SP50. Novel Artificial Intelligence Retrieval-Augmented Generation Model for Hand Surgery

PURPOSE: Hand surgery encompasses a diverse range of procedures addressing complex anatomical structures and functional requirements. Keeping up with the extensive literature and evolving techniques can be challenging for surgeons. Artificial intelligence, particularly in natural language processing, offers potential to assist in medical decision-making. Retrieval-Augmented Generation (RAG) models provide context-aware, evidence-based responses by combining information retrieval with text generation. This study aimed to develop and evaluate a RAG model integrated with the GPT-4 API to assist surgeons in hand surgery planning and management. METHODS: We constructed a RAG model using 4,509 open-access full-text papers from PubMed related to hand surgery. The model was integrated with the GPT-4 API to create an interactive system optimized for clinical queries. The system was tested with various clinically relevant prompts related to hand surgery procedures. Responses were evaluated for accuracy, relevance, and adherence to current clinical guidelines. RESULTS: The integrated RAG-GPT-4 system successfully provided detailed, context-specific information on hand surgery techniques. For instance, when asked about the optimal management of a scaphoid non-union, the system offered comprehensive guidance on surgical options such as vascularized bone grafting, fixation methods, and postoperative rehabilitation. The responses were accurate, relevant, and aligned with current best practices. CONCLUSION: The integration of a RAG model with the GPT-4 API demonstrates significant potential in assisting hand surgeons with decision-making processes. By offering rapid access to relevant, evidence-based information, this system could enhance clinical decision-making and improve patient outcomes. Further clinical validation is needed to fully assess its impact on surgical practice.