基于自然语言问题的电子病历分析工具—QReport  

EMR Statistical Analysis Tools using Questioning Answer Way—QReport

作　　者：孙程琳 夏宇航 刘旭利 高炬[2] 刘珉[2] 殷亦超[2] 阮彤[1] 

机构地区：[1]华东理工大学计算机科学与技术系,上海200237 [2]上海中医药大学附属曙光医院,上海200021

出　　处：《山西大学学报（自然科学版）》2018年第1期23-33,共11页Journal of Shanxi University(Natural Science Edition)

基　　金：国家高技术研究发展计划(863计划)(2015AA020107);国家自然科学基金(61402173)

摘　　要：为提高诊疗水平,临床医生在临床科研工作中,经常要对临床电子病历(Electronic Medical Record,EMR)数据进行各种统计分析。这项工作传统上由医疗信息人员协助,通过不断写SQL或是构建独立的联机事务分析系统完成。对于医生来说,使用EMR进行统计分析的门槛很高。为了降低医生使用EMR的门槛,文章提出了基于自然语言问题的电子病历分析工具—QReport,临床医生输入临床症-治-效相关的各种统计问题,QReport能够自动展现相应的图表。文中使用的症-治-效统计问题集使用专家调查法(Delphi method)向临床医生收集,对于统计自然语言问题,首先对其进行分词、抽取链接和文法解析得到问题解析树,之后生成问题语义图,并翻译为图查询语句在EMR知识库中检索。结果表明,该文法能够覆盖医学期刊中92.26%的统计类问题模式,且文法解析正确率为94.0%,用户主观满意度中位数为3.5(1-5)。In order to enhance medical standards,clinicians always use EMR （Electronic Medical Record） to do statistics with the assist of IT staff who repeatedly write SQL or build OLAP models, which is difficult for clinicians. With the purpose of reducing the barrier for clinician, this paper proposes an EMR analysis tool-- QReport. Clinicians input statistic questions about symptom, diagnosis and curative effect, QReport will return corresponding charts. We use Delphi method to collect statistic questions from clinicians. For the natural language statistic question, firstly, it is segmented and linked to entities in Clinical electronic medical records knowledge base, and we use grammar to analysis the linking result to get a parse tree. Af- ter that we can generate question semantic graph, and then translate it as structured query language to exe- cute in the EMR base. The results show that the grammar can cover 92.26% of the statistic questions in medical journals, the precision of grammar parsing is about 94.0%, and the median of user subjective sat- isfaction is 3.5（1-5）.

关 键 词：电子病历 统计分析问答 图数据库 上下文无关文法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]