Robust and efficient liver and tumor segmentation segmentation tools from CT images are important for clinical decision-making in liver treatment planning and response evaluation. In this work, we report recent advances in an ongoing project Liver Workbench which aims to provide a suite of tools for the segmentation segmentation, quantification and modeling of various objects in CT images such as the liver, its vessels and tumors. Firstly, a liver segmentation segmentation approach is described. It registers a liver mesh model model to actual image features by adopting noise-insensitive flipping-free mesh deformations. Next, a propagation learning approach is incorporated into a semi-automatic classification method for robust segmentation segmentation of liver tumors based on liver ROI obtained. Finally, an unbiased probabilistic liver atlas construction technique is adopted to embody the shape and intensity variation to constrain liver segmentation segmentation. We also report preliminary experimental results.