An imConvNet-based deep learning model for Chinese medical named entity recognition.

Area of Science:

• Natural Language Processing
• Medical Informatics
• Artificial Intelligence

Background:

• Medical big data analysis relies on extracting information from electronic health records.
• Named Entity Recognition (NER) is crucial for text mining in the medical field.
• Challenges in medical NER include limited text resources and specialized terminology.

Purpose of the Study:

• To develop an improved model for Chinese medical Named Entity Recognition (NER).
• To enhance the accuracy of extracting medical information from electronic health records.
• To address the challenges of limited text resources and domain-specific terms in medical NER.

Main Methods:

• A novel model, BERT-imConvNet-BiLSTM-CRF, was developed by integrating an improved convolutional neural network (imConvNet) with BERT, BiLSTM, and CRF layers.
• The imConvNet model extracts additional word vector features to improve NER accuracy.
• The model architecture includes BERT embedding, imConvNet for context features, BiLSTM for long-distance dependencies, and CRF for labeling.

Main Results:

• The BERT-imConvNet-BiLSTM-CRF model achieved an average F1 score of 91.38% on the public yidu-s4k dataset.
• On real electronic medical record text, the model reached an F1 score of 93.89%.
• These results demonstrate superior performance compared to classical models.

Conclusions:

• The proposed imConvNet model significantly enhances the accuracy of Chinese medical named entity recognition.
• The model is effective across various medical corpora.
• This advancement aids in better information extraction from medical texts.