MMedAgent-RL: Optimizing Multi-Agent Collaboration for Multimodal Medical Reasoning

TL;DR

This paper introduces MMedAgent-RL, a reinforcement learning framework for dynamic multi-agent collaboration in multimodal medical reasoning, significantly improving diagnostic accuracy across diverse specialties.

Contribution

It presents a novel RL-based multi-agent system with curriculum learning for flexible, optimized medical reasoning, surpassing existing static collaboration models.

Findings

Achieves 23.6% average performance improvement over baselines.

Demonstrates effectiveness across five medical VQA benchmarks.

Enables dynamic, adaptive collaboration among medical agents.

Abstract

Medical Large Vision-Language Models (Med-LVLMs) have shown strong potential in multimodal diagnostic tasks. However, existing single-agent models struggle to generalize across diverse medical specialties, limiting their performance. Recent efforts introduce multi-agent collaboration frameworks inspired by clinical workflows, where general practitioners (GPs) and specialists interact in a fixed sequence. Despite improvements, these static pipelines lack flexibility and adaptability in reasoning. To address this, we propose MMedAgent-RL, a reinforcement learning (RL)-based multi-agent framework that enables dynamic, optimized collaboration among medical agents. Specifically, we train two GP agents based on Qwen2.5-VL via RL: the triage doctor learns to assign patients to appropriate specialties, while the attending physician integrates the judgments from multi-specialists and its own knowledge to make final decisions. To address the inconsistency in specialist outputs, we introduce a curriculum learning (CL)-guided RL strategy with dynamic entropy regulation, progressively teaching the attending physician to balance between imitating specialists and correcting their mistakes. Experiments on five medical VQA benchmarks demonstrate that MMedAgent-RL outperforms both open-source and proprietary Med-LVLMs. Notably, it achieves an average performance gain of 23.6% over strong baselines.