Fast extraction of tubular and tree 3D surfaces with front propagation methods

Abstract

We present a new fast approach for surface segmentation of thin structures, like vessels and vascular trees, based on Fast Marching and Level Sets methods. Fast Marching allows segmentation of tubular structures inflating a "long balloon" from a user given single point. However, when the tubular shape is rather long, the front propagation may blow up through the boundary of the desired shape close to the starting point. Our contribution is focused on a way to go on front propagation only on the actually moving front and freezing other points. We demonstrate the ability to build a fast and accurate segmentation for those tubular and We structures. We also develop a useful stopping criterion for the causal front propagation. We illustrate our algorithms by applications on the segmentation of vessels in 3D medical images.