基于对抗学习网络的半监督医学图像分割方法  

Semi-supervised Medical Image Segmentation Methods Based on Adversarial Learning Networks

作　　者：张粲 孟雪[1] 王常青 方瑞[1] 钟亚鼎[2] 

机构地区：[1]安徽医科大学生物医学工程学院,安徽合肥230032 [2]安徽医科大学第一附属医院放射科,安徽合肥230032

出　　处：《工业控制计算机》2021年第9期57-59,共3页Industrial Control Computer

基　　金：国家自然科学基金项目(62001005);安徽省高校自然科学研究项目(KJ2017A209);安徽省自然科学基金项目(2008085QH425);安徽医科大学科研基金项目(XJ201811)。

摘　　要：医学图像数据的特殊性使得基于深度学习的全监督算法在图像分割过程中或是缺乏必要的条件,或是成本过高,导致分割效果不佳。因此,诸多研究人员将半监督学习算法和生成对抗网络相结合,应用在医学图像分割任务中,进一步提高分割的精确度。分析了深度学习算法应用在医学图像分割领域中所面临的困难,概述了深度学习算法的分类、生成对抗网络、半监督学习算法及基于对抗学习网络的半监督模型的算法思想、相关原理、模型结构、相关的实现方式及研究进展,并对现有算法存在的问题进行思考,为未来医学图像的精确分割研究提供参考。The particularity of medical image data leads to poor segmentation effect.The fully supervised algorithm based on deep learning either lacks necessary conditions or costs too much in the process of image segmentation.Therefore,many researchers have combined semi-supervised learning algorithm with generative adversarial network in medical image segmentation to further improve the accuracy of segmentation.In this paper,the difficulties faced by deep learning algorithms in the field of medical image segmentation are analyzed.The classification of deep learning algorithms,generation adversarial networks,semi-supervised learning algorithms and semi-supervised models based on adversarial learning networks,algorithm ideas,relevant principles,model structure,relevant implementation methods and research progress are described.At the same time,the problems existing in the existing algorithms are considered.

关 键 词：深度学习 生成式对抗网络 半监督学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] R-05[自动化与计算机技术—计算机科学与技术]