Neuro-symbolic framework aligns LLMs with fuzzy logic for explainable diagnosis

A recent research article introduces a neuro-symbolic reasoning framework that integrates large language models (LLMs) with formal fuzzy logic, aiming to facilitate explainable and verifiable medical diagnoses. This framework incorporates patient narratives and clinical guidelines into a neural knowledge base, where LLMs identify structured medical entities, temporal relationships, and fuzzy symptom patterns. These elements are transformed into a symbolic knowledge base utilizing fuzzy logic and declarative rules. The reasoning process occurs in two stages: first, inductive symbolic generalization to identify diagnostic patterns from the narratives, followed by inference verification through a logic programming engine to derive and confirm diagnoses. This method tackles the challenges of verifiability and interpretability in LLMs for clinical decision-making, which often involves reasoning with incomplete and imprecise patient data. The paper can be found on arXiv with the identifier 2605.25566.