基于边缘监督的肝部超声图像包膜分割网络

Liver capsule segmentation in ultrasound image using edge supervision based network

肝纤维化、肝硬化的早期发现对临床治疗和预后评估具有重要意义。而肝包膜的形态和纹理特征是计算机辅助肝硬化诊断的重要依据。本文提出一种基于边缘监督的肝部超声图像包膜分割网络。该网络以常用的分割模型UNet为基础,引入空洞卷积,扩大感受野;同时,添加了边缘监督模块,从而将特征学习主要聚焦在图像梯度较大的部分;此外,还设计了混合加权损失函数,来缓解肝包膜部分与其他区域之间的极度不平衡情况。实验结果表明,本文提出的ES-UNet网络结构平均Dice系数相比原始UNet提高了0.1715,平均交并比(MIo U)提高了0.0215,其他指标也有较明显的提高,可见,本文算法的各个组件对模型分割性能的优化都有一定的贡献,改进后的模型可以实现肝包膜的精确分割。

The early detection of liver fibrosis and liver cirrhosis is of great significance for clinical treatment and prognosis evaluation, and the morphological and texture characteristics of liver capsule are important for the computer-assisted diagnosis of liver cirrhosis. An edge supervision based network(ES-UNet) is proposed for liver capsule segmentation in ultrasound images. Based on the commonly used segmentation model(UNet), ES-UNet uses atrous convolution to expand the receptive field, and edge supervision module to focus the feature learning on the region with large image gradient. In addition, a mixed weighted loss function is used to reduce the extreme imbalance between the liver capsule and other regions.The experimental results show that compared with those of original UNet model, the average Dice coefficient of ES-UNet is increased by 0.171 5, and the mean intersection over union is higher by 0.021 5, and the other indicators are also elevated significantly, indicating that each component of the proposed algorithm has a certain contribution to the optimization of model segmentation performance. The improved model can achieve accurate liver capsule segmentation.