MedCoRAG: Interpretable Hepatology Diagnosis via Hybrid Evidence Retrieval and Multispecialty Consensus

TL;DR

Experimental results on hepatic disease cases from MIMIC-IV show that MedCoRAG outperforms existing methods and closed-source models in both diagnostic performance and reasoning interpretability.

Abstract

Diagnosing hepatic diseases accurately and interpretably is critical, yet it remains challenging in real-world clinical settings. Existing AI approaches for clinical diagnosis often lack transparency, structured reasoning, and deployability. Recent efforts have leveraged large language models (LLMs), retrieval-augmented generation (RAG), and multi-agent collaboration. However, these approaches typically retrieve evidence from a single source and fail to support iterative, role-specialized deliberation grounded in structured clinical data. To address this, we propose MedCoRAG (i.e., Medical Collaborative RAG), an end-to-end framework that generates diagnostic hypotheses from standardized abnormal findings and constructs a patient-specific evidence package by jointly retrieving and pruning UMLS knowledge graph paths and clinical guidelines. It then performs Multi-Agent Collaborative Reasoning: a Router Agent dynamically dispatches Specialist Agents based on case complexity; these agents iteratively reason over the evidence and trigger targeted re-retrievals when needed, while a Generalist Agent synthesizes all deliberations into a traceable consensus diagnosis that emulates multidisciplinary consultation. Experimental results on hepatic disease cases from MIMIC-IV show that MedCoRAG outperforms existing methods and closed-source models in both diagnostic performance and reasoning interpretability.

Figures (5)

• Figure 1: Comparative Overview of Medical Diagnostic Reasoning Frameworks
• Figure 2: Overall architecture of the MedCoRAG framework, comprising three core components. (1) Abnormal Findings and Preliminary Diagnosis: Abnormal clinical findings are extracted from the patient narrative and standardized via UMLS to generate a focused set of initial diagnostic hypotheses. (2) Hybrid RAG: For each hypothesis, the system retrieves clinical guideline excerpts and UMLS knowledge graph paths, then prunes them using the full clinical context to form a coherent, patient-specific evidence package. (3) Multi-Agent Collaborative Reasoning: A Router Agent assesses case complexity to either activate relevant specialist agents or delegate simple cases to the Generalist Agent; all agents iteratively reason over the shared evidence, trigger re-retrieval when needed, and converge on an interpretable consensus diagnosis through the Generalist Agent.
• Figure 3: Confusion matrix of MedCoRAG on 13 hepatic disease classes.
• Figure 4: Average number of abnormal entities per case across different hepatic diseases. Higher values indicate more complex clinical presentations.
• Figure 5: Average number of hops in knowledge graph paths used during diagnosis. Higher values reflect greater reasoning complexity.