基于全卷积网络的肝脏CT语义分割  

Semantic Segmentation of Liver CT based on Full Convolutional Network

作　　者：徐婷宜 朱家明[1] 李祥健 XU Tingyi;ZHU Jiaming;LI Xiangjian(College of Information Engineering,Yangzhou University,Yangzhou 225000,China)

机构地区：[1]扬州大学信息工程学院,江苏扬州225000

出　　处：《软件工程》2020年第6期20-22,16,共4页Software Engineering

基　　金：国家自然科学基金项目(61873229).

摘　　要：肝脏CT(计算机断层扫描)图像分割为临床肝脏医疗分析提供了可靠依据。文中探索了完全卷积网络(FCN)用于肝脏CT图像中的检测和分割。FCN已被证明是用于语义分段的非常强大的工具,它能接受任意大小的的输入并通过有效地推理与学习产生相应大小的输出。该文将分类网络VGG调整为完全卷积网络,并通过迁移学习将其转移到分割任务,展示了由端到端,像素到像素训练的卷积网络语义分割。此架构能将来自深层粗糙层的语义信息与来自浅层精细层的外观信息相结合,以生成准确而精细的分割。本架构肝脏分割IOU值达到0.9,取得较好的分割效果。Computed Tomography(CT) image segmentation provides a reliable basis for clinical liver medical analysis. A Full Convolutional Network(FCN) is explored for detection and segmentation in liver CT images. FCN has been proven to be a very powerful tool for semantic segmentation. It can accept inputs of any size and generate corresponding output through effective reasoning and learning. This paper adjusts the classification network of Visual Geometry Group(VGG) to a fully convolutional network and transfers it to the segmentation task through transfer learning. It shows an end-to-end, pixel-to-pixel trained convolutional network semantic segmentation. This architecture can combine semantic information from deep rough layers with appearance information from shallow fine layers to generate accurate and fine segmentation. In this architecture, the liver segmentation Intersection-over-Union(IOU) value reaches 0.9, and a good segmentation effect is achieved.

关 键 词：全卷积网络 语义分割 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]