LABORATORY OF MATHEMATICS IN IMAGING Algorithms for Extracting Vessel Centerlines

1 The detection of the vessels centerlines is a useful preprocessing step for 3D quantification of stenosis, topological representation of the vessel tree and registration with an atlas, virtual endoscopy, and visualization of the vascular network. We propose to compare three classes of algorithms leading to centerline representations of the vessels. The first one relies on a pre-segmentation of the vessels, given by a binary image, to compute a topological invariant skeleton. The second method extracts sub-voxel centerlines as ridges of the image intensity. It uses the gradient and the Hessian matrix to interpolate the zero-crossings of the gradient vector in the cross-sectional directions. The third method uses an integration of the gradient information along circles of different radii in the cross-sections, in order to find points located at equal distance from the contours . We present results on a Magnetic Resonance Angiography, show the advantages and the drawbacks of each method and present some perspectives. Grant support: NIH P41-RR13218 and CIMIT