摘要
目的针对术前三维电子计算机断层扫描(computed tomography,CT)图像和术中二维X线图像的配准问题提出了精确高效的全自动配准算法,使得该算法能够适应不同X线图像的风格。方法首先通过数字重建放射影像(digitally reconstructured radiograph,DRR)技术,把CT图像转换成DRR图像,从而将CT和X线图像的配准转换成DRR图像和X线图像的配准。然后在传统的归一化互相关(normalized cross correlation,NCC)和梯度差分(gradiant difference,GD)相似性测度指标基础上提出融合NCC和GD的归一化互相关-梯度差分(normalized cross correlation-gradient difference,NG)指标,并基于NG指标计算DRR图像和X线图像的相似性。通过Powell算法迭代求取相似性测度的极值,从而获得配准矩阵。最后在人体头颅CT数据上采用“黄金标准”判断法量化配准系统的精度,并通过脊柱CT和X线配准案例验证系统在实际场景的性能。结果基于DRR及NG相似性测度的2D/3D图像配准系统的配准距离误差为0.51 mm,角度误差为0.40°。结论基于DRR及NG相似性测度的2D/3D图像配准算法具有较好的配准精度,能适应不同风格的二维输入图像,基于该算法的配准系统具有较好的鲁棒性。
Objective To accurately,efficiently,and automatically register the preoperative computed tomography(CT)image and the inoperative X-ray image in order to improve the adaptability to different X-ray image styles.Methods CT was firstly converted into digital reconstruction radiography(DRR)images through the digital reconstruction radiography technology.The registration of CT and X-ray was converted into the registration of DRR and X-ray images.Then,based on the traditional similarity measures,normalized cross correlation(NCC)and gradient difference(GD),normalized cross correlation-gradient difference(NG)was proposed to calculate the similarity between DRR and X-ray.Powell algorithm was used to get the extremum value of the similarity metric.The registration matrix was obtained after the optimization process.At last,using the CT image of the skull,the“Gold standard”judgment method was used to quantitatively evaluate the registration algorithm.The spine CT and X-ray registration case was also applied to qualitatively evaluate the performance of the registration system in practical application.Results The registration distance errors of the angle and position of the 2D/3D medical image registration algorithm based on DRR and the NG similarity measure were 0.40°and 0.51 mm.Conclusions The 2D/3D image registration algorithm based on DRR and the NG similarity measure can improve the accuracy and adapt to different image styles.The registration system based on the algorithm is robust.
作者
麦永锋
孙启昌
贾鹏飞
陈晓军
MAI Yongfeng;SUN Qichang;JIA Pengfei;CHEN Xiaojun(School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240;Engineering Research Center of Digital Medicine and Clinical Transiation,Ministry of Education,Shanghai 200240;Institute of Medical Robotics,Shanghai Jiao Tong University,Shanghai 200240)
出处
《北京生物医学工程》
2021年第3期263-272,共10页
Beijing Biomedical Engineering
基金
国家重点研发计划项目(2017YFB1302900)
国家自然科学基金(81971709、81828003、M-0019、82011530141)
上海市科委“科技创新行动计划”(19510712200、20490740700)
上海交通大学医工交叉基金(ZH2018ZDA15、YG2019ZDA06、ZH2018QNA23)
厦门市重大科技项目(3502Z20201030)资助。
关键词
2D/3D配准
数字重建放射影像
相似性测度
POWELL算法
归一化互相关
梯度差分
2D/3D registration
digitally reconstructed radiograph
similarity measurement
powell algorithm
normalized cross correlation
gradient difference
