基于肾病专科电子病历构建肾病医学知识图谱  

Constructing a Medical Knowledge Graph of Nephropathy Based on the Electronic Medical Records of Nephropathy Specialists

作　　者：林燕榕 张怡[2] 刘迪 钱东平 斯海燕 姜玉苹 朱江 陆凯东 陈浩 LIN Yan-rong;ZHANGYi;LIUDi;QIAN Dong-ping;SI Hai-yan;JIANG Yu-ping;ZHUJiang;LU Kai-dong;CHENHao(ShenTaiWang Healthcare Technology Limited Company,Nanjing 210023,China;National Key Laboratory for Novel Software Technology at Nanjing University,Nanjing 210023,China)

机构地区：[1]肾泰网健康科技(南京)有限公司,南京210023 [2]南京大学软件新技术国家重点实验室,南京210023

出　　处：《西南大学学报（自然科学版）》2020年第11期52-58,共7页Journal of Southwest University(Natural Science Edition)

基　　金：江苏省科技厅重点研发项目(BE20191611);江苏省南京市发展和改革委员会人工智能企业项目.

摘　　要：对肾病专科电子病历进行命名实体识别和实体关系抽取研究,为避免实体与关系独立抽取产生信息的冗余和联合抽取导致实体识别准确率下降的问题,将实体识别和关系抽取相结合,提高抽取准确率.先用长短期记忆网络—条件随机场(Bi-directional Long Short-Term Memory network,BiLSTM—Conditional Random Field,CRF)模型抽取实体,再使用卷积神经网络(Convolutional Neural Network,CNN)—BiLSTM模型从文本中抽取实体及实体关系,比对2个模型的实体位置,得到最终抽取结果,并将实体标准化后的结果传送到Neo4j数据库中构建知识图谱.该研究针对多数肾病病程长、预后差、治疗周期长的特点,构建一个完备的肾病医学知识图谱展示肾病大数据,为肾病专科临床决策、疾病监控等提供支持,提高医疗服务质量,辅助医学诊断.A study is carried out of named entity recognition(NER)and entity relationship extraction of the electronic medical records of nephrology specialty and,in order to solve the problem of redundancy and joint extraction of information resulting from the independent extraction of entity-relationship extraction,entity recognition and relationship extraction(RE)are combined to improve the extraction accuracy rate.The combined model of Bi-directional Long Short-Term Memory network(BiLSTM)and Conditional Random Field(CRF)is used to extract entities,and then the Convolutional Neural Network(CNN)-BiLSTM model is used to extract entities and entity relationships from the text.A comparison of the two model entity positions gives the final extraction results,and then the results of entity standardization are transferred to the Neo4j database to build a knowledge map.In conclusion,taking into consideration the fact that the majority of kidney diseases are characterized by a long course,a poor prognosis and a long treatment cycle,a complete medical knowledge map of kidney diseases is constructed in this study to show the big data of kidney diseases,which will be able to provide support for kidney disease specialists in their clinical decision-making and disease monitoring,and will help to improve the quality of medical services and auxiliary medical diagnosis.

关 键 词：实体识别 实体关系 Neo4j 知识图谱 联合抽取 长短期记忆网络 

分 类 号：TP392[自动化与计算机技术—计算机应用技术] R586[自动化与计算机技术—计算机科学与技术]