Integration of Retrieval-Augmented Generation Technique for LLM-based Differential Diagnosis Assistant

Artificial Intelligence (AI) is increasingly transforming the medical field, offering significant potential for diagnosis, treatment, and patient care. However, its successful integration relies on healthcare professionals, such as doctors, psychologists, and nurses, trusting the technology’s reliability and accuracy. For Large Language Models (LLMs), this trust requires transparent, verifiable, and rigorously reviewed information sources. This paper presents an AI-powered tool for differential diagnosis and disease comparison, utilizing an LLM enhanced by Retrieval-Augmented Generation (RAG). RAG overcomes traditional LLM limitations by enabling access to external, domain-specific knowledge, ensuring accurate and contextually relevant responses. The system leverages PubMed, a biomedical article aggregator, to extract symptom-related information from scientific literature on various disorders. Evaluations involving psychologist-administered questionnaires demonstrate that combining a similarity score with detailed symptom descriptions provides a clear understanding of relationships between disorders. This approach may enhance diagnostic precision and build trust in AI-driven tools, encouraging their broader adoption in clinical practice.