Large Language Model (LLM)-based multi-agent systems have shown great potential in supporting complex tasks in the medical domain, such as improving diagnostic accuracy and facilitating multidisciplinary collaboration. However, despite the advancement, there is a lack of structured frameworks to guide the design of these systems in medical problem-solving. In this paper, we conduct a comprehensive survey of existing medical multi-agent systems, and propose a medical-specific taxonomy along three key dimensions: team composition, medical knowledge augmentation, and agent interaction. We further outline several future research directions, such as incorporating human-AI collaboration to ensure that human expertise and multi-agent reasoning jointly address complex clinical tasks, designing and evaluating agent profiles, and developing self-evolving systems that adapt to evolving medical knowledge and rapidly changing clinical environments. In summary, our work provides a structured overview of medical multi-agent systems and highlights key opportunities to advance their research and practical deployment.
Lin et al. (2025) studied this question.