主 办:北 京 中 医 药 大 学
ISSN 1006-2157 CN 11-3574/R

北京中医药大学学报 ›› 2021, Vol. 44 ›› Issue (4): 358-365.doi: 10.3969/j.issn.1006-2157.2021.04.011

• 科技之窗 • 上一篇    下一篇

以六种疾病为例研究基于统计注意力的神经网络模型在证名诊断中的应用*

薛哲1, 赵宗耀1, 陈家旭1#, 刘玥芸1, 王喜红2, 许梦白3, 董硕1, 李同同1, 王君1   

  1. 1 北京中医药大学中医学院 北京 100029;
    2 北京中医药大学东方医院;
    3 北京中医药大学东直门医院
  • 收稿日期:2020-10-04 发布日期:2021-04-29
  • 通讯作者: #陈家旭,男,博士,教授,博士生导师,主要研究方向:中医证候的生物学基础, E-mail:chenjiaxu@hotmail.com
  • 作者简介:薛哲,女,博士,讲师
  • 基金资助:
    *国家自然科学基金青年基金项目(No.81803972)

Applied research of SANN model in the diagnosis of traditional Chinese patterns with six diseases data as examples*

Xue Zhe1, Zhao Zongyao1, Chen Jiaxu1#, Liu Yueyun1, Wang Xihong2, Xu Mengbai3, Dong Shuo1, Li Tongtong1, Wang Jun1   

  1. 1 School of Traditional Chinese Medicine, Beijing University of Chinese Medicine, Beijing 100029, China;
    2 Dongfang Hospital of Beijing University of traditional Chinese Medicine, Beijing 100029, China;
    3 Dongzhimen Hospital of Beijing University of traditional Chinese Medicine, Beijing 100029, China
  • Received:2020-10-04 Published:2021-04-29
  • Contact: Prof. Chen Jiaxu, 11 Beisanhuan Donglu, Chaoyang District, School of Chinese Medicine, Beijing University of Chinese Medicine, Beijing 100029. E-mail: chenjx@ bucm.edu.cn
  • Supported by:
    National Natural Science Youth Foundntion(No.81803972)

摘要: 目的 研究基于统计注意力的神经网络(SANN)模型在中医证名诊断中的适用性与先进性,探讨其生成的特征贡献度是否符合中医原理。方法 选择记载于古今医案云平台及中医药杏林园数据库的高血脂、更年期综合征、冠心病、慢性胃炎、慢性肾炎、尿路感染、脂肪肝病案共1 110例。通过人工神经网络(ANN)、随机森林(RF)、支持向量机(SVC)、K-近邻(KNN)、SANN分别建立诊断模型,对比5种模型评价指标。评价指标包括Macro-F1、Macro-Precision、Macro-Accuracy、Macro-Recall。结果 SANN在6种疾病中的Macro-F1平均值为0.78、Macro-Precision平均值为0.79、Macro-Accuracy平均值为0.79、Macro-Recall平均值为0.8,均优于其他4种基准模型,其参数可解释性与导出的特征对类支持度符合中医原理。结论 SANN在中医证名诊断智能化、中医数据的特征筛选、疾病量表研制等任务中具有适用性与先进性,为相关工作提供了创新性的方法参考。

关键词: 基于统计注意力的神经网络, 证名诊断, 特征筛选

Abstract: Objective To study the applicability and strength of the statistical attention-based neural network (SANN) model in the diagnosis of TCM patterns, and to explore whether the generated feature contribution is aligned with TCM principles. Methods A total of 1,110 cases of hyperlipidemia, menopausal syndrome, coronary heart disease, chronic gastritis, chronic nephritis, urinary tract infection, and fatty liver recorded in the ancient and modern medical records cloud platform and the Chinese medicine Xinglinyuan database were selected. Diagnostic models were established through artificial neural network (ANN), random forest (RF), support vector machine (SVC), K-nearest neighbor (KNN), and statistical attention-based neural network model (SANN) respectively. Evaluation indicators include Macro-F1, Macro-Precision, Macro-Accuracy, and Macro-Recall. Results The statistical attention-based neural network model (SANN) in the 6 diseases has an average of Macro-F1 at 0.78, Macro-Precision at 0.79, Macro-Accuracy at 0.79, and Macro-Recall at 0.8, which were better than the other 4 benchmark models. The interpretability of its parameters and the support of derived features conformed to the principles of Chinese medicine. Conclusion The neural network model based on statistical attention (SANN) is applicable and advanced in undertaking tasks such as the intelligent diagnosis of TCM patterns, feature screening of TCM data, and the development of disease evaluation scales, thus providing an innovative methodological reference for related study.

Key words: Statistical Attention-based Neural Network(SANN), diagnosis of patterns, feature screening

中图分类号: 

  • R241.9