A framework for automatic, accurate, and fast 2-D+T/3-D registration applied to trans-catheter aortic valve implantation (TAVI) procedures

Presentation of detailed anatomical structures via 3-D models of the aortic root helps navigation and deployment of the prosthetic valve during TAVI procedures. Accurate 2-D/3-D registration of the 3-D model with one static X-ray image has been achieved using a hybrid method in [1]. In this paper, we extend the above method to register the 3-D model to a sequence of angiographic images that can be used for dynamic 2-D/3-D fusion. The key idea is to iterate the processes of motion-model construction and motion-model-guided 2-D/3-D registration. This iterative process starts from the frames with a good contrast presence, and extends gradually to those frames with less contrast and hence requires the motion-model to boost the registration performance. Image-based automatic motion-phase detection and contrast agent detection are utilized to fully explore the pseudo-cyclic property of the aortic motion and makes our framework fully automatic. Compared to the naïve frame-by-frame 2-D/3-D registration approach, the proposed method reduced the registration error and the computational time by a factor of 6 and 12 respectively, and boosted the success rate from ~50% to ~100%, on five clinical TAVI data.