A Self-supervised 3D/2D Registration Method for Incomplete DSA Vessels

For vascular interventional surgery, the preoperative 3D computed tomography (CT) has complete information of vessels but is not convenient for obvervation, while the intraoperative 2D digital subtraction angiography (DSA) is easy for doctors to monitor the vascular conditions real-timely but information is incomplete in each frame. As a result, 2D/3D registration, which is the technology to fuse information from images of different modal, is useful for the guidance of vascular interventional surgery. In this paper, we proposed a self-supervised 2D/3D vascular registration method to improve the performance on DSAs with incomplete vessels. The proposed method contains a rigid and an elastic registration stage, for regressing the 6-dim parameters to obtain a center image and fine-tuning respectively. In addition, a patch-based content loss is introduced to the rigid registration step to give an appropriate similarity measure for images with incomplete vessels, and a masked elastic module is introduced to simulate the incompletion and deformation caused by breath or heart beats on the real vessels in elastic registration. We evaluated our method on both simulated and real images. Experiments prove that our proposed method is effective to register CT and DSA images.