3D Reconstruction of Blood Vessel from Stereo X-ray Images

Thesis (Master's)--University of Washington, 2017-06We propose a fully automatic system to extract 3D structure of blood vessels from stereo X-ray images. Currently, typical 3D imaging technologies like angiography are expensive and potentially harmful to human body. In addition, for complex images with bones, 3D vessel representation will thus depend on further 3D tracing and segmentation. Because vessels are featureless and have low intensity contrast with background, other reconstruction methods like stereo are additionally challenging. Our system can effectively reconstruct main vessels in following steps. We first do initial segmentation using Markov Random Field and then further refine segmentation in an entropy based post-process. We then extract vessel centerlines and generate trees. Stereo matching is done in a coarse-to-fine scheme: Initial matching using affine transform and dense matching using Hungarian algorithm guided by Gaussian Regression. We test and discuss its performance on stereo X-ray images and synthetic datasets. We also compare our method with human labeling and it achieves an accuracy of 71.08%