A Liver Segmentation Method Based on the Fusion of VNet and WGAN

Abstract

Accurate segmentation of liver images is an essential step in liver disease diagnosis, treatment planning, and prognosis. In recent years, although liver segmentation methods based on 2D convolutional neural networks have achieved good results, there is still a lack of interlayer information that causes severe loss of segmentation accuracy to a certain extent. Meanwhile, making the best of high-level and low-level features more effectively in a 2D segmentation network is a challenging problem. Therefore, we designed and implemented a 2.5-dimensional convolutional neural network, VNet_WGAN, to improve the accuracy of liver segmentation. First, we chose three adjacent layers of a liver model as the input of our network and adopted two convolution kernels in series connection, which can integrate cross-sectional spatial information and interlayer information of liver models. Second, a chain residual pooling module is added to fuse multilevel feature information to optimize the skip connection. Finally, the boundary loss function in the generator is employed to compensate for the lack of marginal pixel accuracy in the Dice loss function. The effectiveness of the proposed method is verified on two datasets, LiTS and CHAOS. The Dice coefficients are 92% and 90%, respectively, which are better than those of the compared segmentation networks. In addition, the experimental results also show that the proposed method can reduce computational consumption while retaining higher segmentation accuracy, which is significant for liver segmentation in practice and provides a favorable reference for clinicians in liver segmentation.

A Lightweight Convolutional Neural Network Model for Liver Segmentation in Medical Diagnosis

“…The training and validation of Deep stacked auto encoder (DSAE) are performed on 2 publicly available datasets (3Dircadb and Sliver'07); the limitations of this method against our methods are obvious that DSAE cannot detect the abnormal liver in the image very well but our method can efficiently segment the abnormal liver very efficiently. In [ 46 ], the DSC of liver segmentation was noted 92% using LiTS17 dataset that is less than the DSC in our proposed work. In [ 35 ], the two datasets, SLiver'07 and 3Dircadb01, that were used to segment the liver show the DSC of 94.80% and 91.83%, respectively, which are obviously much fewer than Ga-CNN method.…”

Automated liver tissues delineation techniques: A systematic survey on machine learning current trends and future orientations

“…To overcome the lack of interlayer information in 2D CNN models that can cause profound loss of segmentation performance, the authors in Ma et al (2021) develop a 2.5-D VNet_WGAN. Moving on, to avoid the sensitivity of liver segmentation models to heterogeneous pathologies and fuzzy boundaries, mainly when the data is scarce, a 3D CNN and a hybrid loss function are deployed in Tan et al (2021).…”

A Review Paper about Deep Learning for Medical Image Analysis

“…The GAN-based approach yielded an average DSC of 95.3% while evaluating 378 CT images. A V-Net and Wasserstein GAN-based model was explored by Ma et al [ 47 ] to improve the efficiency of liver segmentation. The WGAN [ 48 ] model includes Wasserstein distance to fix the issue of GAN training instability.…”