Development and evaluation of a retrieval-augmented large language model framework for enhancing endodontic education

Area of Science:

• Artificial Intelligence in Medical Education
• Endodontics Training
• Multimodal Knowledge Integration

Background:

• Integrating domain-specific knowledge into Large Language Models (LLMs) is challenging for medical education.
• Generic LLMs lack clinical accuracy and contextual grounding for specialized fields like endodontics.
• Effective endodontic training requires both textual evidence and visual procedural demonstrations.

Purpose of the Study:

• To develop and evaluate a Retrieval-Augmented Generation (RAG)-enhanced LLM framework for AI-driven endodontic education.
• To address limitations of generic LLMs in providing clinically accurate and contextually relevant information.
• To improve AI's applicability in specialized dental training environments.

Main Methods:

• Developed Endodontics-KB, a multimodal knowledge platform using a hierarchical RAG architecture.
• Integrated evidence-based literature and visual instructional materials.
• The EndoQ chatbot, augmented with multimodal dental datasets, was benchmarked against GPT-4, Qwen2.5, and DeepSeek R1 using expert-validated questions.

Main Results:

• The framework integrated 2,200 multimodal knowledge units via dynamic semantic indexing.
• EndoQ showed statistically significant improvements over general LLMs in accuracy, clinical relevance, completeness, professionalism, and fluency (p < 0.05).
• Domain experts rated EndoQ highly on a 5-point Likert scale across all evaluated metrics.

Conclusions:

• The RAG framework enhances educational outcomes through precise, context-aware knowledge delivery in endodontics.
• This model is scalable and transferable for AI-enhanced clinical training across medical specialties.
• It significantly advances competency-based pedagogy in dental education.