Segmentation and 3D modeling of liver vessel system

Abstract

In this paper we describe a new approach for segmentation of liver from CT images and further the segmentation of liver vessels to create a visualization model for surgical and modeling purposes. Since usual approaches, based on density models or edge detection, don't work well for liver, we investigate the local density distribution of the liver texture to classify each pixel, whether it lies on the liver-background boundary or outside it. The classifier outputs the boundaries of the liver in each slice, which are used then to create the organ volume. Vessels are segmented then inside the liver volume using a single automatically selected threshold. The result is morphologically closed and smoothed by a Gaussian kernel then. In the last step we produce a 3D skeleton of the vessel system to investigate its topology.