A Two-Stage Proactive Dialogue Generator for Efficient Clinical Information Collection Using Large Language Model.

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Decision Support Systems
• Natural Language Processing

Background:

• Effective patient-doctor interaction is crucial for accurate disease diagnosis.
• Collecting comprehensive patient history, including symptoms and past procedures, enhances diagnostic accuracy but is often time-consuming.
• Computer-assisted systems offer potential optimization for patient information gathering.

Purpose of the Study:

• To develop and evaluate a diagnostic dialogue system for automating patient information collection.
• To enhance the efficiency and effectiveness of disease diagnosis through intelligent conversation agents.
• To address challenges in dialogue generation such as under-exploration and inflexibility.

Main Methods:

• Proposed a diagnostic dialogue system utilizing medical history and conversation logic.
• Employed a two-stage recommendation structure with carefully designed ranking criteria.
• Developed an interactive patient agent to facilitate multi-round clinical queries.
• Leveraged a real-world medical conversation dataset for experimentation.

Main Results:

• The diagnostic dialogue system demonstrated superior performance compared to biomedical language model-based and Deepseek-based approaches.
• Generated clinical queries that effectively mimic the conversational style of human doctors.
• Achieved efficient fluency, professionalism, and safety in generated dialogue.
• Successfully collected relevant disease diagnostic information.

Conclusions:

• The proposed diagnostic dialogue system significantly improves the efficiency of patient information collection for disease diagnosis.
• The system's ability to generate human-like clinical queries enhances the diagnostic process.
• This approach offers a promising computer-assisted solution for optimizing patient-doctor interactions in healthcare settings.