基于混合先验的CT图像重建研究

Research on CT image reconstruction based on mixed prior algorithms

在图像重建中,可以通过贝叶斯方法引入先验知识来抑制噪声提高重建的质量,提出了基于Gibbs先验和解剖中值先验混合的算法,该方法根据Gibbs先验分布和中值先验分布,先用基于Gibbs先验分布的算法多次迭代图像,再结合解剖结构先验,将以后每次迭代后的图像分割成不同的区域,将分割后的图像像素运用于中值先验。仿真实验表明,在相同条件和迭代次数下,该方法重建出来的图像更清晰,图像的相关系数比用Gibbs先验方法和中值先验方法重建出的图像的大。从实际实验数据上看,该方法重建的图像噪声更小,边缘更清晰。

In image reconstruction, the prior knowledge is introduced through Bayesian method to reduce noise and improve the quality of image reconstruction. In this paper, an algorithm is proposed, by which multiple iteration operation to images are done based on Gibbs prior distribution firstly, the images iterated every time are segmented into different regions combined with the anatomical prior technology secondly, and the pixel values of segmented images are used in the median root prior at last. Simulation results show that the new method can get much clearer reconstructed images and greater correlation coefficient of images than Gibbs prior or median root prior in the same conditions and iteration times. And the experimental data show that the noises of reconstructed images are much smaller and the edges are much clearer.