Extracting Vessel Structure From 3D Image Data

This thesis is focused on extracting the structure of vessels from 3D cardiac images. In many biomedical applications it is important to segment the vessels preserving their anatomically-correct topological structure. That is, the final result should form a tree. There are many technical challenges when solving this image analysis problem: noise, outliers, partial volume. In particular, standard segmentation methods are known to have problems with extracting thin structures and with enforcing topological constraints. All these issues explain why vessel segmentation remains an unsolved problem despite years of research. Our new efforts combine recent advances in optimization-based methods for image analysis with the state-or-the-art vessel filtering techniques. We apply multiple vessel enhancement filters to the raw 3D data in order to reduce the rings artifacts as well as the noise. After that, we tested two different methods for extracting the structure of vessels centrelines. First, we use data thinning technique which is inspired by Canny edge detector. Second, we apply recent optimization-based line fitting algorithm to represent the structure of the centrelines as a piecewise smooth collection of line intervals. Finally, we enforce a tree structure using minimum spanning tree algorithm