Segmentation of vessel voxel structures using gradient ridges

To extract shape information, we segment a gray value volume of the brain vessels by means of a rea growing algorithm with two thresholds. Our primary algorithm for computation of these thresholds is based on the assumption that the gray value volume contains much more tissue than vessel voxels. However, in some cases this assumption is not valid. Therefore, we developed a new algorithm to correctly handle these cases. We segmented thirty trouble-free clinical volume data-sets twice, once with the thresholds computed by means of our primary algorithm, and once with the thresholds computed by means of our new algorithm. We evaluated our threshold algorithms on the basis of the shape parameters extracted from the segmented volumes. Our new algorithm gives useful segmentation thresholds for almost all problem volumes for which our primary algorithm fails.