BioBERT Based Named Entity Recognition in Electronic Medical Record

The identification and mining of entities in Electronic Medical Record (EMR) plays an important role in medical diagnosis. This paper used the BioBERT model based on Google BERT model for automatic annotation of clinical problems, treatments and tests in EMR. This method firstly pre-trained BioBERT model on the corpus of medical related fields to convert the text into a numerical vector. Next, the BiLSTM-CRF model was used to train the processed vectors and finally complete the entity tagging. The I2B2 2010 challenge dataset was used in the experiment. The experimental results show that the method can obviously improve performance of named entity recognition (NER) for EMR. The F1 score of the experiment is 87.10%, which meet the needs of clinical system applications and can promote the study of clinical decision in the future.