基于弦图恢复和字典学习的低剂量脑灌注CT优质成像

发布时间：2018-05-04 19:03

本文选题：去卷积 + PCT成像　；参考：《南方医科大学》2017年硕士论文

【摘要】：急性缺血性脑卒中(acute ischemic stroke,AIS),又名急性脑梗死,是临床上最为常见的脑血管疾病,这种脑循环障碍导致的局限性或者全面性神经功能缺损综合征,发病率高,是我国老年人致死、致残的重要原因。脑灌注CT(perfusion CT,PCT)是临床上广泛用于AIS患者的检查手段,但其对病变区域的连续动态扫描使患者承受极大的剂量辐射风险,甚至会诱发一些遗传疾病。如何降低扫描剂量的同时不影响诊断图像质量是当前CT领域亟待攻克的难题。当前,降低CT辐射剂量的方式有很多,在满足临床诊断需要的前提下尽量降低毫安-秒(mAs)是一个简单且经济有效的方法。但是降低mAs,探测器采集的光子数急剧减少,使得投影数据被大量的光子噪声污染,重建图像质量也随之严重受损,在图像中表现为大量的噪声和伪影。针对这种低mAs技术,为获取最后的高质量参数图,我们可由投影数据着手,恢复受噪声污染的弦图数据,或者使用迭代重建的替代方式重建高质量的PCT序列图像;也可以从PCT序列图像本身着手,设计各类滤波器进行图像后处理来恢复高质量PCT序列图像;甚至可以在去卷积估计定量参数中引入合适的正则化形式,直接获取临床诊断可用的血流灌注参数图。根据大量实验数据统计分析,CT投影数据具有一定的统计特性。根据投影数据的统计特性,建立相应的去噪模型,从源头上更好地去除噪声。另一方面,在初步重建的低剂量PCT序列图像中,我们可以观测到序列图像有着相同的背景信息,这部分固定的解剖结构在组织中对比剂浓度随时间变化的过程中基本保持不变,因此,我们可将PCT序列图像分为背景和增强部分来看待。观测某块组织灌注信息像素点CT值得变化(增强情况),不仅与单个像素点的CT值相关,更确切地说是一种区域效果,而非单个像素点小效果。将图像看作许多块来处理,更具有实际意义。根据以上信息,将主要工作归纳如下:(1)提出一种基于惩罚加权最小二乘(penalized weighted least-square,PWLS)投影数据恢复的低剂量脑灌注PCT成像方法。该方法充分考虑到脑PCT投影数据的统计分布特性,利用该特性进行建模,采用PWLS的方法进行数据恢复,利用高斯-赛德尔(Gauss-Seidel,GS)方式进行迭代求解。不仅如此,还在原始投影数据和PWLS恢复后的投影数据之间引入自适应加权处理,更好地恢复投影数据。本方法有以下几大优点:①充分利用投影数据的统计分布特性,对噪声分布了解更为深入,有助于噪声去除;②采用了 GS优化算法,迭代求解;③根据投影数据噪声的水平,引入自适应投影数据加权处理,使得重建结果更为准确。该方法恢复重建的结果较传统滤波形式的结果噪声程度更小。(2)提出一种基于双字典学习(couple dictionary learning,CDL)的低剂量动态脑灌注成像方法。该方法将PCT序列数据中的背景信息和增强信息分离开来,采用 K-SVD(K-singular value decomposition,K-SVD)字典学习的方式分别训练一个二维背景信息字典和三维增强信息字典,用以恢复低剂量PCT序列图像,进一步获取高质量的脑血流灌注参数图。本方法有以下大几优点:①充分考虑PCT序列图像的结构特点,将背景信息和增强信息区分处理;②考虑到低剂量PCT序列图像的噪声程度,在三维增强字典的训练中引入了常规剂量的PCT序列作为外援先验信息。由于适当地引入先验信息与时间信息,恢复的PCT序列较其他几种基于字典学习的结果更优。
[Abstract]:Acute ischemic stroke (acute ischemic stroke, AIS), also known as acute cerebral infarction, is the most common cerebrovascular disease, which is the most important cause of death and disability in the elderly in our country. Cerebral perfusion CT (perfusion CT, PCT) is the clinical cause. It is widely used in the examination of AIS patients, but continuous dynamic scanning of the lesion makes the patient bear a great risk of dose radiation and even induce some genetic diseases. How to reduce the dose of the scan without affecting the diagnostic image quality is an urgent problem in the current CT field. At present, the way to reduce the dose of CT radiation is in a way. Many, it is a simple and economical method to minimize mAs in the condition of meeting the needs of clinical diagnosis. However, the number of photons collected by the detector is reduced dramatically, and the projection data is polluted by a large number of photon noise. The quality of the reconstructed image is seriously damaged, and the image is shown to be a large number of noises in the image and a large amount of noise in the image. For this low mAs technology, in order to obtain the final high quality parameters, we can start from the projection data to restore the string map data contaminated by the noise, or to reconstruct the high quality PCT sequence images using the alternative method of iterative reconstruction, and can also start from the PCT sequence image itself, and design all kinds of filters for post processing to restore the image. The image of high quality PCT sequence is complex, and the suitable regularization can be used in the quantitative parameter of deconvolution estimation, and the blood perfusion parameters can be obtained directly from the clinical diagnosis. According to the statistical analysis of a large number of experimental data, the CT projection data have some statistical characteristics. On the other hand, in the preliminary reconstructed low dose PCT sequence images, we can observe that the sequence images have the same background information. This part of the fixed anatomical structure remains basically the same in the process of the contrast of the contrast agent in the tissue, so we can divide the PCT sequence image into an image. The background and the enhanced part are considered. Observing the pixel point CT of an tissue perfusion information is worth changing (enhancement). It is not only related to the CT value of a single pixel point, but also a regional effect rather than a single pixel. It is more practical to treat the image as a lot of blocks. The following are as follows: (1) a low dose cerebral perfusion PCT imaging method based on the penalized weighted least-square (PWLS) projection data recovery is proposed. This method takes full account of the statistical distribution characteristics of the brain PCT projection data, uses this characteristic to model, uses the PWLS method to restore the data, and uses Gauss Sede. The method of Gauss-Seidel (GS) is iteratively solved. Not only that, it also introduces adaptive weighted processing between the original projection data and the projected data after the PWLS recovery to better restore the projection data. This method has the following several advantages: (1) making full use of the statistical distribution characteristics of the projected data to understand the noise distribution more deeply and helps to help it. Noise removal; (2) using the GS optimization algorithm and iterative solution; thirdly, based on the level of the projection data noise, the adaptive projection data weighting is introduced to make the reconstruction results more accurate. The result of the reconstruction is smaller than the result of the traditional filtering. (2) a new type of dual dictionary learning (couple dictionary LEA) is proposed. Rning, CDL) the low dose dynamic brain perfusion imaging method. This method separates the background information from the PCT sequence data and the enhancement information, and uses the K-SVD (K-singular value decomposition, K-SVD) dictionary to train a two-dimensional background information dictionary and the three dimension enhancement information dictionary to restore the low dose PCT sequence diagram. For example, further obtaining high quality cerebral blood flow perfusion parameters. This method has the following advantages: (1) taking full account of the structural features of PCT sequence images and distinguishing the background information from the enhanced information; secondly, taking into account the noise level of the low dose PCT sequence image, the conventional dose PCT sequence is introduced in the training of the three-dimensional enhancement dictionary. Due to the introduction of prior information and time information, the restored PCT sequence is better than other dictionary based learning results.

【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】：R743.3;R816.1

【参考文献】