This paper aims to propose a noninvasive radiotherapy patient positioning system based on structured light surface imaging, and evaluate its clinical feasibility. First, structured light sensors were used to obtain the panoramic point clouds during radiotherapy positioning in real time. The fusion of different point clouds and coordinate transformation were realized based on optical calibration and pose estimation, and the body surface was segmented referring to the preset region of interest (ROI). Then, the global-local registration of cross-source point cloud was achieved based on algorithms such as random sample consensus (RANSAC) and iterative closest point (ICP), to calculate 6 degrees of freedom (DoF) positioning deviation and provide guidance for the correction of couch shifts. The evaluation of the system was carried out based on a rigid adult phantom and volunteers' body, which included positioning error, correlation analysis, and receiver operating characteristic (ROC) analysis. Using Cone Beam CT (CBCT) as the gold standard, the maximum translation and rotation errors of this system were (1.5 ± 0.9) mm along Vrt direction (chest) and (0.7 ± 0.3) ° along Pitch direction (head and neck). The Pearson correlation coefficient between results of system outputs and CBCT verification distributed in an interval of [0.80, 0.84]. Results of ROC analysis showed that the translational and rotational AUC values were 0.82 and 0.85, respectively. In the 4D freedom accuracy test on the human body of volunteers, the maximum translation and rotation errors were (2.6 ± 1.1) mm (Vrt direction, chest and abdomen) and (0.8 ± 0.4)° (Rtn direction, chest and abdomen) respectively. In summary, the positioning system based on structured light body surface imaging proposed in this article can ensure positioning accuracy without surface markers and additional doses, and is feasible for clinical application.
本文开发了基于结构光体表成像的无创放疗患者摆位引导方法和系统，并评估其临床可行性。首先，选用结构光传感器实时获取放疗摆位场景全景点云，基于光学标定和姿态估计实现点云融合和坐标变换，并参照预设感兴趣区域（ROI）实现体表点云分割。然后，基于随机抽样一致（RANSAC）和迭代最近点（ICP）等算法实现跨源点云全局—局部配准，以计算6自由度摆位偏差并引导移床校正。本研究基于刚性成人模体和志愿者人体对系统进行评估，包括摆位误差测量、相关性分析和受试者操作特征（ROC）分析。模体测试以锥形束计算机断层扫描（CBCT）作为金标准，本文开发系统最大平移和旋转误差分别为（1.5 ± 0.9）mm［垂直（Vrt）方向，胸部］和（0.7 ± 0.3）°［俯仰（Pitch）方向，头颈部］；系统整体输出结果与CBCT验证之间的皮尔逊（Pearson）相关系数分布在［0.80, 0.84］区间内；ROC分析显示，平移和旋转ROC曲线下区域面积（AUC）值分别为0.82和0.85。而在志愿者人体4自由度精度测试中，最大平移和旋转误差分别为（2.6 ± 1.1）mm（Vrt方向，胸腹部）和（0.8 ± 0.4）°［旋转（Rtn）方向，胸腹部］。综上，本文提出的基于结构光体表成像的摆位系统在无体表标记和额外剂量的前提下，可确保摆位精度，具备临床应用可行性。.
Keywords: Noninvasive; Optical surface; Point cloud; Radiotherapy positioning; Structured light.