Zhang Lei, Zhang Yawei, Zhang You, Harris Wendy B, Yin Fang-Fang, Cai Jing, Ren Lei
Medical Physics Graduate Program, Duke University, Durham, NC, USA.
Department of Radiation Oncology, Duke University Medical Center, Durham, NC, USA.
Cancer Transl Med. 2017;3(6):185-193. Epub 2017 Dec 29.
During cancer radiotherapy treatment, on-board four-dimensional-cone beam computed tomography (4D-CBCT) provides important patient 4D volumetric information for tumor target verification. Reconstruction of 4D-CBCT images requires sorting of acquired projections into different respiratory phases. Traditional phase sorting methods are either based on external surrogates, which might miscorrelate with internal structures; or on 2D internal structures, which require specific organ presence or slow gantry rotations. The aim of this study is to investigate the feasibility of a 3D motion modeling-based method for markerless 4D-CBCT projection-phase sorting.
Patient 4D-CT images acquired during simulation are used as prior images. Principal component analysis (PCA) is used to extract three major respiratory deformation patterns. On-board patient image volume is considered as a deformation of the prior CT at the end-expiration phase. Coefficients of the principal deformation patterns are solved for each on-board projection by matching it with the digitally reconstructed radiograph (DRR) of the deformed prior CT. The primary PCA coefficients are used for the projection-phase sorting.
PCA coefficients solved in nine digital phantoms (XCATs) showed the same pattern as the breathing motions in both the anteroposterior and superoinferior directions. The mean phase sorting differences were below 2% and percentages of phase difference < 10% were 100% for all the nine XCAT phantoms. Five lung cancer patient results showed mean phase difference ranging from 1.62% to 2.23%. The percentage of projections within 10% phase difference ranged from 98.4% to 100% and those within 5% phase difference ranged from 88.9% to 99.8%.
The study demonstrated the feasibility of using PCA coefficients for 4D-CBCT projection-phase sorting. High sorting accuracy in both digital phantoms and patient cases was achieved. This method provides an accurate and robust tool for automatic 4D-CBCT projection sorting using 3D motion modeling without the need of external surrogate or internal markers.
在癌症放射治疗期间，机载四维锥形束计算机断层扫描（4D-CBCT）为肿瘤靶区验证提供重要的患者四维容积信息。4D-CBCT图像的重建需要将采集到的投影分类到不同的呼吸相位。传统的相位分类方法要么基于外部替代物，这可能与内部结构存在错误关联；要么基于二维内部结构，这需要特定器官的存在或缓慢的机架旋转。本研究的目的是探讨基于三维运动建模的无标记4D-CBCT投影相位分类方法的可行性。
将模拟过程中采集的患者4D-CT图像用作先验图像。主成分分析（PCA）用于提取三种主要的呼吸变形模式。机载患者图像容积被视为呼气末相位时先验CT的变形。通过将每个机载投影与变形后的先验CT的数字重建射线照相（DRR）进行匹配，求解主要变形模式的系数。主要的PCA系数用于投影相位分类。
在九个数字体模（XCATs）中求解的PCA系数在前后方向和上下方向上均显示出与呼吸运动相同的模式。所有九个XCAT体模的平均相位分类差异均低于2%，相位差<10%的百分比为100%。五例肺癌患者的结果显示平均相位差在1.62%至2.23%之间。相位差在10%以内的投影百分比范围为98.4%至100%，相位差在5%以内的投影百分比范围为88.9%至99.8%。
该研究证明了使用PCA系数进行4D-CBCT投影相位分类的可行性。在数字体模和患者病例中均实现了高分类精度。该方法提供了一种准确且稳健的工具，用于使用三维运动建模进行自动4D-CBCT投影分类，无需外部替代物或内部标记。
