Research On Named Entity Recognition For Vascular Surgery Electronic Medical Records

In the process of medical informatization development,electronic medical records have always been the focus of construction.A large amount of unstructured data has been accumulated in the electronic medical record system of modern hospitals.These data have rich medical research value,but due to the lack of structure and standardization,the potential value of these data is difficult to be tapped.As an important surgical discipline,vascular surgery mainly involves the pathology,diagnosis,treatment,and prevention of human vascular diseases.These studies also require the support of a large amount of structured clinical data.As an important information extraction method,named entity recognition can identify various professional and complex entities from medical electronic medical records.Therefore,in order to further advance the field of vascular surgery,this paper conducts an in-depth study of its named entity recognition method using deep learning techniques.The main work content of this paper is as follows:(1)In view of the lack of research data in the field of vascular surgery,this paper constructs a small-scale specialist data set as the experimental data of this paper based on the course records and discharge records of real patients in the vascular surgery department of a tertiary hospital in Zhejiang Province.The BIOES notation system includes five categories of medical entities.This paper proposes a named entity recognition model MBAC based on Mac Bert’s pre-trained language model and attention mechanism.This model uses Mac Bert to fully consider context information to generate dynamic word vectors,uses Bi GRU to extract features,and captures the sequence interior through a multi-head self-attention mechanism.The relationship between elements,and finally decode the label through CRF.The experimental results show that the precision rate,recall rate and F1 value of the MBAC model on the data set in this paper are all better than those of classic models such as Bi LSTM-CRF and Bert-Bi LSTM-CRF,which verifies the effectiveness of the model.(2)Based on the MBAC model,this paper proposes a named entity recognition model MBDAC-FF-WV based on the integration of glyph features and models.This model addresses the problems of insufficient representation of single character vectors and insufficient recognition capabilities of a single model in the existing models.The presentation layer introduces the glyph information of the two dimensions of Chinese character Sijiaocode and Chinese character Wubi,performs feature fusion of Bi GRU and DGCNN in the feature extraction layer,and adds weighted voting for the output results of multiple models in the voting layer.The experimental results show that the recognition ability of the MBDAC-FF-WV model has been further improved compared with the MBAC model,which verifies the effectiveness of glyph feature embedding and model integration.(3)This paper designs and implements an electronic medical record recognition system for vascular surgery,and uses the research results of this paper for medical staff and scientific researchers.The system adopts browser/server architecture,including four modules of user management,entity recognition,data query,and data analysis,and provides functions such as authority management,text recognition,record addition and deletion,and statistical analysis.In this paper,a detailed functional test of the system is carried out to verify the usability of the system.In summary,this paper constructs a specialized data set in the field of vascular surgery,improves the performance of named entity recognition in this field by designing and improving the deep learning model,and applies the research results to engineering.