Positron Emission Tomography Statistical Iterative Method And The Accelerated Method

Positron emission tomography (PET) is one of advanced noninvasive diagnositic tech-nique of modern medical science, and it acts as one of representative techniques of modern nuclear medical diagnosis. The basic fundamentals and image reconstruction methodology of PET are discussed in the paper. And the later is the main purposes of this dissertation. Gener-ally three reconstruction algorithms are discussed in the paper: PML image reconstructionmodel based on nonlocal means, fuzzy theory and anisotropic diffusion; PML approach base the gray variance of neighborhood pixels and anitropic diffusion; Stactical iteration recon-struction based on modified order subsets.The fundamentals and system structure of PET are introduced in the first part of the pa-per. Furthermore, PET image reconstruction approaches, which are classified into analytical and iterative method, are discussed detailly. The study of tomography reconstruction stems from the analytical method which is characteristic of simplicity and fastness. However, its re-constructed images are very noisy. Iterative algorithms can be separated into algebratic itera-tion method and statistical iteration method. The former one is based on linear equation theory, while the latter one is ground on some kinds of evaluation criterias which contains lots of methods such as least square algorithm, maximum likelihood expectation maximization algo-rithm, bayesian algorithm and so on. Due to introducing the physical properties of system and the stactistical model of projection data into the reconstruction, iterative algorithm can get better images than analytical method.Against defects of P-M model, the gray variance of neighborhood pixels is brung into the diffusion model. And a new method, which combines the proposed diffusion model with re-gularized MLEM algorithm, is proposed in PET image reconstruction. The results show expe-rimentally that it has some performance of improving quality of reconstructed images and the protecting images edges in comparison with other common methods.The method of PET image recontruction combined with anisotropic diffusion filter with median filter (PDEMedian) cannot protect the edges accurately as well as the step artifacts can be found obviously in the image reconstructed by PDEMedian due to the P-M diffusion model s poor properties. Such a novel diffusion model combined nonlocal means with fuzzy theory is introduced into PET image reconstruction. The feasibility and efficiency of the new method are verified experimentally in comparison with other traditional algorithms.Finaly, the dissertation discuses the accelerated algorithm for statistical iteration based on modified ordered subset. And the modified ordered subset, whose number of subset de-creases with iteration, is introduced into iterative method. At the same, the projected data are rebuilted by the method of multi-resolution. The simulation results show the proposed method acceralate convergent rate of algorithm, but not at the expense of the reconstructed image squality.