CRF与规则相结合的医学病历实体识别  

Combining CRF and rule based medical named entity recognition

作　　者：栗伟[1] 赵大哲[1,2] 李博[1] 彭新茗 刘积仁[2,3] 

机构地区：[1]东北大学医学影像计算教育部重点实验室,沈阳110004 [2]东北大学信息科学与工程学院,沈阳110004 [3]东软集团股份有限公司,沈阳110179

出　　处：《计算机应用研究》2015年第4期1082-1086,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61172002;61302012);中央高校基本科研业务费专项资金资助项目(N120518001;N110718001);辽宁省自然科学基金资助项目(2013020021)

摘　　要：针对电子病历结构化中命名实体识别困难的问题,提出了一种基于CRF与规则相结合的医学病历实体识别算法。该算法采用CRF进行病历实体的初始识别,然后基于规则进行病历实体识别结果优化,其中规则包括基于决策树生成的规则和临床知识规则。实验证明,该算法对病历实体进行识别时准确率及召回率分别最高达到91.03%和87.26%,满足临床中系统应用需求,同时实验表明该算法具有很好的鲁棒性和稳定性。In the preprocessing step of electronic medical records analysis,medical named entity recognition is a key issue.This paper proposed a combining CRF and rule based medical named entity recognition algorithm. The algorithm made an initial entity recognition by CRF and then applied a rule based recognition method to improve the accuracy,whose rules included the rules from decision tree and domain knowledge. The results show that the algorithm has high accuracy and recall performance at records entity recognition that is up to 91. 03% and 87. 26%,and meets the requirement of the clinical application.Meanwhile,the algorithm has good robustness and stability on different sizes,types of the dataset.

关 键 词：电子病历 病历实体 命名实体识别 条件随机场 决策树 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]