欠采样条件下的CT图像重建算法研究

发布时间：2018-04-01 12:09

本文选题：低剂量CT　切入点：图像重建　出处：《南方医科大学》2017年硕士论文

【摘要】：计算机断层成像(Computed Tomography,CT)是现代医学诊断及放射治疗的重要技术手段,然而有研究表明,过高剂量的X射线照射会危害受检者健康。在保持现有硬件设备不变情况下,可通过稀疏角度及有限角度扫描来有效降低辐射剂量,但这两种扫描方式皆因数据采样不足而会使重建图像质量发生严重退化。因此,如何在降低辐射剂量的同时确保重建图像质量不退化已成为当前放射成像领域关注的焦点。本文通过系统回顾CT成像理论基础及对经典重建算法进行学习与掌握,针对低剂量CT成像中的稀疏角度及有限角度图像重建问题展开深入研究,并提出以下三种重建算法:第一,针对非局部平均(Non-local Means,NLM)算法在稀疏角度重建的不足,改进并实现了一种基于自适应NLM约束的CT重建算法。该算法定义了一种新型相似窗,并设计一种基于相似窗旋转变换的相似性测度,从而精确衡量像素间的相似性,避免图像细节模糊。其次,设计了一种基于像素梯度大小及当前迭代次数的自适应滤波参数,从而在抑制噪声的同时锐化图像边缘。仿真及真实数据实验结果表明,新算法可以有效消除噪声与伪影并且保持图像的边缘特性。与传统算法相比,信噪比提高了 38%,平均误差则降低了 76%。第二,针对 NLM 约束的代数重建算法(Algebraic Reconstruction Technique,ART)在稀疏角度重建的过平滑问题,改进并实现了一种基于旋转不变性的自适应NLM重建算法。该算法设计了一种基于旋转不变性的相似性测度用于计算相似窗间的相似距离,与此同时,根据像素所处位置特征及重建图像的噪声大小自适应调整滤波参数,从而避免图像边缘区域被过度平滑。仿真及真实数据实验结果表明,本算法可以在平滑噪声和伪影的同时,还原图像的边缘结构细节。与传统ART-NLM算法相比,信噪比提高了 51%,平均绝对误差降低了 74%。第三,针对有限角度CT图像重建存在严重伪影问题,改进并实现了一种基于局部和非局部正则化的有限角度CT图像重建算法。通过改进传统NLM算法以利用非伪影区域正确的图像信息来恢复伪影区域的像素值,然后改用全变差最小化方法校正由非伪影区域引入的伪结构信息。仿真及真实数据实验结果表明,本算法可以大幅度减少重建图像的几何失真伪影,显著提高图像质量。与ART-NLM算法相比,信噪比提高了 57%,平均绝对误差则降低了 52%。本文针对稀疏角度及有限角度CT图像重建问题,改进并实现了三种重建算法,显著提高了重建图像质量。虽然本文所研究的内容在图像精度上取得了一些初步成果,但对于在实际临床应用还有待更深一步的探究。
[Abstract]:Computed Tomography (CTT) is an important technique in modern medical diagnosis and radiotherapy. However, some studies have shown that high doses of X-ray irradiation can harm the health of the patients.Under the condition of keeping the existing hardware equipment unchanged, the radiation dose can be effectively reduced by sparse and limited angle scanning. However, the quality of reconstructed image can be seriously degraded by these two scanning methods due to insufficient data sampling.Therefore, how to reduce the radiation dose while ensuring that the reconstructed image quality does not degenerate has become the focus in the field of radiography.By reviewing the theoretical basis of CT imaging and learning and mastering the classical reconstruction algorithms, the sparse and finite angle image reconstruction problems in low dose CT imaging are studied in this paper.Three reconstruction algorithms are proposed as follows: first, a CT reconstruction algorithm based on adaptive NLM constraints is improved and implemented to overcome the deficiency of non-local average non-local mean mean (NLM) algorithm in sparse angle reconstruction.Secondly, an adaptive filtering parameter based on pixel gradient and the number of iterations is designed to reduce the noise and sharpen the edge of the image.Simulation and real data experiments show that the new algorithm can effectively eliminate noise and artifacts and preserve the edge characteristics of the image.Compared with the traditional algorithm, the SNR is increased by 38 and the average error is reduced by 76.Secondly, an adaptive NLM reconstruction algorithm based on rotation invariance is improved and implemented to solve the problem of over-smoothing in sparse angle reconstruction of Algebraic Reconstruction Technique ART (algebraic reconstruction algorithm) with NLM constraints.In this algorithm, a similarity measure based on rotation invariance is designed to calculate the similarity distance between similar windows. At the same time, the filtering parameters are adjusted adaptively according to the location characteristics of pixels and the noise size of the reconstructed image.Thus, the edge area of the image is not over-smoothed.Simulation and real data experiments show that the proposed algorithm can restore the edge structure details of the image while smoothing noise and artifacts.Compared with the traditional ART-NLM algorithm, the SNR is improved 51%, and the average absolute error is reduced 74%.Thirdly, aiming at the serious artifact problem in finite angle CT image reconstruction, an algorithm of limited angle CT image reconstruction based on local and non-local regularization is improved and implemented.The traditional NLM algorithm is improved to recover the pixel value of the artifact region by using the correct image information of the non-artifact region, and then the pseudo-structure information introduced by the non-artifact region is corrected by the method of total variation minimization.Simulation and real data experiments show that the proposed algorithm can greatly reduce the geometric distortion of reconstructed images and improve the image quality significantly.Compared with the ART-NLM algorithm, the SNR is increased by 57 and the mean absolute error is reduced by 522.In order to solve the problem of sparse and finite angle CT image reconstruction, three reconstruction algorithms are improved and implemented in this paper, and the quality of reconstructed images is improved significantly.Although some preliminary achievements have been made in the image accuracy of the content studied in this paper, it needs to be further explored in the practical clinical application.
【学位授予单位】：南方医科大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R814.42;TP391.41

【参考文献】