The Research And Optimization Of PET Image Statistical Iterative Reconstruction Algorithm

Positron emission tomography is the most advanced detection technology in nuclearmedicine today, has been a hot research topic in the field of medical imaging. PET principleis that a certain amount of radionuclide injected into the human body decay and producepositrons. The negatrons of human body annihilate with positrons, generating a pair ofphotons opposite direction. Using the annular detectors to count the photons, we can get thedensity distribution of the radionuclide of human body, can thereby identify whether thedetected organ have lesions, cancer or other information.This paper describes the principle of PET imaging firstly, and then discusses severalclassic PET reconstruction algorithms which mainly divided into two categories: analyticmethod and the iteration method. The classic analytical method is filtered back projectionalgorithm, has small amount of calculation and fast imaging speed advantage, but cannoteffectively suppress noise. Iterative algorithm considers the physical characteristics of theimaging systems and the statistical model of observation data in the iterative process, so thatthe reconstruction results are better than the analytical method. Classical statistical iterativealgorithm has maximum likelihood estimation, least-squares algorithm, maximum aposteriori method based on Bayesian theory and so on. For statistical iterative algorithmconvergence speed is slow, it is also proposed a variety of accelerated algorithm, such as theHudson and Larkin proposed the classic ordered subsets expectation maximization algorithm,the broad space update the expectation maximization algorithm (SAGE) and Byrne etc.proposed expectation maximization algorithm based on block iteration (BI-EM) etc..The classic ordered subset expectation (OSEM) algorithm can speed up the convergencerate, but after a certain number of iterative reconstruction quality will decline. To solve thisproblem, this paper studies an ordered subsets statistics iteration accelerated algorithm based on wavelet. Multi-resolution technology applied to each subset of OSEM, is designed tosuppress noise, while stable solution process. The experimental simulation results show thatthis method not only overcome the shortcomings of image degradation, and still has theadvantages of speed up the convergence speed. The experimental data also proved that notonly can achieve better visual effect and still achieve a higher signal-to-noise ratio.The classical MLEM algorithm not only has slow convergence speed, but also cannoteffectively suppress noise. Therefore, usually join the regularization to improve the MLEM ofthe reconstructed image in the iterative process. This paper presents a new effectivede-noising algorithm based on wavelet shrinkage and anisotropic diffusion, the algorithm andMLEM method are combined to form a high quality reconstruction method, and applied tothe PET reconstruction. Experimental results show that the algorithm can obtain a highersignal-to-noise ratio and better image visual effect.Finally, this paper studies a median priori PET image reconstruction algorithm based onde-noising mixed model. Maximum a posteriori algorithm based on Bayesian by introducingthe constraints of the prior distribution of the image, which can effectively improve the imagequality, but not suitable prior distribution will lead to reconstructed image over smooth orstepped edge artifacts. In the median prior distribution of PET image reconstruction algorithm,then using the combination of wavelet shrinkage and forward-and-backward anisotropicdiffusion filter, proposed an improved median prior MAP reconstruction algorithm. Theexperimental results show that the algorithm in terms of SNR, RMSE, CORR and imagevisual effect is improved greatly.