Higher order tensor-based segmentation and n-furcation modeling of vascular structures

Official URL: http://campar.in.tum.de/files/STENT/STENT2014/CVII_STENT_2014_proceedings.pdf

A new vascular structure segmentation method, which is based on a cylindrical flux-based higher order tensor (HOT), is presented. On a vessel structure, HOT naturally models branching points, which create challenges for vessel segmentation algorithms. In a general linear HOT model, embedded in 3D, one has to work with an even order tensor due to an enforced antipodal-symmetry on the unit sphere in 3D. However, in scenarios such as in a bifurcation, the antipodally-symmetric tensor models of even order will not be useful. In order to overcome that limitation, we embed the tensor in 4D and obtain a structure that can model asymmetric junction scenarios. Thus, we will demonstrate a seed-based vessel segmentation algorithm, which exploits a 3rd or 4th order tensor constructed in 4D. We validate the algorithm on both synthetic complex vascular structures as well as real coronary artery datasets of the Rotterdam Coronary Artery Algorithm Evaluation framework.