Fast blood vessel segmentation for surgical and interventional planning and navigation

A fast segmentation of tubular tree structures like vessel systems in volumetric datasets is of vital interest for many medical applications. This is especially valid for planning and navigation in catheter based interventions and liver resection surgical procedures. For catheter based navigation, blood vessel segmentation helps in planning the placement of stents and valves, and also in navigating to the desired location. For liver resection, a good visualisation of the blood vessels in relation to tumours provides better information about spatial relationship, which is very important in planning an optimal resection plane. Also a fast blood vessel segmentation will make it feasible to update the models intra-operatively. In this thesis, we have developed a couple of methods for fast and user-friendly blood vessel segmentation. The methods work from a user-initiated seed, by tracking and segmenting the blood vessels to the ends of the vessel tree using a local structure analysis approach. To perform the structure analysis, we propose novel multiscale modified vesselness and circleness filters. The bifurcation cross-sections of the blood vessel were found by either detecting multiple peaks in the filtering output, or by estimating the sudden change in radius of the bifurcation, or by estimating significant change in the compactness and radius variance of the vessel cross-section. The novelty in our final algorithm is in performing the whole blood vessel segmentation by use of only 2D analysis on the blood vessel cross-sections, which makes it faster than performing a 3D image analysis. Our methods were validated using synthetic as well as medical images, and also by clinically testing the method on liver hepatic and portal vein segmentation. The results have shown that the methods work in just seconds for images related to catheter navigation and in a couple of minutes for liver resection planning images. On medical validation of the liver blood vessel segmentation, our method detected 100% of blood vessels at and above 3mm radius and 80% at 2.5mm radius, which are the most clinically relevant blood vessels for liver resection planning