基于多尺度非完整数据的中医个体知识图谱构建及其诊疗规律研究

The long clinical practice of diagnosis and therapy was the essential way to form the stable system of “theory, principle, formula and herb” in traditional Chinese medicine (TCM). While, how could the pieces of personal experience accumulated into practice guideline, it is one of the key points to enhance the level of treatment variation based on syndrome differentiation. Early study has shown that the discrimination of syndrome types and herbs given to a same patient successively by three experienced doctor have a big difference, while they are all effective. Hypothesis was proposed that the personal treatment based on syndrome differentiation and the TCM doctor’s “theory, principle, formula and herb” could be taken into a self-consistency knowledge system, which could be expressed by the information visualization techniques of Knowledge Graphs. Structured design was chosen in the prospective data collection of 1500 outpatients treated continuously by three senior TCM doctors in two years. Realized the feature of the multi-scale nonholomic data in the treatment variation based on syndrome differentiation, method like Group structure thinning algorithm, random forest and other pattern recognition approach would be applied to construct the sub-net of the effective patient population, core herb formula and others. Information about the “diseases-syndrome patterns-treatment-effect” would be integrated to build up the personal knowledge graph. In the process, the management system of the update and clinical application could be formulated, and the personal experience in diagnosis and treatment would be explored and visualized. The study would provide more effective Statistical models and data analysis methods for the study and application of the individual regulation of the therapy based on syndrome differentiation by the construction technology of Knowledge Graph. Meanwhile, the study would lay the methodology foundation for researches about the inheritance and innovation of the distinguished veteran doctors of TCM.

漫长的中医临床诊疗实践，是形成“理法方药”体系稳定的必由之路，然而碎片化的个体诊疗经验如何上升为诊疗规律，是提高中医辨证论治水平的关键问题之一。前期研究表明：同一患者依次接受三位知名中医诊治，治疗有效前提下，辨证与方药存在较大差异。本研究提出“中医医师个体辨证论治及其理法方药是一种自洽的知识体系，可以运用知识图谱可视化建模技术表达出来”的工作假说，通过结构化设计、前瞻性采集3位高年资中医师门诊病例连续数据1500例；针对辨证论治诊疗数据的多尺度、非完整特点，拟采用结构组稀疏算法、随机森林等模式识别方法，建立有效患者群特征和核心处方等子网络；整合“病-证-治-效”信息构建医师个体知识图谱，形成其更新与临床应用机制，发现并可视化个体诊疗规律。本研究为知识图谱建模技术在辨证论治个体诊疗规律方面的研究与应用提供更为有效的统计模型及数据分析方法的同时，为名老中医经验传承与创新研究奠定方法学基础。

本项目以构建中医医师个体知识图谱模型，发现与展示中医专家个体诊疗知识体系，揭示其诊疗规律为目标，验证了辨证论治是一类具有复杂概念层次和网状结构的独特知识体系，是医师“症（证）、治、效”紧密相关的自洽体系，可以运用知识图谱可视化建模技术表达出来”的假说，以中医理论为指导，以失眠、慢性胃炎为例，应用临床实际诊疗数据1607例（8037诊次），构建了个体医师证-治-效紧密结合的知识框架和知识图谱，形成了名中医临床经验挖掘、可视化展示的基本规则和操作路径，并解决了个体医师辨证论治知识体系构建研究中的关键问题，为继续开展中医临床辨证论治思想传承与创新方法研究奠定了基础。（1）形成了体现辨证论治临床诊疗规律的知识本体框架构建方法和路径，提出中医师个体化辨证论治知识体系顶层本体，包含疾病、证候、治疗、疗效4级概念关系，是中医个体化知识图谱构建研究的有机组成；（2）形成了辨证论治真实诊疗过程数据采集与预处理方法，以失眠为示例形成了数据采集、数据库构建、数据预处理、质量控制方法等，是构建知识图谱的关键环节；（3）形成了个体医师知识图谱构建及可视化方法，以中医诊疗指南、标准和教材的知识分类，构建辨证论治知识体系的共性框架，作为知识图谱构建的共性本体；以医师个体实际诊疗数据为数据实体，在有效病例数据基础上，构建知识图谱，同时结合医师论著研究、专家访谈、复杂网络等多途径优化知识图谱，是个体医师辨证论治的图谱构建的基本规则；（4）对医师个体知识图谱规则的应用研究：将知识图谱构建方法在名医传承知识挖掘中进行应用，实践了可视化展示、语义搜索、治疗方案推荐等功能, 为分析和展示医师临证思维及其诊疗特点提供了新工具和系统流程。.总之，本研究将知识图谱构建技术与辨证论治疗效评价需求相结合，为挖掘、展示医师隐性思维及其诊疗方案提供了新途径，为中医师经验传承和创新奠定方法学基础，为今后开展相应机制研究奠定了临床基础。