Medical assertion classification in Chinese EMRs using attention enhanced neural network

Area of Science:

• Natural Language Processing
• Medical Informatics
• Machine Learning

Background:

• Electronic medical records (EMRs) contain valuable natural language data crucial for healthcare analysis.
• Accurate extraction of medical concepts and their semantic context, including assertion status, is vital for clinical applications.
• Existing medical assertion classification (MAC) methods often rely on traditional machine learning, requiring manual feature engineering and struggling with unlabeled data.

Purpose of the Study:

• To develop an advanced deep neural network for medical assertion classification in Chinese EMRs.
• To enhance information extraction by integrating external medical knowledge into a neural network architecture.
• To improve the accuracy and efficiency of classifying medical problem assertion types (e.g., diseases, symptoms).

Main Methods:

• A novel deep neural network architecture combining GRU and CNN models was proposed.
• A medical knowledge attention layer was developed to integrate entity representations from medical dictionaries.
• The attention layer uses dictionary-based entity representations as a query for encoding, enhancing context understanding.

Main Results:

• The proposed deep learning model demonstrated superior performance compared to existing methods in experimental evaluations.
• The integration of external medical knowledge via the attention layer significantly improved assertion classification accuracy.
• The model effectively classifies assertion types for medical problems like diseases and symptoms within Chinese EMRs.

Conclusions:

• The novel deep neural network architecture effectively classifies medical assertion types in EMRs.
• Integrating external medical knowledge through an attention mechanism enhances the performance of medical information extraction.
• This approach provides a foundation for more sophisticated health data analysis and clinical decision support systems.