基于正侧位锥束CT投影快速构建脊椎三维模型

Rapid construction of three-dimensional spine model based on positive and lateral cone-beam CT projection

目的探讨基于C形臂的两张正侧位2D锥束CT(CBCT)投影图像进行3D模型重建的效果。方法采用半自动化的二维投影图像特征点提取算法,选定18点的特征点集并提取其对应的正侧位二维投影图像平面坐标,然后针对C形臂CBCT系统建立坐标系,推导特征点三维空间坐标与其在投影图像中平面坐标之间的几何关系,代入转换公式获得特征点集的三维空间坐标。利用薄板样条法对三维脊椎基础模型进行空间非刚性插值获得三维脊椎目标模型。将L3石膏模型置入C形臂CBCT系统,获取375幅圆周扫描图像,利用FDK算法重建三维模型并进行表面重采样得到三维脊椎参考模型,对其进行不规则形状调制得到三维脊椎基础模型,利用本文方法对三维脊椎基础模型进行空间非刚性插值得到三维脊椎目标模型,并设置对照组对本文方法的精度进行评价。结果相对于特征点手动提取和边缘增强提取算法,采用半自动化特征点提取算法构建获得的三维脊椎目标模型与参考模型的误差降低至1mm以内。结论采用本文方法可构建出近似的、精度较高的脊椎三维模型,为基于C形臂CBCT的手术导航提供3D图像支持。

Objective To explore the 3D reconstruction based on two positive and lateral cone-beam CT （CBCT） projection images. Methods Firstly, a feature point extraction algorithm based on semi-automated 2D projection image was proposed. The feature points of 18 points were selected and the corresponding normal lateral position 2D projection image plane coordinates were extracted. Then, a coordinate system was established for C-arm CBCT imaging system. The geometric relationship between the 3D space coordinates of the feature points and the corresponding projection coordinates in the ima- ging plane was deduced. Getting the geometric relationship into the transformation formula and the 3D space coordinates of the feature points were obtained. Finally, the thin plate spline space non rigid interpolation method was used to transform the 3D spinal basic model into a 3D spinal target model. The L3 plaster model was put into the C-arm CBCT imaging sys- tem, and the 375 circumferential scanning images were obtained. FDK algorithm was used to reconstruct the 3D model, and its surface resampling was used to get the 3D spine reference model. 3D spinal basic model was obtained by irregular shape modulation of 3D spinal reference model. In this paper, the spatial non rigid interpolation method was used to trans- form the 3D spinal basic model into a 3D spinal target model, and the control group experiment was set up to evaluate the accuracy of this method. Results Compared to the feature point manual extraction and edge enhancement extraction algo- rithm, this paper proposed the semi-automatic feature point extraction algorithm to construct the 3D spine target model and the reference model, and the accuracy error was improved within 1 mm. Conclusion The method of this paper can be used to construct approximate 3D spine model with high accuracy for image guided surgery based on C-arm CBCT.