A parallelized 4D reconstruction algorithm for vascular structures and motions based on energy optimization

摘要

In this paper, we present a parallel 4D vessel reconstruction algorithm that simultaneously recovers 3D structure, shape, and motion based on multiple views of X-ray angiograms. The fundamental goal is to assist the analysis and diagnosis of interventional surgery in the most efficient way towards interactive and accurate performance. We start with a fully parallelized algorithm to extract vessels as well as their skeletons and topologies from dynamic image sequences. Then, instead of resorting to registration, we present an algorithm to formulate the reconstruction problem as an energy minimization problem with color, coherence, and topology constraints to reconstruct the 3D vessel initially, which is robust to combat noise and incomplete information in images. Next, we incorporate temporal information into our energy optimization framework to track and reconstruct 4D kinematics of the dynamic vessels, which is also capable of recovering previous incomplete and misleading shapes acquired from static images otherwise. We demonstrate our system in coronary arteries reconstruction and movement tracking for percutaneous coronary intervention surgery to help medical practitioners learn about the 3D shapes and their motions of the coronary arteries of specific patient. We envision that our system would be of high assistance for diagnosis and therapy to treat vessel-related diseases in a clinical setting in the near future.