A Unified Minimal Path Tracking and Topology Characterization Approach for Vascular Analysis

We present a unified coarse-to-fine approach for extracting the medial axis representations (centerlines) of human vasculature in contrast enhanced (CE)-CTA/MRA. The proposed method constitutes two separate analysis stages that are successively applied (and repeated) for a refined extraction. The former stage involves the use of a graph-based optimization algorithm that identifies the minimum-cost paths between user-specified seed points. The costs of all feasible paths are efficiently computed via the medialness filter, which is a contrast- and scale-invariant local operator sensitive to the presence of tubular structures. Nonetheless, image noise and the presence of nearby blood vessels can affect the quality of detection and delineation. In the latter stage, we thereby employ a novel multiscale orientation descriptor so as to guide/stop additional minimal path extraction steps. Specifically, the descriptor is designed to classify a point of interest as vessel or non-vessel, as well as to obtain a reliable estimate of the number and directions of the vascular segments (branches) at a vessel point. Our method improves the accuracy of extraction by robustly identifying critical configurations such as bifurcations, endpoints, or non-vessel points, and thereby delineating/eliminating missing/spurious vessel branches.