Eigenvector Grouping for Point Cloud Vessel Labeling

Patryk Rygiel, Maciej Zieba, Tomasz K. Konopczynski

Abstract: Segmentation of coronary arteries from Coronary Computed Tomography Angiography (CCTA) is an essential step in developing various noninvasive diagnostic methods. In this work, we tackle the task of vessel labeling on coronary artery voxel-based prediction by use of point cloud artificial neural network. We propose a novel point aggregation technique Eigenvector Grouping (EVG), tailored to the analysis of tubular-like structures. We further utilize a specifically designed post-processing technique Component-Wise Majority Point Voting (CMPV), to refine point cloud segmentation by enforcing class consistency among connected components. We show that our solution outperforms previously proposed methods in the vessel labeling task on a CCTA dataset especially, in the presence of disrupted segmentations.