Extraction of hepatic veins in contrast enhanced CT with application to interventional planning

The Liver performs several important tasks that are essential for survival. However, liver cancer, the third most common type of cancer, affects these functions significantly. Different treatment options are available, but a surgical resection, if possible, offers the best prognosis for the patient. Thus, the decision, whether a surgical resection is feasible, is important and must be taken with care in a pre-interventional planning stage. Modern volumetric imaging techniques such as CT or magnetic resonance imaging (MRI) are utilized to decide which treatment is best for the patient and to plan the intervention. However, the amount of anatomical details visible in the acquired volumes is steadily increasing. This comes along with an increasing amount of data per patient. Manual examination is time consuming and prone to errors. As a matter of fact, several software systems were proposed to support the surgeon during the planning phase. The extraction of blood vessels plays an important role in these applications. The segmentation of vessels is a challenging problem that has to deal with acquisition-dependent problems such as noise, contrast, spatial resolution, and artifacts. Furthermore, blood vessel specific characteristics like high variability of size and curvature result in additional difficulties for segmentation algorithms. The liver, in particular, exhibits another challenge to vessel segmentation algorithms. Its supply and drain vessel systems are densely distributed within the liver, and because of partial volume effects and motion artifacts, they seem to be connected at some points. The focus of the present thesis is the robust extraction of hepatic veins in multiphase CT volumes. Therefore, an image processing pipeline is presented that covers vessel enhancement, vessel segmentation, graph creation and tree reconstruction. The pipeline was used to develop an application for interventional planning. It allows for the simulation of intraoperative hepatic vein clamping for (sub-)segment oriented liver resections and the execution of risk analysis to judge surgical risk during an atypical resection. Furthermore, results of the present thesis were also successfully used in an application for intraoperative navigation to extract liver vessels in 3D ultrasound data and matching of anatomical vessel trees and graphs of the liver for registration of 3D volumes