【作者】 惠成煊； 盛玉霞； 柴利； 郭诗尧； 熊丹；

【Author】 HUI Chengxuan;SHENG Yuxia;CHAI Li;GUO Shiyao;XIONG Dan;School of Information Science and Engineering,Wuhan University of Science and Technology;

【机构】 武汉科技大学信息科学与工程学院；

【摘要】 正电子发射断层成像(Positron Emission Tomography,PET)在病理诊断上具有重要作用。为了减少病人在放射性药物下的暴露时间,往往需要降低放射性同位素药物的使用量。低用量导致较低的光子计数量,从而使得PET重建图像分辨率非常低。针对此问题,本文对最大似然期望最大化方法(Maximum Likelihood Expectation Maximization,MLEM)进行改进,提出了一种新的动态PET图像时空域重建方法。本文方法在MLEM迭代之前利用PET数据时域和空域先验特征之间的相似性构建时域和空域马尔可夫矩阵;在每次MLEM迭代后首先计算马尔可夫矩阵的最优幂次,然后使用高次幂马尔可夫矩阵对迭代后的数据进行修复,接着将修复后的数据进行下次MLEM迭代,重复多次来提升图像重建效果。在仿真实验和活体数据实验中,与MLEM后滤波方法和最新的核重建方法相比,本文方法有着更好的重建效果。更多还原

【Abstract】 Positron emission tomography(PET) forms images of functional information in the human body,which has broad application in pathological analysis.In order to reduce the concerns of potential risk under radiation exposure,it is necessary to reduce the amount of radio-tracer used.However,low dose usage often results in low counting statistics,which reduces the resolution of the reconstructed image.In this paper,we modify the Maximum Likelihood Expectation Maximization algorithm(MLEM) and propose Markov matrix-based method to solve this problem.Before MLEM iteration,a temporal and a spatial Markov are computed based on spatiotemporal similarities among prior data;after each MLEM iteration,firstly we compute the optimal power of both temporal and spatial Markov matrix,then we multiply the iteration data by the optimal power of both spatial and temporal Markov matrix and introduce the processed data to the next MLEM iteration,finally we repeat this process to achieve overall denoising effect.In both simulation and in vivo data experiments,compared with MLEM with Gaussian post-filtering and the recently developed kernelized MLEM method,the proposed method can achieve better reconstruction performance.更多还原