Phase Analysis Based Vessel Recompletion with Incomplete DSA for Dynamic Coronary Roadmap

Digital subtraction angiography(DSA) is a technique used extensively in interventional operation, which visualizes blood vessels and catheters through contrast agent. However, DSA image displays only local vessels due to the decreasing concentration of the contrast through blood flow. The rest of the image tend to be either unclear or unnoticeable. To solve the issues above, Dynamic Coronary Roadmap are presented aiming to create a dynamic and complete map of vessels, which is helpful for surgeons or surgical robots in clinical surgery. In this paper, we propose a simple and effective method, utilizing a segmentation network and an optical flow network to obtain a mask of the complete vessels for each frame in a DSA sequence. The segmentation model can be any of the common networks, such as Unet. For the optical flow model, this paper proposes a self-supervised training methodology to address the problem of insufficient training data for elastic deformation models. The results of this experiment indicates that the proposed algorithm is able to reconstruct complete vessels in a high recall.