2D-3D配准中的初始位姿估计方法

Initial Pose Estimation Method in 2D/3D Registration

针对2D-3D医学图像配准过程中存在的初始参数变换范围小和配准对象单一等局限性,提出了一种基于参数映射和图像分割的初始位姿估计方法,扩大了三维物体空间参数的起始运动偏差范围,解决了骨科手术导航系统在配准过程中可估计参数少和多骨块图像配准难等问题.首先,使用区域生长算法对多骨块影像进行图像分割与感兴趣目标的提取,完成待配准目标数据的获取.其次,基于正交融合投影原理构建投影成像模型,将空间刚体变换参数分解到正侧位平面上,建立空间参数与平面参数间的映射关系.然后,将投影生成的正交双平面模板与对应目标图像进行匹配,并将得到的正侧位平面配准参数转换为空间参数,从而完成三维物体初始位姿的有效估计.最后基于颅骨、完整股骨和骨折股骨的CT数据对该方法进行验证,并与传统2D-3D图像配准方法进行对比研究.实验结果表明,所述方法可使5个空间参数的初始变换范围达到±30 mm或±20°,并且对在该范围内的不同研究对象都展现出良好的初始位姿估计效果.与传统图像配准方法相比,基于上述方法进行2D-3D迭代优化的整体配准时间平均缩短46.5%,单个平移参数配准精度最高提升69.2%,单个旋转参数配准精度最高提升91.1%,同时最终配准结果也兼具良好的鲁棒性.

In this paper,we propose a new initial pose estimation approach based on parameter mapping and image segmentation,focusing on the problems of small initial parameter transformation range and single registration object in the 2D/3D medical image registration process.This approach expands the range of initial motion deviation of 3D object spatial parameters and solves the issues regarding few estimable parameters and multibone image registration difficulty of the surgical navigation system during the registration process.First,the region growth algorithm is used to segment the multibone image and extract the target of interest to complete the target-data acquisition to be regis-tered.Second,a projection imaging model is constructed based on the principle of orthogonal fusion projection to decompose the spatial rigid body transformation parameters into the positive lateral plane and establish the mapping relationship between spatial and plane parameters.Then,the orthogonal biplane template generated using the projection is matched with the corresponding target image,and the registration parameters of the obtained positive and lat-eral plane are converted into spatial parameters,to achieve the effective estimation of the initial pose of the 3D ob-ject.The method was verified based on computed tomography data of the skull,intact femur,and fractured femur.Then,the method was compared with the traditional 2D/3D image registration method.The experimental results showed that the proposed method could make the initial transformation range of the five spatial parameters reach±30 mm or±20°,and different research objects within this range show a good initial pose estimation effect.Com-pared with the traditional image registration method,the overall registration time of the 2D/3D iterative optimization based on the above method is 46.5%shorter on average,the registration accuracy of a single translation parameter is up to 69.2%,and the registration accuracy of a single rotation parameter is up to 91.1%.Simultaneously,the final registration results have adequate robustness.