Multi-Expert LLM Framework for Clinical Diagnosis
A new multi-agent framework, RE-MCDF, addresses challenges in clinical diagnosis from electronic medical records (EMRs), particularly in neurology. EMRs are heterogeneous, sparse, and noisy, causing single-agent LLMs to produce self-reinforcing errors. RE-MCDF introduces closed-loop reasoning among multiple expert LLMs to mimic rigorous, evidence-driven clinical processes. Unlike existing multi-agent systems with shallow interactions, RE-MCDF models logical dependencies among diseases, such as mutual exclusivity and pathological compatibility, enabling it to rule out clinically implausible hypotheses. The framework aims to improve diagnostic accuracy by leveraging collaborative validation and structured reasoning.
Key facts
- RE-MCDF is a closed-loop multi-expert LLM reasoning framework for clinical diagnosis.
- It targets knowledge-grounded diagnosis from electronic medical records (EMRs).
- EMRs in neurology are heterogeneous, sparse, and noisy.
- Single-agent LLMs are vulnerable to self-reinforcing errors.
- Existing multi-agent frameworks have shallow and loosely structured interactions.
- RE-MCDF models logical dependencies among diseases, including mutual exclusivity and pathological compatibility.
- The framework can rule out clinically implausible hypotheses.
- It aims to reflect rigorous, evidence-driven processes used by clinical experts.
Entities
—