From 2D markers in MIP to 3D vessel segmentation: A fuzzy paradigm for connected filtering

Connected filters, and in particular those relying on component trees, can be involved in vessel filtering and segmentation tasks. We propose a new component tree-based segmentation method that enables to easily obtain vascular volumes from 3D data, by simply providing 2D markers from MIP visualisations. This approach gathers the advantages of component trees (low algorithmic cost) and the user-friendly (fuzzy) handling of 3D images via 2D representations, leading to an ergonomic and fast tool. It has been successfully used for segmenting a large dataset of MRAs and CTAs visualizing Willis polygons, in the clinical context of stenosis detection.