Abstract: For improving the quality of positron emission tomography (PET) images, this paper proposes a new Bayesian image reconstruction algorithm by combining anisotropic median-diffusion filter with median root prior algorithm. Iterations of the proposed method can be divided into two steps: firstly, suppressing noise with the anisotropic median-diffusion filter; secondly, reconstructing image with median root prior algorithm. Simulation experiment  results present that the proposed algorithm can effectively suppress noise and accurately preserve edges information in PET image reconstruction. Furthermore, in comparison to other similar reconstruction algorithms, the proposed method absorbs the advantages of the anisotropic median-diffusion filter and is less sensitive to the selection of the image    gradient threshold and diffusion number, thus making the application of PET image reconstruction feasible.

Key words: positron emission tomography (PET), anisotropic median-diffusion, median root prior, image reconstruction, suppressing noise

摘要： 为了有效提高正电子发射断层成像图像的质量，通过把各向异性中值扩散滤波器融合到中值根先验算法中，提出了一种新的基于Bayesian理论的图像重建算法。新算法的每次迭代过程都可以分为两步：首先用各向异性中值扩散滤波器抑制重建图像中的噪声；然后用中值根先验算法重建图像。仿真实验结果表明，在正电子发射断层成像中，新算法不仅能有效地抑制噪声，还能精确地保护图像的边缘。此外，与其他类似算法相比，新算法吸收了各向异性中值扩散滤波器的优点，在迭代过程中对梯度阈值和扩散次数不敏感，易于实现，实用性强。

关键词: 正电子发射断层成像（PET）, 各向异性中值扩散, 中值根先验, 图像重建, 抑制噪声