3D/2D Vascular Elastic Registration Based On Coronary CTA And DSA Images

Cardiovascular disease,mainly coronary heart disease,is a major global problem that restricts human health.2D DSA image-guided percutaneous coronary intervention is currently the main treatment method of coronary heart disease.The goal of 3D/2D vascular elastic registration is to anatomically align the vascular structures in the 3D coronary CTA images acquired by before surgery with the vascular structures in the intraoperative real-time 2D coronary DSA images.Through 3D/2D vascular elastic registration,2D DSA images can be enhanced after intraoperative imaging,and the correlation between preoperative diagnosis and intraoperative information can be realized,so as to provide doctors with more auxiliary information during operation,to improve the precision and success rate of surgery,and also improve the prognosis of patients.Existing 3D/2D vascular elastic registration methods are insufficient in registration accuracy or algorithm time-consuming.In this thesis,three 3D/2D vascular elastic registration methods are proposed,which solve the problems of current methods from different perspectives.In the proposed matching-deformation registration algorithm based on point sets from 3D/2D vascular centerlines,a registration cost function including distance constraint,deformation length change constraint and link constraint is designed,which improves the accuracy and robustness of registration.In the proposed matchingdeformation registration algorithm based on curves from 3D/2D vascular centerlines,a novel restricted breadth-first search algorithm is used to ensure the accuracy of the registration algorithm and decrease the algorithm time-consuming.Finally,an elastic registration method based on deep graph convolutional network is proposed.The elastic deformation field is obtained based on CTA vascular centerline,CTA image projection and DSA image through a feature extraction module and an elastic registration module.In the network,the 2D vascular centerline is not needed any more.The average registration distance error of the elastic registration method based on the deep graph convolutional network on clinical data is 2.08 mm,and the average time-consuming is 1.58 s,which can preliminarily meet the requirements of the actual intraoperative scene.