基于深度学习的医疗数据智能分析算法研究  

Research on intelligent analysis algorithm of medical data based on deep learning

作　　者：朱铁兵[1] 柏志安[1] ZHU Tiebing;BAI Zhian(Computer Center,Ruijin Hospital Affiliated to Medical School of Shanghai Jiaotong University,Shanghai 200025,China)

机构地区：[1]上海交通大学医学院附属瑞金医院计算机中心,上海200025

出　　处：《电子设计工程》2021年第12期84-88,共5页Electronic Design Engineering

基　　金：上海市科委重点攻关项目(10231204103)。

摘　　要：当前电子病历(EMRs)命名实体识别(NER)的研究主要集中在验证传统文本中的命名实体识别方法是否对电子病历有效。到目前为止,仍未从多类分类的角度提出通过深度学习提高命名实体识别性能的模型。文中通过对一个真实的EMRs语料库进行标注并完成模型的训练与性能评估,然后提出一种基于深度学习卷积神经网络(CNN)的EMRs命名实体多类分类方法。该方法分为两个阶段:在第一阶段中对EMRs进行预处理,以嵌入单词来表示样本;在第二阶段,将训练数据分割成多个子集,并在每个子集上训练出CNN二元分类模型。通过数据测试实验,将此方法与其他传统方法进行详细的对比,验证了该多类分类方法的有效性。The current research on Named Entity Recognition(NER)of Electronic Medical Records(EMRs)mainly focuses on verifying whether the named entity recognition methods in traditional texts are effective for EMRs.So far,there is no model to improve the recognition performance of named entity through deep learning from the perspective of multi class classification.In this paper,a real EMRs corpus is labeled,and the model training and performance evaluation are completed.A multi class classification method of EMRs named entities based on deep learning Convolutional Neural Network(CNN)is proposed.The method is divided into two stages:in the first stage,EMRs is preprocessed to embed words to represent samples;in the second stage,training data is divided into multiple subsets,and CNN binary classification model is trained on each subset.Through data testing experiments,this method is compared with other traditional methods in detail,and the effectiveness of the multi class classification method is verified.

关 键 词：EMRs 卷积神经网路 命名实体识别 深度学习 语言处理 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]