Liver Segmentation Using Deep Learning

Abstract

Throughout the most recent couple of years, major breakthroughs were achieved in many computer visions tasks, such as image classification and segmentation by using the application of deep learning. The programmed liver division from Computed Tomography (CT) pictures has become a significant territory in clinical research, including radiotherapy, liver volume estimation, and liver transplant medical procedures. This research introduces a framework for the automated segmentation of liver and lesions in images of the CT and Magnetic Resonance Images (MRI) abdomen using Cascaded Fully Convolutional Neural (CFCN) networks for the segmentation of large-scale medical trials and quantitative image analysis. We train and cascade two Fully Convolutional Neural (FCN) networks for the combination of liver segmentation and lesions. We train an FCN as a first step to segment the liver as an input of Region of Interest (ROI) for a second FCN. The second FCN segments only lesions inside phase one's estimated liver ROIs. CFCN models have been trained on a 100-volume abdominal CT dataset. Validation tests on additional data sets show that linguistic liver and lesion segmentation based on CFCN achieves Dice scores above 94% for the liver with computation times below 100s per volume. On 38 MRI liver tumor volumes and the public 3DIRCAD dataset, we further experimentally demonstrate the robustness of the proposed method.