Abstract—Automatic liver segmentation in CT images is an important step for computer-aided diagnosis and computeraided hepatic surgery. Recently, though numerous methods based on deep learning such as U-Net have been proposed for automatic liver segmentation, it is still a challenging topic because of its low contrast and variations of liver shape. Additionally, limited training data for deep learning is another challenging problem. In this paper, we propose an automatic liver segmentation using U-Net with a Wasserstein Generative Adversarial Network (GAN). The Wasserstein GAN was used to improve U-Net’s training, especially training with a small data set. We demonstrated that liver segmentation accuracy (Dice value) with 33 and 392 training data sets was improved from 88% to 92% and from 92% to 93%, respectively.
Index Terms—liver, segmentation, deep learning, GAN, WGAN
Cite: Yuki Enokiya, Yutaro Iwamoto, Yen-Wei Chen, and Xian-Hua Han, "Automatic Liver Segmentation Using U-Net with Wasserstein GANs," Journal of Image and Graphics, Vol. 7, No. 3, pp. 94-101, September 2019. doi: 10.18178/joig.7.3.94-101
Copyright © 2012-2020 Journal of Image and Graphics, All Rights Reserved