低剂量CT图像质量改善算法研究

发布时间：2018-11-29 12:41

【摘要】：X射线计算机断层扫描(Computed Tomography,CT)技术在农林业应用、工业无损检测、材料学以及医学诊断等领域发展迅速,特别是在临床医学领域发挥巨大作用。X射线辐射会对病患造成一定程度的伤害,并诱发癌症等疾病,因此如何在获取解剖信息清晰且密度分辨率高的重建图像的同时尽可能地减少CT辐射剂量已成为CT研究者们的奋斗目标。降低管电流是一种减少辐射剂量的有效方法,此方法会导致投影数据产生噪声,进而损失低剂量CT重建图像的质量。本文主要采用改进重建算法、对投影数据进行去噪以及对重建图像的噪声滤除三种方法进行噪声去除和伪影抑制,主要创新工作如下:1.为克服全变分算法容易造成阶梯伪影和过度平滑的问题,构造一个加权方差和图像梯度共同作用的边缘指示函数,该扩散函数与全变分(Total Variation,TV)模型结合得到基于加权方差的TV模型。进而,将新模型引入到惩罚加权最小二乘重建(Penalized Weighted Least Square,PWLS)算法中得到一种基于加权方差TV的统计迭代重建去噪算法。采用两步进行新模型的优化估计,首先,采用交替方向迭代法将联合问题分解为两个子问题,接着,分别采用梯度下降法和可分离抛物线替代法求解。通过视觉效果和量化指标分析,新算法重建图像质量在得到明显改善的同时边缘细节分辨率高。2.由于中值滤波不仅能够消除脉冲噪声,还可以较好地保留图像边缘,给出一种基于中值非局部先验的最大后验概率投影域滤波算法。该算法先对投影图像进行中值滤波,再根据图像块间的相似性进行自适应非局部降噪,采用高斯-赛德尔方法得到所提模型的最优解,最后经过滤波反投影(Filtered Back Projection,FBP)算法得到最终CT重建图像。采用修正后的大脑模型进行仿真实验,所提算法不仅在平滑投影图像噪声和抑制条形伪影方面表现良好,且可获得高信噪比图像。3.直觉模糊熵(Intuition Fuzzy Entropy,IFE)能够自适应地区分图像平坦区域和边缘细节区域,便将其与各项异性扩散模型的扩散函数共同作用得到一种基于IFE的边缘扩散函数。同时,采用新指示函数对广义全变分(Total Generalized Variation,TGV)模型改进得到新的自适应TGV正则化滤波模型。最后,通过一阶原始-对偶算法求解新模型获取最终重建图像。仿真模型和实际数据实验结果均表明新算法在抑制噪声和去除条状伪影方面表现突出,同时较好地保留低剂量CT复原图像的纹理结构特征。
[Abstract]:X-ray computed tomography (Computed Tomography,CT) technology is developing rapidly in the fields of agriculture and forestry, industrial nondestructive testing, materials science and medical diagnosis, etc. In particular, it plays an important role in clinical medicine. X-ray radiation can cause a certain degree of injury to patients and induce diseases such as cancer. Therefore, how to obtain the reconstructed image with clear anatomical information and high density resolution while minimizing the radiation dose of CT has become the goal of CT researchers. Reducing the tube current is an effective method to reduce the radiation dose. This method can cause the noise of the projection data, and then lose the quality of the low dose CT reconstruction image. This paper mainly adopts improved reconstruction algorithm, denoising projection data and noise filtering of reconstructed image for noise removal and artifact suppression. The main innovative work is as follows: 1. In order to overcome the problems of step artifacts and excessive smoothing caused by the total variation algorithm, a boundary indicator function with weighted variance and image gradient is constructed. The diffusion function is combined with the total variational (Total Variation,. TV) model combined with TV model based on weighted variance. Furthermore, the new model is introduced into the penalty weighted least square reconstruction (Penalized Weighted Least Square,PWLS) algorithm to obtain a statistical iterative reconstruction denoising algorithm based on weighted variance TV. The optimal estimation of the new model is carried out with two steps. Firstly, the joint problem is decomposed into two sub-problems by alternating direction iteration method, and then the gradient descent method and the separable parabola substitution method are used to solve the joint problem. Through visual effect and quantization index analysis, the reconstruction image quality of the new algorithm is improved obviously and the resolution of edge detail is high. 2. Because median filter can not only eliminate impulse noise, but also preserve image edge, a projection domain filtering algorithm based on median nonlocal priori is presented. The algorithm first carries on median filtering to the projection image, then adaptively non-local noise reduction according to the similarity between the image blocks. The optimal solution of the proposed model is obtained by using the Gauss-Seidel method. Finally, the filtered backprojection (Filtered Back Projection, is used. FBP) algorithm to get the final CT reconstruction image. The modified brain model is used for simulation experiments. The proposed algorithm not only performs well in smoothing projection image noise and suppressing bar artifact, but also can obtain high SNR image. Intuitionistic fuzzy entropy (Intuition Fuzzy Entropy,IFE) can self-adaptively distinguish the flat region from the edge detail region, and then work together with the diffusion functions of various anisotropic diffusion models to obtain an edge diffusion function based on IFE. At the same time, a new adaptive TGV regularization filter model is obtained by using a new indicator function to improve the generalized total variational (Total Generalized Variation,TGV) model. Finally, the first-order primitive-dual algorithm is used to solve the new model to obtain the final reconstructed image. Both the simulation model and the experimental results show that the new algorithm is very effective in noise suppression and strip artifact removal, while preserving the texture features of low dose CT reconstructed images.
【学位授予单位】：中北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】