Named Entity Recognition of PCI Surgery Information Based on BERT+BiLSTM+CRF

Percutaneous coronary intervention (PCI) is a vital treatment method for coronary artery disease, but the unstructured nature of its clinical data makes it challenging to utilize directly. The data for this study was obtained from the Cardiovascular Treatment Center of the People’s Hospital of Liaoning Province, China. A representative dataset of 5.8% of PCI patients’ surgical records was selected for labeling, and a language model-based PCI surgical information entity recognition model was developed. First, Encoder Representations from Transformers (BERT) was employed to express the semantic relationship between characters accurately. Then, BiLSTM was used as a feature extractor to extract contextual relations, and finally, conditional random field (CRF) was applied to optimize the prediction results. Experimental results demonstrated that the F1 score in the PCI surgical information entity recognition model reached 85.49%, which is 25.66% higher than the traditional HMM and 0.94% higher than BiLSTM in deep learning.