Magisterarbeit Robust Segmentation of Tubular Structures in 3D Volume Data

Segmentation of tubular structures like blood vessels and airways in 3D volume data is of vital interest for medical applications like diagnosis and surgical planning. The aim of this diploma thesis is to facilitate the efficient analysis of vessels by developing an automatic segmentation method. First, the method uses a vessel detection filter, which is based on a novel multiscale medialness function. The filter allows to distinguish between tube-like and other structures and provides an estimate of the tube’s radius. Second, centerlines of the tubes are extracted and the vessel tree is reconstructed by taking the physiological properties of the vessels into account. The centerline and radius information is further used to build an initial tube representation. Third, the final segmentation step uses the tube representation to initialize and constrain a level set method for tubular structures. Computer generated phantom data sets are used for evaluation of different levels of known properties of CT data sets. Results show the robustness of the developed method against noise and anisotropic voxels. Finally, experiments with two real CT data sets demonstrate the applicability of the method to different tubular structures.