Study And Implementation Of Positron Emission Tomography Statistical Reconstruction Algorithms

Positron emission tomography (PET) is one of the most advanced technologies in thefield of nuclear medicine. It uses the radionuclide injected into the body for imaging，whichcan reflect dynamically the physiological and biochemical information of organism. It is thekey problem of positron emission tomography to how to reconstruct the high-quality images.So the reconstruction algorithm has always been the research focus.PET reconstruction algorithms are divided into analytical and iterative method. Theanalytical method is represented by filtered back projection. Its implementation method issimple, but the quality of reconstructed image is poor. The iterative method, which is dividedinto algebraic and statistical iterative method, adds the physical and demographic factors intothe reconstruction process. While the basic theory of algebraic iterative algorithm is linearequations, statistical iterative algorithm is optimal estimation. Because the imagereconstructed by statistical iterative algorithm which can more accurately reflect the PETimaging process is better, this study focuses on the statistical iterative algorithm. The main jobis:(1) Based on the method of coupled feedback and anisotropic diffusion, the maximum aposterior algorithm is improved. In new algorithm, the information of coupled feedbackmodel, which is used to update the projection data, come from the image denoised byanisotropic diffusion model.(2) To improve the poor image reconstructed by penalized exponential expectationmaximization algorithm, an improved denoising model is added into the algorithm. Based onthe kernel anisotropic diffusion, the new denoising model denoises image in the nuclear space,using the anisotropic diffusion method based on fuzzy non-local mean theory.(3) The bigger step size method is added to the maximum likelihood expectation maximization algorithm, to accelerate the convergence speed. At the same time, the newalgorithm filter images using bidirectional diffusion model. Experimental results show thatthe new algorithm can speed up the convergence as well as improve the image quality.