The proper separation of hepatic veins is crucial for discriminative visualization during surgical planning and intraoperative navigation, automated analysis and liver segment approximation. However, due to several reasons liver and portal vein may be connected at some points and may contain loops. In this work, we propose a novel graph-based method to separate interconnected hepatic veins. Furthermore, we show how to transfer the graph-based separation results back to the imaging data to improve the initial segmentation and discriminate portal and liver vein. In addition, we propose a majority voting operation to correct some segmentation imperfections. Promising qualitative and quantitative evaluation results are presented.