Research on the Structuring of Electronic Medical Records Based on Joint Extraction Using BART

Medical institutions commonly utilize electronic medical records (EMRs) to document patients’ medical conditions, which contain invaluable medical information. However, EMRs often consist of semi-structured or unstructured data, posing significant challenges in processing and analysis. In this paper, addressing the requirements for subsequent tasks such as clinical decision-making, we present the process of structuring EMRs, focusing on lung cancer EMRs. This process encompasses EMR structure analysis, data preprocessing, information extraction, and data integration. Notably, entity and entity relationship extraction are pivotal steps in this workflow. To accomplish this, we employ a joint extraction model using BART for information extraction tasks in lung cancer EMRs. When compared to existing models, our model achieves an F1 score of 64.86%. Furthermore, we validate the model’s generalization capability by conducting experiments on a pediatric epilepsy dataset, ultimately achieving the structuring of EMRs tailored to the requirements of subsequent tasks.