A method for detecting blood vessel bifurcations in digital medical images includes inflating a sphere from a first center point inside a segmented blood vessel until a surface of the sphere intersects a surface of the blood vessel, searching within the inflated sphere for a second center point that has a sphere intersecting a surface of the blood with a maximum radius, assigning all voxels of the maximal radius sphere to a root node of a shape-tree, increasing the radius of the maximal radius sphere and computing a voxel difference set with respect to the previous maximal radius sphere, computing one or more connected components Cm in the voxel difference set, assigning voxels of each connected components to a different child node of the shape tree, connecting each child node with the root node, and calculating features from the shape tree for training a classifier to detect blood vessel bifurcations.
