Medical Event Extraction For Chinese Electronic Medical Records

Medical event extraction is a fundamental issue in natural language processing technology in the medical field,aiming to extract specific event information from unstructured electronic medical records.It is a key step in constructing medical texts and a prerequisite for clinical applications.Although the performance of currently proposed medical event extraction methods is continuously improving,they still face some key issues: First,due to the lack of a Chinese medical event corpus,the cost of manual labeling of medical data is high,and labeling samples are scarce.Second,there are many difficult medical terms in electronic medical records,and existing medical event extraction models often ignore the importance of contextual semantic information for content understanding.Aiming at the above issues,this article takes the electronic medical record of tumors as the research object and uses deep learning,pseudo tag learning,and semantic features to conduct research on medical event extraction tasks.The main work is as follows:Firstly,a medical event extraction model based on pseudo-tag data enhancement is proposed TEC＿MEE.Aiming at the scarcity of labeled medical data samples,a pseudolabel selection algorithm is proposed,which makes full use of unlabeled data,uses trained models to predict unlabeled data,and generates pseudo-label data for data enhancement.Based on the enhancement of pseudo tag data,a medical event extraction model(TEC＿MEE)based on Transformer Encoder and CRF is constructed to extract the attributes of specified events from unstructured Chinese electronic medical record text.Experimental results show that compared to Bi GRU-CRF,the proposed TEC＿MEE model obtained better medical event extraction results after being enhanced with pseudo tag data,with an F1 value increase of 4.11%.Secondly,a medical event extraction model based on semantic features is proposed.To solve the problem of missing contextual semantic information in Chinese electronic medical records(EMR)text,a method based on the Ro BERTA model was proposed to construct a word vector containing richer contextual information,and then the semantic feature fusion BERT-Conditiona Layer Norm model was used to extract medical events.Using the relative distance of the trigger word in the electronic medical record and the relative distance of the word "transfer" in the electronic medical record as semantic features,a medical event extraction model integrating semantic features is constructed to extract various attributes of tumor events.Experimental results show that the constructed SF＿BERT model,the model extraction effect can be greatly improved,and compared to Transformer-Bi LSTM-CRF,the F1 value increased by 5.98%,respectively.This method has a better extraction effect.Thirdly,implement a tumor event extraction application based on fused semantic features.Realize the visualization of the application of medical event extraction technology in tumor diseases,and provide certain assistance for the diagnosis and treatment of tumor diseases.A medical event extraction model incorporating semantic features is used to extract various attributes of tumor events.Using Qt Designer to design a user interaction interface,through tumor grading,detection of metastatic sites,and determination of the size of the primary lesion,develop a tumor diagnosis and treatment plan,and conduct subsequent diagnosis and treatment plan evaluation and data analysis and statistics,can help clinicians diagnose and treat tumors more effectively.