OSEM重建算法及其改进算法的研究和比较

Research and comparison on OSEM and its improved reconstruction algorithms

使用OSEM算法重建CT图像,当子集水平选择较大时,重建图像收敛速度快,但会随迭代次数的增加发散;当子集水平选择较小时,重建图像收敛速度慢,图像的高频信息会丢失。为此提出将重建SPECT图像的Count-Regulated OSEM算法(CROSEM)应用于CT图像的重建。对比CROSEM算法、OSEM算法和子集序列EM算法(SSEM)对Sheep_Logan模型和实际的固体火箭发动机模型的重建结果,对比重建图像的收敛速度、重建图像的质量以及算法子集水平选择的不定性。实验结果表明,CROSEM算法相比于其它两种算法,重建图像的收敛速度更快、质量更好,子集水平是一个固定值,其具有更好的实用性。

When using OSEM algorithm to reconstruct CT image, different OS levels have different effects on reconstructed results. If the OS level is large, the reconstructed image has high convergence speed, but with the increase of the number of itera- tions, the reconstructed image radiates. If the OS level is low, the reconstructed image has low convergence speed, and the high frequency information of the image losses. To solve these problems, a Count-Regulated OSEM algorithm applied to reconstruct SPECT image was proposed to reconstruct CT image. The Sheep _ Logan model and the real solid rocket engine model were re- constructed separately using CROSEM, OSEM and subset sequence EM （SSEM）. The convergence speed and the quality of re- constructed images, and the way to choose the OS level in the process of reconstruction were compared. Experimental results demonstrate that, compared with the other two algorithms, the reconstructed image using CROSEM has higher convergence speed, and its quality is better, moreover, its OS level is a fixed value, so CROSEM is more practical.