Automatic annotation of liver computed tomography images based on a vessel-skeletonization method

Anatomical analysis of liver region is an essential and key step for liver-related disease diagnosis and treatment. One of the challenging issues is to annotate the functional regions of liver automatically or semi- automatically by analyzing Computed Tomography (CT) images. The present study developed a complete liver annotation system with an improved vessel-skeletonization method is proposed for CT images. In the first step, an automatic level set method and a customized region-growing method are applied to extract the liver region including vessels and tumors. Next, a modified iterative thinning method is developed to obtain the geometric structure of liver vessels and mark a vessel skeleton. The three-dimensional information is transformed into a tree data structure for storage. Based on the branch distribution of portal vein skeleton, a model-based method with a modified nearest neighbor segment approximation (NNSA) algorithm is adopted for the functional liver anatomy. Three experiments involving five 64-row liver CT datasets are performed. The accuracies of segmentation and annotation results were validated by an experienced doctor. Compared with different methods, our proposed vessel skeletonization method can simultaneously preserve the connectivity of the vasculature topology and generate the skeleton in a shorter time. Furthermore, our proposed annotation system can provide both visual and measurable information of livers. These experimental results demonstrate the usefulness and effectiveness of our proposed method. Our liver annotation system is helpful to evaluate the function of liver system and support diagnosis of liver disease.