计算机科学与探索 ›› 2023, Vol. 17 ›› Issue (2): 303-323.DOI: 10.3778/j.issn.1673-9418.2208052

• 前沿·综述 • 上一篇    下一篇

深度学习在舌象分类中的研究综述

吴欣,徐红,林卓胜,李胜可,刘慧琳,冯跃   

  1. 1. 五邑大学 智能制造学部,广东 江门 529020
    2. 维多利亚大学,澳大利亚 墨尔本 8001
    3. 上海中医药大学 教学实验中心,上海 201203
  • 出版日期:2023-02-01 发布日期:2023-02-01

Review of Deep Learning in Classification of Tongue Image

WU Xin, XU Hong, LIN Zhuosheng, LI Shengke, LIU Huilin, FENG Yue   

  1. 1. Faculty of Intelligent Manufacturing, Wuyi University, Jiangmen, Guangdong 529020, China
    2. Victoria University, Melbourne 8001, Australia
    3. Experiment Center of Teaching & Learning, Shanghai University of Traditional Chinese Medicine, Shanghai 201203, China
  • Online:2023-02-01 Published:2023-02-01

摘要: 随着技术的快速发展和计算能力的提高,深度学习在舌象分类领域将得到广泛的应用。舌诊图像的舌象分类是中医舌诊客观化的重要组成部分。传统的舌诊是在基础理论的指导下,借助个人经验所做出的理解和判断,因而会具有一定的差异性和模糊性,影响诊断的可重复性。为了减少主观判断的误差,许多研究人员致力于通过深度学习实现中医舌诊的客观化、定量化和自动化。主要对基于深度学习的舌象分类方法研究现状进行分析梳理和归纳总结。在舌象分类研究中,以各类深度学习方法作为研究对象,将其划分为基于早期神经网络、卷积神经网络、区域卷积神经网络、迁移学习以及其他方法进行总结分析;对舌诊中的中医证候和疾病以及体质分类进行了讨论;用Kaggle上的公开舌诊数据集进行5折交叉验证实验,数据集为小样本齿痕舌,评估了基于深度学习和迁移学习分类方法;对舌诊图像质量、构建数据集方式、特征提取、单标签和多标签分类的研究发展进行了探讨和展望。

关键词: 中医舌诊, 深度学习, 舌象分析, 舌象分类

Abstract: With the rapid development of technology and the improvement of computing power, deep learning will be widely used in the field of tongue classification. The classification of tongue image is an important part of tongue diagnosis in traditional Chinese medicine (TCM). Traditional tongue diagnosis is dependent on understanding and judgment skills gained from personal experience under the guidance of basic theory, which leads to ambiguity and variability, affecting diagnostic reproducibility. In order to reduce the error of subjective judgment, many researchers have devoted themselves to realizing the objectification, quantification and automation of tongue diagnosis in TCM through deep learning. This paper analyzes and summarizes the research status of tongue image classification methods based on deep learning. In the study of tongue image classification, various deep learning methods are used as the research objects. The research objects are divided into categories based on early neural networks, convolutional neural networks, regional convolutional neural networks, transfer learning and other methods for summary analysis. TCM syndromes/diseases in tongue diagnosis and classification of physical constitution are discussed. A 5-fold cross-validation experiment is conducted with the public tongue diagnosis dataset on Kaggle. The dataset is a small sample of the dentate tongue, and the classification methods based on deep learning and transfer learning are evaluated. The research development of single-label and multi-label classification is discussed and prospected.

Key words: traditional Chinese medicine tongue diagnosis, deep learning, tongue image analysis, tongue image classification