Research And Application About Image Reconstruction Of PEM

Early diagnosis and treatment plays an important role in reducing the mortality for breastcancer. Researches about PET indicate that it has an extremely high sensitivity to tracer what makesPET could realize imaging tumors in a high accuracy. However, the sensitivity of whole-body ishigh enough in detection of breast. Therefore, PEM is developed to detect breast cancer which isanother application of PET and it makes a big contribution in the development in imagingexamination of breast tumors. For the image reconstruction of PEM, the algorithm of that is soimportant that we make a real good study in the algorithms. The main work is as follows:(1)PET image reconstruction by filtered back projection method: In this chapter, we firstdescribe the foundation of filtered back projection method——Radon transformation and Fourierslice theorem, deduced reconstructing equation and simply describe the implementation process.Secondly, Shepp-Logan map is used in our stimulation experiments as a standard image that wereconstruct images from the data obtained by projecting the standard image in three differentinterval of angles including10o、5oand1o. Results show that the smaller interval makes the muchclearer images. And then, we researched the effect of window function in the reconstruction byadding0.01Poisson noise on the standard image and check the differences between imagesreconstructed by normal FBP and by the FBP algorithm with a window function. The results showethat after adding a window function, the noise of image is reduced. On the other hand, it turns outthat we could hardly recognize much more details from the image since the high-frequencycomponents has been eliminated by the window function. Overall, Filtered-Back Projection is notso good in reconstruction.(2)Image reconstruction based on iterative algorithm: This chapter introduces the basicknowledge of iterative algorithm including algebraic reconstruct technique (ART), maximumlikelihood expectation method (MLEM) and ordered subset expectation maximization (OSEM).Besides that analysis among these three algorithms indicate that OSEM is much more prominent.Further more, we take a study of the quality of imaging though the same projection data butdifferent amount of subsets which means the amount of projection direction are different. Resultsshow that it is much more reasonable in each subset containing10or less projection direction. Inaddition, for the actual data contains a lot of noise, we also studied the impact of the number ofiterations in the presence of noise on the image quality. The result shows that the optimal number ofiteration does exist but it depends on the noise in reconstruction.(3)Application of images reconstruction algorithm in PEM: In this chapter, breast model and mice tumor experiments are taken to verify the reasonability of those algorithm proposed above. Inthe experiment of breast model, single-slice rebinning is realized in the preprocessing and theimages are reconstructed by FBP and OSEM separately. Results show in this experiment the qualityof images reconstructed by FBP is much better. However, the artifacts are much less in imagesreconstructed by OSEM. In mice tumor experiment, three rebinning methods are all realized andcomplete the reconstruction. In one hand, for the method of rebinning, the quality of imagesreconstructed from data recombined by single-slice rebinning (SSRB) and Fourier rebinning (FORE)are much better than that by multi-slice rebinning (MSRB). On the other hand, for reconstructionalgorithms, it turns out that OSEM algorithm behaves better not only in noise suppression but alsoin the quality of imaging.