Abstract
Liver vessel segmentation from computed tomography (CT) images is important in clinical liver surgical planning. However, liver vessel segmentation is a challenging task due to the low quality of vessel information in the CT images, which leads to errors in vessel detection and vessel type classification due to the complex vessel structures, such as portal vein, and hepatic vein to the goal of this thesis is to improve the vessel segmentation result and classify the liver venous vessel into meaningful part. We propose an integrated framework for reconstructing 3D liver venous vessel model from 3D CT images. The proposed framework consists of vessel detection, vessel connectivity, vessel classification and vessel radius refinement. First, we employ the tubular-filter based approach to detect vessel structure inside the CT images and construct the reasonable vessel tree structure to bridge all the gaps between vessels by using the proposed similarity score. Then, we apply the random walker algorithm with simple user interaction to classify the liver venous vessel into portal vein and hepatic vein. Finally, we refine the vessel segmentation result by estimating vessel radius with vessel tracing and curve fitting. We evaluate the proposed algorithm on 20 CT datasets and experimental results show that our algorithm improves the mutual overlap rate by 7.57% when compared to the original tubular filter.