优化X射线相衬成像辐射剂量效率的研究

As a novel X-ray imaging method, the grating-based X-ray differential phase contrast (DPC) imaging method has attracted a lot of research interests. Unlike the conventional X-ray absorption imaging which generates the object attenuation information, the DPC imaging enables the detection of object refraction information. Studies have shown that DPC images are more sensitive in distinguishing different tissue types, for example, the normal soft tissues and tumors. One of its promising potential clinical applications is in breast cancer screenings, especially for patient with dense breast. This is because of the reduced sensitivity of the conventional absorption images in catching the abnormalities in the dense breast. By jointly using the conventional absorption information and the complimentary DPC information, one can achieve better sensitivity and specificity in detecting the tumors inside a dense breast. .Despite of these promising potentials, however, it is still a challenge to enhance the radiation dose efficiency for a future clinical DPC imaging system. There are two major causes for the reduced radiation dose efficiency: the increased image noise and the decreased signal accuracy. Studies have shown that the statistical DPC signal extraction method is able to reduce the DPC image noise, but the reason why it is so and the lowest noise level in theory are not clear by far yet. In addition, the radiation dose efficiency in DPC imaging could be improved if the moiré artifacts can be mitigated. However, there is no such efficient artifact correction algorithm available. Due to these above two facts, the aims of this projection are straightforward and contain two-folds: first, build up a theoretical framework to analyze and demonstrate the lowest noise level in DPC imaging; second, develop a new artifact correction algorithm based on the deep learning technique. By minimizing the DPC image noise and maximizing the DPC signal accuracy, as a result, the overall radiation dose efficiency for a future clinical DPC imaging system would be greatly improved.

X射线相衬成像方法能够探测物体的相位信息，极大的提高分辨物体内部不同组分的灵敏度，是一种重要的新型X射线成像方法。为将其应用到临床诊断中，仍需要解决系统辐射剂量效率低的问题。其中，相位信号噪声高、信号精度差是导致辐射效率低的两个主要原因。预研究表明：统计信号提取方法比传统方法能更好的降低相位信号噪声，但其降噪机理和最小噪声极限尚不明确；另外减少相衬图像莫尔伪影可以提高相位信号精度，从而提高辐射剂量效率，但尚无较好的莫尔伪影校正算法。为此，本项目拟：（1）基于光子涨落的统计模型，从理论上研究相衬信号最小噪声极限，确立最优的统计相位信号提取方法，降低相衬图像噪声水平；（2）利于深度学习方法开发新型图像域相衬图像莫尔伪影的校正算法，提高相衬图像信号精度。本项目的研究成果将为降低相衬图像噪声提供理论依据，为提高相衬图像信号精度提供算法支持，最终实现优化现有X射线相衬成像系统辐射剂量效率的目标。

X射线光栅相衬成像方法能够探测物体的相位信息，可以极大的提高对物体内部组分的分辨能力，是一种重要的X射线成像方法。为将其应用到实际临床医学诊断中，本项目重点解决X射线光栅相衬成像系统辐射剂量效率低的问题。其中，相位信号噪声高、精度差是本研究主要关注的因素。通过本项目的研究，解决了理论和实验两个方面的问题，取得了如下三个方面的重要结果：（1）建立了基于光子涨落的统计模型，通过Cramér-Rao下界法求解了相位信号噪声的最小理论表达；（2）依据理论得到的相位信号最小噪声极限，研究确立了最佳的相位信号提取方法，降低了相衬图像噪声水平；（3）开发了新型基于深度学习技术的图像域相衬图像残余莫尔伪影的校正算法，提高了相衬图像信号精度。本项目的研究成果丰富了相衬成像理论，为降低相衬图像噪声提供了理论指导，同时有效提高了相衬图像质量，最终达到了优化系统辐射剂量效率的目标。