面向肺部肿瘤精确放疗的4D-CBCT优质重建新方法研究

During the process of radiation therapy for lung cancer, significant motion induced displacement can be occurred in the respiratory cycle. In this case, the motion-blur and artifacts in the 3D-CBCT will seriously affect the accurate tumor positioning and the following adjustments of the treatment plan . On the other side, the respiratory correlated 4D-CBCT can provide accurate tumor shape and motion information. However, restricted by the hardware, the projection number for a single phase is very low, and the associated image reconstructed by using conventional reconstruction algorithms is constrained by serious streak artifacts, which cannot meet the clinical requirements. Thus, based on our previous studies, we are going to conduct a study focusing on high quality 4D-CBCT image reconstruction. The research priorities of this project include: (1) based on the basic theory of 4D–CBCT imaging, we will construct the peculiar optical statistical model and physical model, which provides the theoretical support for 4D-CBCT reconstruction; (2) based on the constructed optical model and physical model, we will build the reconstruction framework. We will focus on respiratory information induced prior design to achieve get simultaneous motion estimation and image reconstruction, and we will also further discusses the design of the optimization algorithm under the new model to obtain accurate and robust reconstructions; (3) to carry out clinical evaluation of the current reconstruction algorithms.

在肺癌肿瘤放射治疗过程中肺内肿瘤随呼吸运动产生显著的位移，常规用于引导放疗的3D-CBCT图像含有严重的呼吸运动模糊和伪影，难以实现肿瘤的精确定位与治疗计划后续的调整，而基于不同呼吸时相的4D-CBCT可以提供准确的肿瘤形态以及运动信息。然而受硬件限制，4D-CBCT中单个呼吸相位的投影数目较少，常规解析算法重建图像中含有大量伪影，无法满足临床要求。囿于此，在申请人前期研究的基础上，针对4D-CBCT的优质重建展开系列研究，重点内容包括:(1)从4D-CBCT成像底层重建理论出发，构建4D-CBCT成像中特有的光学统计和物理成像模型，为4D-CBCT重建提供理论支撑；（2）构建以光学模型为基础的重建模型，重点探讨呼吸运动信息导引的先验模型构建，以实现运动估计和图像重建联合进行，并进一步探讨新模型下优化算法的设计，以期实现4D-CBCT的准确、鲁棒重建；（3）开展对新算法广泛的临床评估研究。

肺癌肿瘤放射治疗过程中肺内肿瘤随呼吸运动产生显著的位移，常规用于引导放疗的3D-CBCT图像含有严重的呼吸运动模糊和伪影，难以实现肿瘤的精确定位与治疗计划后续的调整，而基于不同呼吸时相的4D-CBCT可以提供准确的肿瘤形态以及运动信息。然而受硬件限制，4D-CBCT中单个呼吸相位的投影数目较少，常规解析算法重建图像中含有大量伪影，无法满足临床要求。囿于此，本研究针对四维锥形束CT的优质重建展开系列研究，具体地：1） 提出了一种基于运动补偿后序列图像的全变分正则化先验构建模型，并提出了相应的优化算法，相应的论文发表于Physics in Medicine and Biology 杂志；2）构建了包含基于多相位投影数据的保真项和基于压缩感知理论的全变分正则化的代价函数。对于投影数据保真项的设计，不再局限于当前相位投影数据，而是利用多个相位投影数据通过变形的成像模型联合构建。对于复杂代价函数的优化，我们利用GPBL（Gradient-Projection-Barzilai-Linesearch, GPBL）算法来实现,相应的论文发表于IEEE-MIC 会议及南方医科大学学报；3）利用运动轨迹搜索机制，提出了运动信息导引的时空先验模型，算法实现过程中将图像分成三维的块，然后利用估计运动变形场对图像块沿相位轴进行搜索，然后利用张量分解（HOSVD）策略对图像块进行提升和优化，相位的论文发表于Scientific Report 杂志；4）利用DTS 投影数据本身构建了呼吸信号模型，并结合动态规划技术和边缘检测技术实现基于膈肌的呼吸信号实时动态提取，及4D-DTS 的优质重建，相应论文发表于Medical Engineering and Physics 杂志；5）针对常规PICCS算法中先验图像中易引入运动模糊的问题，提出了运动补偿先验图像用于PICCS算法4D-CBCT重建策略.