智能提取胃癌分期相关信息研究  

Intelligent Extracting Method for Gastric Cancer Staging Information

作　　者：黄文鹏 李莉明[1] 程铭 李爱云 梁盼[1] 雍刘亮 高剑波[1] HUANG Wenpeng;LI Liming;CHENG Ming;LI Aiyun;LIANG Pan;YONG Liuliang;GAO Jianbo(Department of Radiology,The First Affiliated Hospital of Zhengzhou University,Zhengzhou Henan 450052,China;Department of Information,The First Affiliated Hospital of Zhengzhou University,Zhengzhou Henan 450052,China)

机构地区：[1]郑州大学第一附属医院放射科,河南郑州450052 [2]郑州大学第一附属医院信息处,河南郑州450052

出　　处：《中国医疗设备》2021年第1期40-43,52,共5页China Medical Devices

基　　金：国家自然科学基金(81701687,81671682)。

摘　　要：目的采用自然语言处理技术从非结构化手术记录中智能提取胃癌分期的相关信息,并评估其效能。方法从电子病历系统中搜集2016至2018年确诊为胃癌并行手术的病例共632人,分析其手术记录,根据临床问题确定临床实体和属性。由两名医生进行标注,结果作为金标准。按3:1将数据集随机分为训练组和验证组。提取记录信息主要包括两步,首先采用识别医学实体,其次采用提取属性。采用精确度,召回率和F值评估模型效果。结果模型分析了共21319个实体,4390个属性。模型建立中,精确匹配医学实体识别平均精确度0.84,召回率0.87,F值0.85。属性平均精确度0.86,召回率0.88,F值0.87。松弛匹配的F值大于精确匹配下的F值。158份验证组手术记录中,19.62%出现浆膜受侵,37.34%出现肿大淋巴结,4.43%出现腹膜转移。结论本文提出了一种新的混合方法从手术记录中智能提取胃癌分期相关信息,未来将有可能在不同系统疾病的电子病历中应用。Objective To extract staging information of gastric cancer from unstructured surgical records automatically with natural language processing and evaluate the performance.Methods From 2016 to 2018,a total of 632 gastric cancer patients who underwent surgery were collected from the electronic medical record system,and their surgical records were analyzed to determined the entities and attributes according to clinical problems.Two experienced clinicians annotated entities and attributes which were served as gold standard.On a scale of 3:1,632 cases were randomly divided into training group and validation group.The extraction of recorded information mainly consists of two steps:firstly,the identification of medical entities;secondly,the extraction of attributes.Precision,recall rate and F-measure were used to evaluate the performance.Results A total of 21319 entities and 4390 attributes were analyzed.The average precision,recall and F-measure of clinical entities were 0.84,0.87 and 0.85 under strict matching criteria.The average precision,recall and F-measure of attributes were 0.86,0.88 and 0.87 under strict matching criteria.F-measures under relaxed matching criterion were all greater than that under strict matching.In validation group,19.62% patient have serosal invasion,37.34% patient have enlarged lymph nodes and 4.43% patient have peritoneal metastasis.Conclusion This study presents a new hybrid method to extract gastric cancer staging information and will be likely to be applied in electronic medical records of different systems in the future.

关 键 词：腹膜转移 电子病历 实体识别 信息提取 胃癌 

分 类 号：R735.2[医药卫生—肿瘤]