摘要
[目的/意义]介绍深度学习的概念、发展过程以及三种典型深度学习模型:卷积神经网络(CNN)、深度信念网络(DBN)和堆叠自动编码器(SAE),并梳理三种模型的发展历程,对深度学习在医学图像分析中的应用研究状况进行综述。[方法/过程]通过文献调研对深度学习模型在医学图像分析领域的诸多应用进行整理,并在此基础上突出一些关键的应用领域,进而讨论深度学习基本模型存在的问题。[结果/结论 ]深度学习模型目前存在五个问题:模型结构单一、训练方式仍需改进、训练时间过长、对无标记数据添加标签、克服对抗样本,在实际工作中应提出相关解决措施。从跨组织合作、大的图像数据集、深度学习方法的进步三个方面对深度学习在医学图像分析领域的发展前景进行展望。
[Purpose/significance]This paper introduces the concept and development process of deep learning and three typical deep learning models:convolutional neural network(CNN),deep belief network(DBN)and stacked automatic encoder(SAE).[Method/process] Based on the literature research,this paper sorts out the application of deep learning model in the field of medical image analysis,highlights some key application fields on this basis,and then discusses the problems existing in the basic model of deep learning.[Result/conclusion] At present,the deep learning model has five problems:the model structure is simple,the training method still needs to be improved,the training time is too long,the non-label data are labeled,and the counter samples are removed,all of which need to be solved in practice.Finally,the authors predict the future of deep learning in the field of medical image analysis in terms of cross-organizational cooperation,large image data sets and the progress of deep learning methods.
作者
黄江珊
王秀红
Huang Jiangshan;Wang Xiuhong(Institute of Science and Technology Information,Jiangsu University,Zhenjiang 212013,China)
出处
《图书情报研究》
2019年第2期92-98,112,共8页
Library and Information Studies
关键词
深度学习
医学图像分析
卷积神经网络
深度信念网络
堆叠自动编码器
deep learning
medical image analysis
convolutional neural network
deep belief network
stacked automatic encoder
