基于系统建模的低剂量CT重建研究

发布时间：2018-04-17 08:38

本文选题：低剂量CT + 先验图像　；参考：《南方医科大学》2014年博士论文

【摘要】：X射线计算机断层成像(CT)以其高的时间分辨率、空间分辨率及对比度分辨率成为现代影像学的杰出代表,在临床诊断和治疗中广泛使用。但过量照射X光可诱发癌症、白血病或其他遗传性疾病,因此CT的剂量安全问题成为业界关注的焦点。如何以最小的代价和最低的X射线剂量获得最佳的CT图像质量,已经成为临床最为迫切的需求。降低X射线剂量的方法有多种,如通过降低管电流和缩短曝光时间来降低X射线球管的毫安秒(mAs)或者减少扫描角向采样数目等直接性的措施来降低X谢线剂量。但与此同时,上述扫描模式下解析重建的图像质量将严重退化,图像中的噪声和伪影影响了临床诊断和应用。近年来,针对低剂量CT图像优质重建的研究方兴未艾,主要集中于对解析重建的低剂量CT图像进行后处理滤波、对低剂量CT投影数据进行恢复,以及包括代数迭代和统计迭代的迭代重建算法。其中,针对迭代重建算法的研究尤其热烈。相对于解析重建算法,迭代重建算法在CT固有的物理系统建模方面具有较大的优势,如可精确模拟系统成像几何,X线多能光谱、线束硬化、散射、噪声等。因此,迭代重建算法能保持或者提高图像空间分辨力的同时抑制伪影和噪声。进一步的,统计迭代重建算法能基于光子统计学建立多种更精确的噪声模型,在降低图像噪声方面有更佳的表现。常规的,在迭代重建过程中,准确建模投影数据的测量方程是得到优质重建的前提和基础,而先验知识的合理利用对于求解的稳定性及获得高精度重建图像具有非常重要的意义。目前,在疾病的临床诊断和治疗过程中,常用到反复的CT扫描,如在灌注CT成像、4D-CT成像、CT图像引导的活体组织穿刺检查以及图像引导放疗等过程中。以图像引导放疗为例,除计划阶段CT扫描外,在整个放射治疗阶段,在分次治疗前需对患者进行CBCT扫描来进行定位,在此情况下,患者所接受的累加辐射剂量较常规CT检查扫描将会很高,反复CT扫描增加了罹患肿瘤的风险。在上述反复CT扫描过程中,对于同一个病人,先后两次CT扫描的图像之间除了几何位置和某些不自主运动引起的不同外,大部分解剖结构是相同的。换句话说,两次扫描获取的CT图像之间存在大量的结构冗余信息。传统的针对低剂量CT的研究方法仅仅引入图像自身的局部邻域的约束作为先验信息,未能考虑同一病人先前扫描获取的图像所能提供的先验信息来导引当前CT的图像的重建。鉴于此,本论文针对基于先前图像的先验信息的合理引入、先验信息引入后算法的优化等问题进行了深入的研究,同时本文对基于平板探测器的CBCT(Cone BeamCT)成像系统的噪声特性进行了进一步实验分析和验证,为后续迭代重建建立了理论和系统模型基础。归纳起来,本文的主要工作有： 1)为合理的引入先验图像信息,同时突破传统先验单纯依赖于目标图像局部邻域内的像素灰度信息的约束,本文利用非局部的思想通过在固定大小的搜索窗内检测基于图像块匹配的像素相似性而形成正则化项来实现基于先验图像的先验信息的有效引入。新方法在能够引入先验图像信息的同时弱化了对当前待重建目标图像和先验图像间配准精度的要求。数值仿真,物理体模以及临床数据实验表明,新方法能够提高当前低剂量图像的重建质量,同时不引入伪结构信息。 2)针对脑灌注CT成像的两大特点：(1)灌注扫描前的平扫描标准剂量图像的分辨率高且噪声低,可以为低剂量灌注序列增强图像的重建提供极为丰富的形态结构冗余信息；(2)对同一感兴趣层面反复扫描所得的增强序列图像除增强信息外,各图像之间的相对形变量较小；本文提出一种标准剂量平扫描图像导引的低剂量CT脑灌注序列图像重建方法。新方法首先利用利用标准剂量平扫描图像丰富的冗余信息实现基于改进的非局部均值算法的低剂量增强扫描图像的优质恢复,然后对恢复后的图像做去卷积迭代重建。数值仿真体模和病人数据实验表明,该算法能有效抑制图像中噪声,提高图像信噪比,从而使得相应血流动力学参数的计算更加准确。 3)为解决先验图像和当前待重建图像之间的结构不匹配问题,本文提出了一种利用当前测量所得投影数据与先验图像配准的方法获取与当前待重建图像相近或类似的先验图像用于基于先验图像的CBCT少角度重建。具体来说,本文算法是通过当前测量投影数据与先验图像在当前成像几何下的前向投影数据的匹配来实现图像域变形场的估计,然后将该变形场作用于先验图像来获取配准的先验图像,用于基于先验图像的重建策略中。特别的,本文采用的重建算法为PICCS(Prior Image Compress Constrained Sensing)算法。本文提出的算法能够有效避免图像空间配准时FDK算法重建的图像中噪声和条形伪影对配准精度的影响。XCAT仿真实验以及临床数据实验结果表明,本文提出的方案能够获取优质先验图像,重建结果优于传统PICCS算法。 4)鉴于CBCT成像系统中平板探测器与常规扇形束CT探测器物理设计上的不同,本文实验性的针对瓦里安TrueBeam系统中的平板探测器采集信号进行噪声相关性研究。通过对固定角度下反复测量的的探测数据的分析得出投影数据在探测器单元之间存在噪声相关性。实验结果表明,二维探测器八邻域内的噪声相关性是明显可见的,且一阶邻域内的相关系数明显高于二阶邻域内相关系数。同时,该噪声相关性与扫描所用X线剂量无关。本文将该噪声相关性应用于基于投影数据噪声模型的惩罚加权最小二乘恢复算法中,重建图像结果表明,该噪声相关性的引入能够提高重建图像质量。该实验研究为CBCT图像重建提供了更精准的系统建模。
[Abstract]:X - ray computed tomography (CT) with its high temporal resolution, spatial resolution and contrast resolution has become an outstanding representative of modern imaging, widely used in clinical diagnosis and treatment. But excessive irradiation of X light can cause cancer, leukemia or other genetic diseases, so dose CT security issues become the focus of attention of the industry. How to minimize the cost and the lowest CT X ray dose obtained the best image quality, has become the most urgent demand for clinical.
Method of reducing the X radiation dose has a variety of, such as by reducing the tube current and shorten the exposure time to reduce the X ray tube milliamperemeter (mAs) or reduce the scan angle to the number of sampling direct measures to reduce X Xie line dose. But at the same time, the image quality of the scanning mode of analytical reconstruction will be severely degraded that image noise and artifacts affect clinical diagnosis and application. In recent years, the study of low dose CT image quality reconstruction is just unfolding focuses on image analysis, the reconstruction of low dose CT postprocessing filter for low-dose CT projection data recovery, and iterative reconstruction algorithms including algebraic and statistical iteration. Among them, research on the iterative reconstruction algorithm is particularly warm. Compared with the analytical reconstruction algorithm, iterative reconstruction algorithm has more advantage in the modeling of physical systems inherent in the CT, such as fine Indeed the imaging geometry simulation system, X-ray energy spectrum, beam hardening, scattering, noise and so on. Therefore, while suppressing noise artifacts and iterative reconstruction algorithm can maintain or improve spatial resolution of the image. Further, statistical iterative reconstruction algorithm can establish a more accurate noise model based on photon statistics, have better performance to reduce the image noise. The conventional iterative reconstruction, in the process, the measurement equation of projection data is obtained accurately modeling the premise and foundation of high quality reconstruction, and the rational use of prior knowledge is very important for solving the stability and achieve high precision of image reconstruction.
At present, the clinical diagnosis and treatment of disease, commonly used to repeated CT scan, such as perfusion CT imaging, 4D-CT imaging, biopsy examination and image guided radiotherapy CT image guidance. In image guided radiotherapy as an example, in addition to planning CT scan, treatment stage in the whole radiation. In the time before treatment for patients with a CBCT scan to locate, in this case, the cumulative radiation dose received by the patients compared with the conventional CT scans will be high, repeated CT scanning and raises the risk of cancer. In the repeated CT scanning process, for the same patient, has between two CT scan images in addition to different geometry and some involuntary movements caused by the most anatomical structure is the same. In other words, there is a large amount of redundant information structure between CT image two scans obtained for traditional. Study on the method of low dose CT only into the local neighborhood of the image itself as constraint of prior information, failed to consider the image of the same patient previously scanned can provide information to guide the current CT image reconstruction. In view of this, this paper introduced the previous image based on prior information, in-depth study on the optimization of a priori information after the introduction of this paper on the flat panel detector based on CBCT (Cone BeamCT) the noise characteristics of the imaging system for further analysis and experimental verification, based on model theory and system for the subsequent iterative reconstruction. To sum up, the main works of this paper are:
1) for the a priori information of the image, pixel information and break through the traditional a priori simply rely on the local neighborhood within the constraints of the target image, using non local thought through the detection of pixel image block matching based on the similarity and the formation of a regularization term to achieve effective introduction of the prior image based on prior information in the search window fixed size. The new method weakens on the reconstructed object image and prior image registration accuracy requirements can be introduced in the prior image information at the same time. The numerical simulation shows that, the physical phantom and clinical data of the experiment, the new method can improve the image quality of low dose, while not introducing pseudo structural information.
2) according to the two characteristics of CT perfusion imaging: (1) high dose image scan standard perfusion scanning before the resolution and low noise, can provide abundant information for the redundant structure of low dose perfusion sequence to enhance image reconstruction; (2) to enhance the sense of the same image sequence in addition to enhance information the outer layer of interest face repeatedly scanned, relative deformation is smaller between each image; this paper presents a scanning image guided flat standard dose of low dose CT cerebral perfusion image sequence reconstruction method. The new method first use standard dose scan image redundant information rich can enhance the quality of low dose improved scanning image restoration the non local means algorithm based on image restoration, and then after the deconvolution iterative reconstruction. Numerical simulation shows that the phantom and patient data experiments, this algorithm can effectively suppress the noise,. The high image signal to noise ratio (SNR) makes the calculation of the corresponding hemodynamic parameters more accurate.
3) to solve the prior image and the current image to be reconstructed structure mismatch between the problem, this paper proposes a method of using the measured projection data and prior image acquisition and image registration prior the image to be reconstructed or similar to the prior image CBCT based on small angle reconstruction. Specifically, the algorithm is through the estimation of the current measurement of projection data and prior image in the image geometry prior to projection data, to realize the image deformation, and the deformation field to the prior image to obtain the prior registration for image based on prior image reconstruction strategy. In particular, the reconstruction algorithm used in this paper is PICCS (Prior Image Compress Constrained Sensing) algorithm. This algorithm can effectively avoid image space registration FDK reconstruction algorithm in noise and strip The influence of artifact on registration accuracy is demonstrated by.XCAT simulation experiment and clinical data experiment. The results show that the proposed scheme can get high-quality prior images, and the reconstruction result is better than the traditional PICCS algorithm.
4) in view of the flat panel detector CBCT imaging system and the conventional fan beam CT detector on different physical design, this paper studies the noise correlation experimental system for Varian TrueBeam flat panel detector signal acquisition. Through repeated measurements of the fixed angle of the detection data analysis shows the presence of noise in the projection data correlation between detector unit. The experimental results show that the noise correlation in the neighborhood of the two-dimensional detector eight is visible, and the correlation coefficient of first order in the neighborhood was higher than the two order neighborhood correlation coefficient. At the same time, the noise correlation with scanning X-ray dose related. The noise correlation applied to the penalized weighted least-squares projection data based on the noise model. The reconstruction of image restoration algorithms, the results show that the introduction of the noise correlation can improve the quality of image reconstruction. The experimental research The study provides a more accurate system modeling for CBCT image reconstruction.

【学位授予单位】：南方医科大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：R814.2

【参考文献】