中文电子病历数据元抽取方法

Purpose/Significance Extracting data elements that comply with national standards from EMR (Electronic Medical Records) can help to achieve fine-grained sharing of EMR data. Method/Process This paper proposes a method for extracting data elements from Chinese EMRs. Firstly, it uses the ALBERT, BILSTM and CRF models to perform sequence labeling on EMRs, and generates a set of candidate data elements based on labeling results. Then, for any candidate data element, its contextual information is collected to form an enhanced key vector. Finally, the similarity between the vector and the standard vector is calculated to determine whether the candidate data element is valid. Result/Conclusion The results show that the F1 value is 90.32%, indicating good performance. The shortcomings are the small size of the experimental dataset and the uneven distribution of data element types.