X射线真彩色CT图像重建研究

True-color CT (spectral CT /multi-energy CT) is an important development due to the merits of k-edge imaging and material decomposition in biomedical imaging. The most critical limitations with current true-color x-ray detectors are threefold: cost, size and radiation dose. Current true-color detectors are both small and expensive. Radiation dose is a concern for any x-ray-based system, and energy bins for each true-color channel require higher total exposure relative to a single-channel energy-integrating detector. Furthermore, the maximum detectable x-ray flux is limited by the detector readout speed (due to pulse pileup). In this proposal, we aim to develop an image reconstruction method to improve the image quality while reducing the cost and radiation for true-color CT. Our research is divided into two parts: one is the image reconstruction on each energy bin and the other is the color mapping. Our solutions are three folds. Firstly, the reconstruction on each energy bin is a low-count issue. However, the data on one energy channel is related to another one. Recently, a dictionary learning based sparsity constraint has been proved to perform well for low-count data, with preserving details and suppressing noise as much as possible. We propose to incorporate the correlations among multi-energy bins and the dictionary learning based constraint into the reconstruction on each energy bin. Second, interior tomography can reduce the detector size to get a ROI image. Since most applications on medical imaging focus on a small ROI, we propose to use interior scanning in true-color CT considering the limitation on true-color detector size and cost. The global gray-scale reconstruction can provide a good prior for interior tomography on each energy bin using dictionary learning technique. The third aspect of this proposal is to develop quantitative color-mapping method. The present color mapping method in true-color CT cannot make a direct relationship between the color and the material. In this proposal, we will investigate the physical meaning behind each principle component which is produced from the multi-energy reconstructions and work out a quantitative color mapping strategy.

使用光子计数探测器的X射线CT系统能同时采集多个能量通道的投影数据，进行彩色成像，称为光谱CT（Spectral CT）、多能谱CT（Multi-energy CT）、真彩色或彩色CT（True-color/color CT）。本项目围绕真彩色CT的图像重建问题展开研究。针对多能谱数据每个能谱通道上光子数少、噪声大的问题，提出使用基于字典学习等稀疏性约束的统计迭代重建与多能谱数据特性相结合的重建方法；针对大尺寸的光子计数探测器制造工艺难度大、价格高的问题，提出使用内部扫描方式，并提出在多光谱内重建中使用全局灰度重建结果作为先验约束的多光谱内重建方法；针对真彩色图像的彩色生成问题，提出将统计分析方法与成像物理学过程相结合的非线性彩色映射和定量彩色生成方法。通过本项目的研究形成一套完整的高性能真彩色CT图像重建方案，提高真彩色CT的成像质量、降低成像成本，为其实际应用打下基础。

传统CT在重建过程中考虑的是不同物质对X线的衰减不同，即X线衰减的物质依赖性。多能谱CT则利用多组不同能量下的投影数据，在考虑X线衰减的物质依赖性的同时还充分考虑了X线衰减的能量依赖性，能够进行彩色成像，是当前CT领域的一个研究热点。然而，多能CT目前面临着多方面的挑战：包括由于总剂量限制所带来的低剂量多能CT重建问题；由于探测器制造工艺和成本限制带来的多能内重建问题；以及如何充分利用多能信息进一步提高成像质量的问题等。在本项目中，我们通过对低剂量CT、内重建、双能CT等问题分别进行深入研究的基础上，结合多能数据的特点，解决上述问题，为进一步提高多能CT重建质量提供了解决方案。主要包括：.1..提出在统计迭代重建框架下，结合基于图像块的稀疏性和相似性约束来提高重建质量，其中，基于图像块的稀疏性和相似性约束分别通过基于字典学习的稀疏表示和非局部相似性方法来实现；对迭代重建涉及的参数选择问题进行了研究，提出了一种基于平衡准则的自适应参数选择算法；提出了一种将基于截断Hilbert变换的内重建与全变分最小化约束结合，通过基于截断Hilbert变换的内重建实现快速重建，并通过全变分最小化约束来有效抑制噪声的内重建方法。.2..提出将字典学习方法应用于双能CT中，分别从正常剂量的高、低能图像集中训练两组字典，将其应用于基于基本材料分解的双能CT迭代重建框架中；提出了一种双能CT混合内重建方法，对高电压数据进行内部扫描，通过投影补全和基于TV最小化约束来对这种混合双能内部扫描数据进行重建。.3..提出将字典学习的方法应用于多能CT重建中，从结合了所有能量通道的宽能投影数据的重建结果中训练字典，将该字典应用于每个能量通道的重建中提高重建质量；在此基础上，通过利用低秩和张量来描述不同能量通道之间各个图像的相关性，提出了结合字典学习和低秩张量的多能CT重建算法。