面向肺癌筛查的低剂量CT成像: 基于投影筛选与CT图像库学习的精准重建

Lung cancer is the leading cause of morbidity. Accurate early lung cancer screening it the key to reduce the mortality of lung cancer. Low-dose CT (LDCT) is the most effective way for early lung cancer screening of high-risk population. LDCT has the potential radiation risk of cancer. However, the CT projection is deteriorated because of the decreasing dose. It also leads to the high noise and severe artifacts of the reconstruction image itself, which directly affects the accuracy of early lung cancer detection. For the above problem, this project proposes a new LDCT imaging technique for early lung cancer screening. It focuses on solving the key problems of how to judge and correct the deteriorated projections and the poor reconstruction of the image details, realizing the precise lung reconstruction of LDCT. This technique firstly uses the threshold method to select and correct the deteriorated projection to remove their effect. Then from the lung image patches of high quality CT, a few patches similar to the target patch in LDCT are learned. Based on them, the regularization of the objective function is constructed. Finally, the noise model of projection is combined to solve the problem iteratively and the precise reconstruction of lung details will be realized. The study of this project not only provides more accurate reconstruction algorithm and system development for LDCT lung cancer screening, but also provides the application base to carry out the lung cancer screening and reduce the mortality of lung cancer for people at high risk in our country, therefore it has important research significance and social significance.

肺癌是致死率最高的恶性肿瘤，准确的早期肺癌筛查是降低其死亡率的关键。低剂量CT(LDCT)是实现早期肺癌筛查的最有效手段，然而LDCT因剂量降低而导致投影恶化，使得重建图像自身存在高噪声和严重伪影，直接影响肺癌的准确诊断。针对这一问题，本项目提出一种面向早期肺癌筛查的LDCT成像新方法，重点解决其中高污染投影的鉴别、校正和图像细节重建不佳的关键问题，实现LDCT肺部图像的精准重建。该方法首先使用阈值法对投影数据进行筛选和校正，以去除被污染投影的影响；然后基于高质量CT肺部图像库，通过学习训练出与LDCT肺部目标图像块相似的优选样本，以构建目标函数约束项；最后结合投影噪声模型对目标函数进行迭代求解，实现肺部细节保持的图像精准重建。本项目的研究，不仅为LDCT肺癌筛查提供更精准的重建算法和系统开发基础，而且为我国高危人群开展肺癌筛查、降低肺癌死亡率提供应用基础，因此具有重要研究意义和社会意义。

肺癌是致死率最高的恶性肿瘤，准确的早期肺癌筛查是降低其死亡率的关键。低剂量CT(LDCT)是实现早期肺癌筛查的最有效手段，然而LDCT因剂量的降低而导致投影恶化，使得重建图像自身存在高噪声和严重伪影，直接影响肺癌的准确诊断。针对这一问题，本项目提出一种面向早期肺癌筛查的LDCT成像新方法，重点解决其高噪声、图像细节重建不佳的关键问题，实现LDCT肺部图像的精准重建。项目主要研究内容包括1）肺部成像模型的构建及低剂量CT投影数据的获取；2）低剂量CT投影中的高污染投影选择和处理方法；3）基于正常剂量CT图像库学习的正则项构建方法；4）肺部低剂量CT成像评价。针对以上研究内容，首先基于正常投影数据构建了肺部低剂量CT投影仿真模型，获得低剂量肺部CT投影；其次构建了一种基于高质量CT图像库学习的非局部均值正则项：通过高质量图像构建图像块数据库，作为训练样本，利用低剂量图像块与之相似性，挖掘先验信息，将构建的正则项引入结合了投影惩罚加权最小二乘处理的目标函数，进行迭代重建。因先验信息来自于大量训练样本，避免了高质量图像和LDCT图像的配准过程，减少重建误差；同时与基于自身图像的正则项相比，减少了重复扫描带来的剂量损伤；最后将所提方法用于肺部LDCT迭代重建，可有效抑制图像噪声，保持图像细节和纹理特征，降低成像剂量至正常剂量的1/10。本项目所提方法为LDCT肺癌筛查提供更精准的重建算法和系统开发基础，有望应用于我国高危人群的LDCT肺癌筛查，具有重要研究意义和社会意义。通过本项目研究，课题组构建了肺部的高质量CT图像数据库和不同剂量水平的低剂量CT投影，可为后续相关研究提供帮助。此外课题组还构建了多种低剂量X线发光成像策略，为本项目拓展了相关应用，储备了系列X线发光投影数据，可用于新型X线荧光分子成像研究。