摘要
荧光分子断层成像技术(fluorescence molecular tomography,FMT)系统中为获得体内光源的结构信息,需要利用CT体数据。FMT系统在进行光学图像与CT图像的配准时,由于两种模态图像的成像原理、图像风格和图像维度等方面的差异,导致传统配准方法耗时长、效果差。本研究提出了一种基于T2DR-Net(texture transfer and dense registration net)与互信息的光学-CT图像配准方法,实现FMT系统中白光图像与CT图像的配准。该方法将光学-CT图像配准分为粗配准和精配准两个部分。在粗配准阶段,利用CycleGAN实现了FMT白光图像和CT投影像的纹理迁移,以降低两种图像纹理差异对图像配准的影响,并提出了DenseReg-Net模型获取白光图像和CT投影像粗配准参数;在精配准阶段,通过互信息方法进一步对两种模态图像配准,并得到最终的配准结果。利用1330张光学图像和39711张CT投影像作为样本集来验证配准方法的有效性,实验结果表明,本研究提出的光学-CT图像配准方法,相关系数为0.8797±0.0175,结构相似性为0.8683±0.0051,模型配准时间为(2.88±1.39)s。模型的配准效果及其稳定性优于传统方法。此外,与传统方法相比,速度提升了约60倍。
In order to obtain the structural information of the in vivo light source,the fluorescence molecular tomography(FMT)system needs to use computed tomography(CT)volume data.When the FMT system performs the registration of the optical image and the CT image,due to the differences in the imaging principles,image styles,and image dimensions of the two modal images,the traditional registration method takes a long time and has poor results.We proposed an optical-CT image registration model based on texture transfer and dense registration net(T2DR-Net)and mutual information to realize the registration of white light images and CT images in the FMT system.The method included two parts:rough registration and fine registration.In the rough registration stage,in order to reduce the impact of the difference of the two image textures on the image registration,CycleGAN was used to realize the texture transfer of FMT white light image and CT projection image,and the DenseReg-Net model was proposed to obtain the white light image and CT projection image rough registration parameters.In the fine registration stage,the mutual information method was used for registration correction,and the final registration result was obtained.We used 1330 optical images and 39711 CT projection images as sample sets to verify the validity of the registration method.The results of the registration showed that this optical-CT image registration model had a correlation coefficient of 0.8797±0.0175 and a structural similarity of 0.8683±0.0051,the registration time of this model was(2.88±1.39)s.The registration effect and stability of the model are better than that of traditional methods,and it is about 60 times faster than the traditional method.
作者
崔建良
陈春晓
陈志颖
姜睿林
CUI Jianliang;CHEN Chunxiao;CHEN Zhiying;JIANG Ruilin(Department of Biomedical Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China)
出处
《生物医学工程研究》
2022年第2期143-150,共8页
Journal Of Biomedical Engineering Research
关键词
图像配准
荧光成像
2D/3D配准
纹理迁移
互信息
光学/CT图像配准
Image registration
Fluorescence imaging
2D/3D registration
Texture transfer
Mutual information
Optical/CT image registration
作者简介
通信作者:陈春晓,Email:ccxbme@nuaa.edu.cn。
