Translating Natural Language Questions into SNOMED Expression Constraint Language

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Clinical Terminology

Background:

• SNOMED Expression Constraint Language (ECL) is vital for clinical decision support and data analysis.
• The complexity and steep learning curve of ECL hinder its adoption by non-specialists.

Purpose of the Study:

• To develop a novel approach using Large Language Models (LLMs) to translate natural language questions into SNOMED ECL.
• To make SNOMED CT concepts more accessible by simplifying ECL query generation and explanation.

Main Methods:

• Leveraging state-of-the-art LLMs for bidirectional translation between natural language and ECL.
• Developing custom datasets and a novel pipeline integrating multiple AI agents.
• Training and evaluating the LLM-based model for accuracy in ECL query generation.

Main Results:

• The proposed LLM-based model achieved 83.78% accuracy in translating natural language questions to SNOMED ECL.
• Demonstrated the model's capability for bidirectional tasks: generating ECL and explaining ECL queries in natural language.

Conclusions:

• LLMs show significant potential in bridging the gap between complex clinical terminology (SNOMED ECL) and natural language.
• This approach can enhance healthcare interoperability, clinical decision support, and data analysis by improving accessibility to SNOMED CT.
• This research pioneers the use of AI for SNOMED ECL translation, paving the way for broader clinical adoption.