第15卷第3期 智能系统学报 Vol.15 No.3 2020年5月 CAAI Transactions on Intelligent Systems May 2020 D0L:10.11992tis.201811015 融合迁移学习和神经网络的皮肤病诊断方法 商显震，韩萌，王少峰，贾涛，许冠英 (北方民族大学计算机科学与工程学院，宁夏银川750021) 摘要：针对医学特征对患者病情发展的时间顺序无法有效表达，医学特征构建工作耗费大量人工成本，以及 皮肤病数据样本数量较少等问题，提出了融合迁移学习和神经网络的皮肤病轴助诊断方法。该方法将T©xtL STM(long short term memory neural network for text),TextCNN(convolutional neural network for text)RCNN(re- current convolutional neural networks for text classification)等3种基于神经网络的文本分类模型应用于皮肤病辅助 诊断，同时融入迁移学习技术，能够在一定程度上将皮肤病专业书籍中的理论知识迁移到诊断模型中。在皮肤 病多分类实验中，本文方法的正确率优于对比方法；在皮肤病二分类实验中，本文方法的召回率优于对比方 法。迁移学习对实验结果的积极影响率高于75%。 关键词：皮肤病诊断：神经网络：迁移学习；文本分类：卷积神经网络：循环神经网络：长短期记忆网络：辅助诊断 中图分类号：TP391.1文献标志码：A文章编号：1673-4785(2020)03-0452-08 中文引用格式：商显震，韩萌，王少峰，等.融合迁移学习和神经网络的皮肤病诊断方法J.智能系统学报，2020,15(3)： 452-459. 英文引用格式：SHANG Xianzhen,,HAN Meng,WANG Shaofeng,etal.A skin diseases diagnosis method combining transfer learn- ing and neural networks J.CAAI transactions on intelligent systems,2020,15(3):452-459. A skin diseases diagnosis method combining transfer learning and neural networks SHANG Xianzhen,HAN Meng,WANG Shaofeng,JIA Tao,XU Guanying (School of Computer Science and Engineering,North Minzu University,Yinchuan 750021,China) Abstract:To address the problems that medical features can not effectively express the chronological order of a patient' s condition,feature construction incurs high labor costs,and the number of diagnosed cases of skin diseases is relatively low,this study proposes binary classification and multi-classification diagnostic methods based on neural network and transfer learning of multisource data for diagnosing skin diseases.The text classification model based on three neural network models,namely,TextLSTM(long short term memory neural network for text),TextCNN(convolutional neural network for text),and RCNN(recurrent convolutional neural networks for text classification),is applied to dermatologic- al auxiliary diagnosis.At the same time,the method incorporates transfer learning,which can transfer theoretical know- ledge of skin diseases obtained from books to the diagnostic models to a certain degree.Results show that the accuracy rate of the multi-classification diagnostic method is higher than that of the binary classification diagnostic method.By contrast,the recall rate of the binary classification diagnostic method is higher than that of the multi-classification dia- gnostic method.Thus,transfer learning has a positive effect on more than 75%of the experimental results. Keywords:skin disease diagnosis;neural network;transfer learning;text classification;convolutional neural network; recurrent neural network;long short term memory neural network;auxiliary diagnosis 皮肤病是一种常见病和多发病山，不但使患 收稿日期：2018-11-21. 基金项目：国家自然科学基金项目(6156300I):宁夏自然科学 者承受生理的病痛，而且给患者的社会生活造成 基金项目NZ17115):计算机应用技术宁夏回族自治 许多负面影响。针对皮肤病的诊断方法的相关研 区重点学科项目(PY1703). 通信作者：韩萌.E-mail:2003051@nun.edu.cn 究具有重要的意义。现有的很多疾病诊断方法DOI: 10.11992/tis.201811015 融合迁移学习和神经网络的皮肤病诊断方法 商显震，韩萌，王少峰，贾涛，许冠英 （北方民族大学 计算机科学与工程学院，宁夏 银川 750021） 摘 要：针对医学特征对患者病情发展的时间顺序无法有效表达，医学特征构建工作耗费大量人工成本，以及 皮肤病数据样本数量较少等问题，提出了融合迁移学习和神经网络的皮肤病辅助诊断方法。该方法将 TextL￾STM(long short term memory neural network for text)、TextCNN(convolutional neural network for text) 以及 RCNN(re￾current convolutional neural networks for text classification) 等 3 种基于神经网络的文本分类模型应用于皮肤病辅助 诊断，同时融入迁移学习技术，能够在一定程度上将皮肤病专业书籍中的理论知识迁移到诊断模型中。在皮肤 病多分类实验中，本文方法的正确率优于对比方法；在皮肤病二分类实验中，本文方法的召回率优于对比方 法。迁移学习对实验结果的积极影响率高于 75%。 关键词：皮肤病诊断；神经网络；迁移学习；文本分类；卷积神经网络；循环神经网络；长短期记忆网络；辅助诊断 中图分类号：TP391.1 文献标志码：A 文章编号：1673−4785(2020)03−0452−08 中文引用格式：商显震, 韩萌, 王少峰, 等. 融合迁移学习和神经网络的皮肤病诊断方法 [J]. 智能系统学报, 2020, 15(3): 452–459. 英文引用格式：SHANG Xianzhen, HAN Meng, WANG Shaofeng, et al. A skin diseases diagnosis method combining transfer learn￾ing and neural networks[J]. CAAI transactions on intelligent systems, 2020, 15(3): 452–459. A skin diseases diagnosis method combining transfer learning and neural networks SHANG Xianzhen，HAN Meng，WANG Shaofeng，JIA Tao，XU Guanying (School of Computer Science and Engineering, North Minzu University, Yinchuan 750021, China) Abstract: To address the problems that medical features can not effectively express the chronological order of a patient’ s condition, feature construction incurs high labor costs, and the number of diagnosed cases of skin diseases is relatively low, this study proposes binary classification and multi-classification diagnostic methods based on neural network and transfer learning of multisource data for diagnosing skin diseases. The text classification model based on three neural network models, namely, TextLSTM(long short term memory neural network for text), TextCNN(convolutional neural network for text), and RCNN(recurrent convolutional neural networks for text classification), is applied to dermatologic￾al auxiliary diagnosis. At the same time, the method incorporates transfer learning, which can transfer theoretical know￾ledge of skin diseases obtained from books to the diagnostic models to a certain degree. Results show that the accuracy rate of the multi-classification diagnostic method is higher than that of the binary classification diagnostic method. By contrast, the recall rate of the binary classification diagnostic method is higher than that of the multi-classification dia￾gnostic method. Thus, transfer learning has a positive effect on more than 75% of the experimental results. Keywords: skin disease diagnosis; neural network; transfer learning; text classification; convolutional neural network; recurrent neural network; long short term memory neural network; auxiliary diagnosis 皮肤病是一种常见病和多发病[1] ，不但使患 者承受生理的病痛，而且给患者的社会生活造成 许多负面影响。针对皮肤病的诊断方法的相关研 究具有重要的意义。现有的很多疾病诊断方法[2-4] 收稿日期：2018−11−21. 基金项目：国家自然科学基金项目 (61563001)；宁夏自然科学 基金项目 (NZ17115)；计算机应用技术宁夏回族自治 区重点学科项目 (PY1703). 通信作者：韩萌. E-mail: 2003051@nun.edu.cn. 第 15 卷第 3 期 智 能 系 统 学 报 Vol.15 No.3 2020 年 5 月 CAAI Transactions on Intelligent Systems May 2020