Semiautomated Editing of Vessel Segmentation Masks

This thesis describes a technique for editing segmentation results of vessels, which should enhance usage and reduce work duration for physicians by using a simple and fast way of interaction. Moreover also a quick calculation of an accurate result was of primary interest. Since vascular structures are vulnerable to diseases, vessels are the main focus of this thesis. Nowadays, Image Analysis is able to facilitate the medical diagnosis procedure. Since stroke treatment is time-crucial, appropriate algorithms should be fast and enable an accurate depiction of the arteries to simplify the diagnostic process. However, because automatic segmentation is often quite inaccurate and manual segmentation is tedious, neither of these two methods alone is often adequate for usage. Because of this we suggest to combine the fast automatic segmentation and the exact manual editing done by clinical experts. To reduce effort and working time of the medical staff, this thesis describes different techniques, which were developed to modify and, more importantly, to improve the segmentation results. The segmentation mask can be altered as its components can be separately removed and independent elements can be connected. A framework was implemented, with which a user is able to perform these tasks interactively. The deletion process is supported by various metrics, which enable the search and removal of similar structures. Also this framework assists the reconnection of vessels by finding the most likely connection by the means of image intensities and their gradients. The main goal of this thesis was to facilitate and accelerate the editing process by implementing fast semi-automatic algorithms. Intuitive interaction methods also had a major impact on the design.