Liver segmentation in CT images is essential in many clinical applications, such as pathological diagnosis of liver disease, visualization, surgical planning, image-guided surgery, volumetric measurement to estimate liver function reserve and select the appropriate treatment. However, segmentation of the liver is still a very difficult task due to the complex context, fuzzy and blurred boundaries between abdominal organs, various appearance of liver, scale variety, the presence of artifacts, and the low contrast between the liver and the surrounding organs and tissues. Recently, Deep models have been very successful in many tasks such as detection, classification and segmentation on non-medical images due to the sufficient availability of labeled data. In the medical field, labeled data is still limited due to privacy concerns and the costly need for experts in the labeling process. In this paper, to address these challenges, we present a new automatic and efficient model combining efficient pre-processing and a based capsules network. Our proposed network is an end-to-end learning process that can eliminate redundant computations and reduce the risk of over-fitting on limited training data. The model has been validated on 3D liver segmentation CHAOS challenges and Sliver07 and have obtained competitive segmentation results compared to state-of-the-art on such a limited amount of labeled data.
Volume 12 | 04-Special Issue
Pages: 1016-1024
DOI: 10.5373/JARDCS/V12SP4/20201574