法(Maximum-Likelihood Expectation-Maximization,ML-EM)能够针对系统模型的物理效应和探测数据和噪声的统计泊松特性建立数学模型,其重建的图像质量要优于传统的FBP方法。然而,单纯的传统ML-EM方法收敛速度较慢,而且在迭代过程中会产生质量退化的图像而导致的棋盘效应,从而导致非收敛的迭代过程。近年以来,基于马尔可夫随机场(Markov Random Fields,MRF)的贝叶斯(Bayesian)重建方法或者最大化后验估计(Maximum A Posteriori,MAP)的方法已经在包括PET图像重建在内的图像重建中得到广泛的应用。Bayesian方法通过正则化在迭代过程中引进放射性示踪剂浓度在空间上概率分布的先验信息,能够明显改善重建图像质量以及迭代过程的收敛性,该方法已被证明了其在理论上的正确性和实际上的有效性。虽然引入了图像先验信息的Bayesian方法能够在很大程度上改善迭代重建的效果,但是依赖于传统的局部空间邻域信息的Bayesian方法只能为重建提供有限的先验信息。而且在当前PET问题的研究中,通常将时间和空间分开来考虑。但在实际问题中,时间和空间是相辅相成、互相影响、互相作用的。PET测量数据中已经隐含了很多生理信息,但往往未能直接表示出来。传统的间接参数估计方法如加权最小二乘参数估计方法(Weighted Least SquareMethod,WLS),需要首先从测量数据重建出PET动态放射性活度图像,然后在这些时间上连续的重建图像应用适当的动力学模型上估计感兴趣的生理参数,然而PET重建图像中往往含有极大的噪声,这些噪声将影响参数估计的准确性。本文对于PET重建算法的研究工作同样也是基于PET研究中的两个主要研究方向,作者做了以下关于PET图像重建和动力学参数估计方面的工作:1,提出一种新的综合了QM先验、QP先验和QTO先验的MRF二次混合多阶先验模型,新的混合先验可依据目标图像不同位置的特性自适应的决定QM先验、QP先验和QTO先验的作用效果,其重建结果比单独使用QM先验、QP先验的效果好。同时也优于与传统的TV先验和中值根先验的重建结果。2,从引入更多先验知识指导PET图像重建角度进行研究,对采用不同邻域大小的常用先验和非局部先验进行大量对比实验。从实验结果中可以看到,简单的增大邻域大小是无法有效的结合空间大尺寸信息至Bayesian重建。另一方面非局部先验能够利用目标图像中大尺寸信息或全局信息,其对发射断层重建作用比传统的QM先验和Huber先验有效,而且鲁棒性好。3,提出一种新的基于二房室模型和空间邻域信息的时空先验模型,新的时空先验综合考虑时间和空间因素,根据探测器得到的sinogram数据直接求解出动力学参数,同时重建PET动态图像。实验中给出了新方法与传统方法的定量比较。模拟实验证明基于该时空先验的动态PET重建图像和动力学参数图像均要优于传统方法重建结果。试验中将以上三种方法应用于相应的PET断层发射成像和动力学参数估计中,相关试验分析表明:本文所提出的基于MRF和优化理论的新算法均能够分别在不同程度上提高PET图像重建和动力学参数估计的质量。更多还原

【Abstract】 Positron emission computed tomography (PET) is the effective medical imaging technique that provides functional information of physiological activity by displaying the concentration distribution of radioisotope labeled tracer (chemical compounds or biological molecular) which pre-injected into the human body before imaging process. PET is able to represent heart and brain metabolism and functions on molecule level by imaging techniques, and has shown great performance in oncology, cardiopathy, neurology and new medicine studies. The purpose of PET reconstruction is to get physiological parameters about the region of interest (ROI) through tracer activity distribution. So, it is the direction of development that PET activity image reconstruction and parameters estimate simultaneously through the relationship between physiological parameters and tracer activity distribution. There are two main research directions in PET reconstruction study: 1, accurate reconstruction static activity images of the radioactive tracer spatial distribution; 2, dynamic PET activity image reconstruction and precise physiological parameters estimate.But positron emission tomography is an ill-posed inverse problem because the observed projection data are contaminated by noise due to low count rate and physical effects. Though needing less computation cost, traditional filter back projection (FBP) method often reconstruct noisy images of low quality. Better expressing system models of physical effects and modeling the statistical poisson character of the data, the famous maximum-likelihood expectation-maximization (ML-EM) approach outperforms the FBP method with regard to image quality. However, pure traditional ML-EM approach suffers slow convergence and the reconstructed activity images always start deteriorating to produce "checkerboard effect" as the iteration proceeds.In recently years, Bayesian methods or equivalently MAP (Maximum A Posteriori) methods has been widely used in image reconstruction. Bayesian methods incorporate MRF prior information of objective tracer concentration distribution data into the ML-EM algorithm through regularization or prior terms and have been proved theoretically correct and practically effective compared to other methods. Compared to traditional ML-EM algorithm, Bayesian reconstruction shows a better performance in both improving convergence behavior and producing more appealing images. Bayesian reconstruction can greatly improve reconstruction by incorporating image prior information. However we also find that, heavily relied on the information within a limited neighborhood, conventional Bayesian methods can only contribute limit spatial local prior information to reconstruction.And at persent PET srudy, space and time are divided into two problems. In conventional indirected parameter estimate, such as Weighted Least Square Method (WLS), the changing activity of the injected radiotracer is conventionally measured through multiple consecutive PET image reconstructions. The image of the radioactivity distribution in each frame is reconstructed independently and the whole set of frames is then used to estimate the distribution of the physiological parameter of interest by the application of an appropriate pharmacokinetic model to the time radioactivity curve either of appropriately selected functional regions or of each image element. However, the noisy PET reconstruction image will influence the accuracy of parameters estimate.Our work on PET reconstruction is based on the two mian research direction in PET study: how to further improve the quality for activity image reconstructions and how to further improve accuracy of parameters estimate. We have done following work:1, proposing a noval Markov Quadratic Hybrid Multi-Order Priors which has the effects of QM prior, QP prior and QTO prior adaptively according to different properties of different positions in objective image effectively. The new MRF hybrid prior outperforms unitary QM prior and unitary QP prior.2, studing how to incorporate more prior knowledge to guide PET image reconstruction, a great quantity of experiments were carried out with common prior with different size and and non-local a priori a priori. From the analyses and experiments presented in this paper, we can see that just enlarging the sizes of neighborhood can not effectively incorporate more large-scale knowledge into Bayesian reconstruction. On the other hand, the nonlocal prior, which is devised to exploit the large-scale or global connectivity and continuity knowledge in the image, demonstrates a more effective and robust regularization for emission reconstruction than the conventional local QM prior and Huber prior.3, proposing a noval spatio-temporal prior based on two-tissue compartmental model and neighborhood information. Time and space factors are general take into account in the new spatio-temporal prior. Based on spatio-temporal prior, kinetic parameters and PET activity images could be estimated and reconstructed synchronous. And in the simulation experiment, activity reconstruction image and parameter estimation based on spatio-temporal prior has better quality than that with the conventional method.更多还原