李嘉敏 陈洪宇

摘要：目的 通过人工神经网络技术构建可根据症状体征预测用药的医案模型，以《临证指南医案·湿》医案为例，分析方证之间的网状关系。方法 对《临证指南医案·湿》医案进行筛选，将症状、体征、处方药物等数据规范化后录入。采用Python语言编程，PyBrain模块构建并训练神经网络模型，MatPlotLib模块绘制误差曲线、预测的拟合曲线，评估灵敏度与特异度，NetworkX模块实现方证网状关系的可视化表达，分析处方内的药物及配伍关系与其针对的病证病机或病理环节之间的关系。结果 构建的医案神经网络模型预测灵敏度96.15%、特异度75.00%，实现了方证网状关系的可视化映射及其单个、单组、两组节点间多角度的分析。结论 人工神经网络能较好模拟医案知识的方证关系，网状关系的可视化组合与呈现可为医案文献的方证知识发现提供可行方法。

关键词：医案；叶天士；临证指南医案；方证；人工神经网络；知识发现

DOI：10.3969/j.issn.1005-5304.2017.09.023

中图分类号：R2-05；R249 文献标识码：A 文章编号：1005-5304（2017）09-0091-05

Relationship Between Syndromes and Prescriptions of Damp Disease： a Neural Network-based Study on Cases from Lin Zheng Zhi Nan Yi An Shi LI Jia-min， CHEN Hong-yu （Guangxing Hospital Affiliated to Zhejiang Chinese Medical University， Hangzhou 310007， China）

Abstract： Objective Taking medical cases in Lin Zheng Zhi Nan Yi An Shi as examples to analyze the network relationship between the syndromes and prescriptions through building a medical case model forecasting medication via artificial neural networks for the syndromes and prescriptions in medical cases. Methods The study screened medical cases in Lin Zheng Zhi Nan Yi An Shi， and standardized and entered the data with Python language programming. PyBrain module was used to build and train a network model. The MatPlotLib module drew the error curve and the predicted fit curve， and evaluated the sensitivity and specificity. NetworkX module realized the visual expression of the network relationship between the syndromes and prescriptions， and analyzed the medicine within the prescriptions and compatibility relationship and the relationship between the pathogenesis and pathology. Results The sensitivity of the constructed medical case network model was 96.15% and the specificity was 75.00%. The visual mapping of the network relationship between the syndromes and prescriptions and the analysis on single， single group， and multi-angle were realized. Conclusion Neural network is capable to simulate the relationship between syndromes and prescriptions of medical knowledge. The visual combination and manifestation of network can provide a feasible solution for the knowledge discovery in medical literature.

Key words： medical cases； YE Tian-shi； Lin Zheng Zhi Nan Yi An； syndromes and prescriptions； artifial neural network； knowledge discovery

清代著名醫家叶天士曾叹“吾吴湿邪害人最广”，现代研究也表明，湿邪参与多种慢性病变发生发展过程，如微循环代谢紊乱、能量代谢障碍、局部组织炎症反应等[1]，因此对湿邪所致疾病的辨治研究值得重

通讯作者：陈洪宇，E-mail：hzchenhy@126.com

视。《临证指南医案》[2]出自临床实践，以医案形式记录诊疗经验和思路方法，具有很高的学术价值。通过分析其中的方证规律，研究处方内的药味及其配伍关系与其针对的病证、病机或病理环节之间的联系，有助于阐明医案中的辨证原理。……