Empowering Medical Multi-Agents with Clinical Consultation Flow for Dynamic Diagnosis

TL;DR

This work addresses dynamic medical diagnosis by formulating the problem as a $POMDP$ with three agents and introducing a RL‑driven multi‑agent framework anchored in clinical consultation flow. A hierarchical action set, derived from medical textbooks, constrains the doctor’s decisions across inquiry, examination, and diagnosis phases, while memory‑augmented reasoning guides action selection. Evaluations on the MVME benchmark show state‑of‑the‑art improvements in both score‑based and entity‑overlap (ICD‑10) metrics, with notable gains from RL in early phases that propagate to better diagnostics and reduced premature closure. The approach demonstrates stronger persistence in information gathering, more adaptive turn‑level interactions, and robustness to patient perturbations, signaling practical potential for real‑world dynamic, multimodal diagnosis. $POMDP$ formulation, hierarchical actions, and clinical knowledge integration collectively enable more reliable, transparent, and interactive AI‑assisted diagnosis in clinical workflows.

Abstract

Traditional AI-based healthcare systems often rely on single-modal data, limiting diagnostic accuracy due to incomplete information. However, recent advancements in foundation models show promising potential for enhancing diagnosis combining multi-modal information. While these models excel in static tasks, they struggle with dynamic diagnosis, failing to manage multi-turn interactions and often making premature diagnostic decisions due to insufficient persistence in information collection.To address this, we propose a multi-agent framework inspired by consultation flow and reinforcement learning (RL) to simulate the entire consultation process, integrating multiple clinical information for effective diagnosis. Our approach incorporates a hierarchical action set, structured from clinic consultation flow and medical textbook, to effectively guide the decision-making process. This strategy improves agent interactions, enabling them to adapt and optimize actions based on the dynamic state. We evaluated our framework on a public dynamic diagnosis benchmark. The proposed framework evidentially improves the baseline methods and achieves state-of-the-art performance compared to existing foundation model-based methods.

Figures (4)

• Figure 1: (1) Illustration of multi-agent dynamic diagnosis framework. (2) Main issues resulting in recent premature closure challenge in the interaction.
• Figure 2: (1) Illustration of the proposed hierarchical action set derived from consultation flow. (2) An example of the Dynamic diagnosis pipeline.
• Figure 3: Distribution of (a) dynamic diagnosis interaction turns and (b)-(f) evaluation scores in the 5 aspects.
• Figure 4: Case Study. An illustration of how the proposed framework mitigates the challenge of premature closure.