Mean Curvature Diffusion Method for PET Image Processing

• Main Authors: Yin-Chiao Tsai, 蔡銀嬌
• Other Authors: Kang-Ping Lin
• Format: Others
• Language: en_US
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/43197194597925985414

碩士 === 中原大學 === 電機工程研究所 === 90 === Positron Emission Tomography (PET) is a tomographic technique to display metabolic activity in slices through a patient''s body. The popular reconstruction methods today in PET are Filtered Backprojection (FBP) and iterative reconstruction algorithm. FBP is based on a Fourier Transform algorithm and is extremely fast, but the reconstructed image may suffer from annoying streak artifacts. Iterative reconstruction, like Maximum Likelihood-Expectation Maximization (ML-EM) algorithm, depresses the noise problem, but the algorithm is iterated too long, such that the reconstructed image starts to degrade. In this paper, Mean Curvature Diffusion (MCD), a nonlinear filtering technique, will be applied in the processing of the Filtered Backprojection reconstruction. The Mean Curvature Diffusion approach not only can depress noise but can also reserve the outlines of tissues. Using the combination of Mean Curvature Diffusion and Filtered Backprojection methods, a reconstructed PET image of good quality can be obtained quickly. In our study, the effect of MCD filtering in depressing the noise of PET image was investigated. The filtering of Mean Curvature Diffusion is applied to both the projection image (sinogram) prior to reconstruction and to the FBP reconstructed image. Preliminary studies on simulating target (phantom) of “sphere-ellipsoid”, we can rapidly get a good reconstructed image by using FBP reconstruction technique and the later MCD filtering method.