Research On Information Extraction Technology For Electronic Medical Record

With the development of the domestic medical information industry and the normalization of the medical data standard system,the value of researching clinical electronic medical record data has also increased.The digging of clinical electronic medical record data can promote the development of the smart medical industry.Information extraction technology is an important means of knowledge extraction for electronic medical record texts.In the field of clinical electronic medical records,the research of information extraction technology is of great significance to the conduction of personalized medical services,clinical decision support? follow-up management work and so on.Information extraction technology can obtain medical knowledge effectively from the electronic medical record text.In this article,information extraction technology mainly refers to named entity recognition technology and entity relationship extraction technology.The entity recognition technology aims to identify various types of medical entities in the electronic medical record text,and the entity relationship extraction technology aims to extract the relationship between various medical entities in the electronic medical record text.Compared with other types of texts,electronic medical records has some problems such as blurred boundaries,less labeled data,irregular writing and so on.The above problems increase the difficulty of named entity identification and entity relationship extraction.In order to extract the medical entities in the electronic medical record and the relationships between the entities effectively,this article does the following work:Entity recognition in electronic medical records: This paper proposes a method of electronic medical record entity recognition based on the pre-training model EMRBERT.First,use BERT pre-training model to provide basic parameters for EMR-BERT,next use relevant electronic medical records corpus to provide EMR-BERT with pretraining data.Then combine Bi-LSTM-CRF model for entity extraction.Finally compare the result with the traditional model,the recall rate of 64.97% and F value 62.14% are respectively increased 4.65% and 2.16%.The experimental results show that the model can effectively solve the problem of physical identification of electronic medical records.Entity relationship extraction in electronic medical records: For the relationship extraction task,select the benchmark datasets i2b2 2010 electronic medical record datasets of the task for experiments.This paper proposes a BiGRU-CNN model based on the attention mechanism to evaluate the micro-average indicators of the extraction results of the eight medical entity relationships.The result is as follows:the micro-average precision rate,micro-average recall rate and micro-average F1 value were 68.9%,64.6% and 66.7%.The experimental results show that the model can effectively solve the entity relationship extraction problem of electronic medical records.