A Robust Deterministic-Based Automatic Vessel Centerline Extraction Algorithm in 3-D Binary Volumes

In this manuscript we propose a new fully automated approach for centerline extraction in vascular trees from a 3D medical binary volume that delivers smooth and robust results. A 3D binary volume which has been segmented is taken as an input data. Distance map is generated from the tridimensional Euclidian distance transform of the 3D binary volume data followed by the Hessian matrix computation for each target voxel based on the distance map. We utilized the distance transform because it indicates the distance between the point to the closest surface of the vessel. Analysis of Hessian matrix eigenvalues for each target voxel is used to extract the direction of the centerline. Since the proposed approach is totally automated, the user is not required to carry out any data preprocessing. The initial point from the centerpoint list is taken as the seed point through which the algorithm propagates to generate the centerline. We also tackled the complex issue of bifurcation where the vessel splits into multiple branches. We have demonstrated the performance of our algorithm on 6 different datasets of coronary artery, where the outcome was one voxel thick, connected centerline. Based on this centerline we can estimate the different geometrical parameters of the vessels, like radius, curvature, and length.