基于层叠条件随机场模型的电子病历文本信息抽取  

TEXT INFORMATION EXTRACTION FOR ELECTRONIC MEDICAL RECORD BASED ON CASCADED CONDITIONAL RANDOM FIELD MODEL

作　　者：梁立荣 李长伟 沈晔 周立娟 景行 童朝晖 Liang Lirong;Li Changwei;Shen Ye;Zhou Lijuan;Jing Hang;Tong Zhaohui(Department of Clinical Epidemiology Research,Beijing Institute of Respiratory Medicine,Beijing Chaoyang Hospital,Capital Medical University,Beijing 100020,China;Department of Epidemiology and Biostatistics,College of Public Health,University of Georgia,Athens 30602,Georgia,USA;Beijing INFI-sagacity Technology Co.,Ltd.,Beijing 100085,China;Department of Respiratory and Critical Care Medicine,Beijing Institute of Respiratory Medicine,Beijing Chaoyang Hospital,Capital Medical University,Beijing 100020,China)

机构地区：[1]首都医科大学附属北京朝阳医院北京市呼吸疾病研究所临床流行病学研究室,北京100020 [2]乔治亚大学公共卫生学院流行病学与生物统计学系,佐治亚雅典30602 [3]北京无极慧通科技有限公司,北京100085 [4]首都医科大学附属北京朝阳医院北京市呼吸疾病研究所呼吸与危重症医学科,北京100020

出　　处：《计算机应用与软件》2019年第10期47-54,112,共9页Computer Applications and Software

基　　金：北京市第五批高层次卫生技术人才(学科骨干)项目(2015-3-026)

摘　　要：自然语言处理技术已用于非结构化中文电子病历信息抽取,并且新的算法或模型不断出现,但其应用效果的证据较少。共收集北京某大型三甲综合医院呼吸专科住院电子病历38218份,通过对数据预处理,抽象文本特征与定义语法规则,产生训练数据集和测试集,构建层叠条件随机场模型,并评估该模型的识别效果。结果表明,针对入院记录、出院记录、辅助检查报告3大类共39种非结构化文本,该模型可准确、快速地处理病历文本信息,应用效果较为理想。Natural language processing technology has been used to extract unstructured Chinese electronic medical record information,and new algorithms or models are emerging,but there is little evidence of its application effect.A total of 38 218 electronic medical records were collected from the respiratory department of a large third-class general hospital in Beijing.Through data preprocessing,abstracting text features and defining grammatical rules,training data sets and test sets were generated,cascaded conditional random field model was constructed,and the recognition effect of the model was evaluated.For 39 unstructured texts in three categories: admission record,discharge record and auxiliary examination report,the results show that the model can process the text information of medical records accurately and quickly,and the application effect is ideal.

关 键 词：医疗电子病历 自然语言处理 机器学习 层叠条件随机场模型 信息抽取 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]