This research proposes a Neuro-Symbolic RAG for Clinical Decision Support (NSR-CDS) architecture that integrates Retrieval-Augmented Generation with neuro-symbolic reasoning to address limitations in current AI-powered clinical decision support systems, including lack of interpretability and inability to incorporate structured medical knowledge effectively.
Key findings
NSR-CDS combines neural retrieval mechanisms with symbolic knowledge graphs and logical inference.
The architecture features a dual-path retrieval system and a neuro-symbolic reasoning engine.
A multi-modal explanation generator provides both clinical rationales and logical justifications.
The approach ensures factual consistency through symbolic validation while maintaining neural generation flexibility.
Limitations & open questions
Challenges in handling automated chunking of clinical guidelines.
Ensuring temporal consistency of retrieved information remains an issue.