A Character-level Convolutional Recurrent Interaction Network for joint traditional Chinese medicine clinical named...

Area of Science:

• Medical Informatics
• Natural Language Processing
• Traditional Chinese Medicine

Background:

• Traditional Chinese Medicine (TCM) electronic medical records (EMRs) contain valuable patient data structured by four diagnostic dimensions.
• Current deep learning models for TCM EMR analysis lack the ability to fully leverage these dimensions and ancient Chinese semantic features.
• This limits the effective structuring and modeling of TCM medical data.

Purpose of the Study:

• To develop a joint named entity and relation extraction method for TCM EMRs.
• To address the limitations of existing models in capturing the four diagnostic dimensions and ancient Chinese semantics.
• To improve the extraction of clinical entities and their relationships within TCM data.

Main Methods:

• Proposed a Character-level Convolutional Recurrent Interaction Network (CCRIN) model.
• Treated the four diagnostic dimensions as relationships, locations as head entities, and symptom attributes as tail entities.
• Integrated Chinese character embeddings, contextual convolutional features, and a self-attention mechanism for multi-task learning.

Main Results:

• The CCRIN model demonstrated superior performance in extracting TCM entities and relations.
• Empirical studies on the NYT and TCM-cases datasets validated the model's effectiveness.
• The model successfully captured semantic information from ancient Chinese and incorporated the four diagnostic dimensions.

Conclusions:

• The proposed CCRIN model offers an effective approach for analyzing TCM EMRs.
• This method enhances the extraction of clinical entities and their associative relationships based on the four diagnostic dimensions.
• The findings contribute to better structuring and modeling of TCM medical data using advanced NLP techniques.