Diploma and Master Theses (authored and supervised):
"Coronary Artery Tracking with Rule-based Gap Closing";
Supervisor: E. Gröller, K. Bühler, S. Zambal;
Institut für Computergraphik und Algorithmen,
Coronary artery diseases are among the leading causes of death in the industrial countries. The high death rate leads to an increased demand of diagnosis and treatment of these diseases. Additional to the conventional coronary angiography, the CT angiography is mainly used in the extended diagnostics of coronary artery diseases. This modality allows a detailed assessment of the coronary vessels and potentially present stenoses. For supporting the radiologist during the evaluation of the coronary arteries by the help of computer-aided diagnostic methods, robust and e cient procedures for the tracking of coronary arteries are needed. The approach presented in this thesis uni es the strong points of existing methods delivering high accuracy with the strong points of methods achieving high overlap. Therefore the approach presented in this thesis aims at highly accurate results in combination with high overlap of the investigated coronary artery vessel tree. The approach is divided into three phases: 1) calculation of seed points, 2) tracking of vessel segments, and 3) construction of the coronary artery trees. Phase 1 & 2 are executed in an automatic manner. First potential seed points for the tracking of vessel segments are identi ed. During the second phase, vessel segments located at these seed points are tracked by use of a cylindrical shape model. By use of rule-based anatomical heuristics, the third and nal phase combines vessel segments to form complete coronary artery trees. This phase requires minimal user interaction, as the location of the root of the left and right coronary artery tree needs to be speci ed. Beside the detailed description of the algorithm, the integration into a professional radiology workstation is demonstrated. The results obtained by the evaluation on 24 CTA datasets show a high overlap (OV) of 89.5% in combination with very precise accuracy (AI) of 0.24 mm in comparison to an expert-annotated reference segmentation.
Electronic version of the publication:
Created from the Publication Database of the Vienna University of Technology.