A Joint Deep Learning Approach for Automated Liver and Tumor Segmentation

Abstract

Hepatocellular carcinoma (HCC) is the most common type of primary liver cancer in adults, and the most common cause of death of people suffering from cirrhosis. The segmentation of liver lesions in CT images allows assessment of tumor load, treatment planning, prognosis and monitoring of treatment response. Manual segmentation is a very time-consuming task and, in many cases, prone to inaccuracies. Therefore, automatic tools for tumor detection and segmentation are highly desirable. We propose a network architecture that consists of two consecutive nested fully convolutional neural networks together with a joint minimization strategy. The first sub-network segments the liver whereas the second sub-network segments the actual tumor inside the liver. We compare the nested network architecture to a one-step approach, where a neural network performs both segmentation tasks simultaneously. Both architectures are trained on a subset of the LiTS (Liver Tumor Segmentation) Challenge and evaluated on data provided from the radiological center in Innsbruck. The nested approach is shown to significantly outperform the one-step network in terms of various accuracy measures.