Fast fully-automated model-driven liver segmentation utilizing slicewise applied levelsets on large CT data

Modern Computed Tomography scans are acquired down to a slice thickness of 0.5 mm thus yielding a huge number of 2D slices to be examined by the physician. Hence the need for automated computer assisted diagnostics, e.g. in the field of abdominal scans for liver tumor diagnostics and surgery planning, arises. In this work a fully-automated algorithm for robust and accurate segmentation of the liver parenchyma, a prerequisite for liver lobe classification and resection planning, is presented. A first estimate for liver segmentation is achieved by applying a normalized liver model to the CT data. Based on this pre-segmentation parameters for level set segmentation on a slice-by-slice strategy are assessed, thus enabling a fully-automated segmentation of the liver parenchyma. The slice-by-slice level set propagation utilizes fast-marching and threshold level set implementations.