Toward “Vibe Medicine”: A Self-Evolving Multi-Agent Framework for Clinical Decision Support

Large language models (LLMs) and AI agents have shown strong potential in medical imaging analysis, diagnosis, and treatment planning. However, most current medical AI systems still rely on pre-trained knowledge and fixed workflows. This limitation hinders their ability to learn from long-term clinical feedback, patient outcomes, or previous treatment experience, making it difficult for them to adapt to the complexity of real-world clinical practice.

To address this challenge, a team led by Dr. Lian Zhang from the First Hospital of Hebei Medical University, proposed the concept of “Vibe Medicine” and developed VIBEMed (Versatile Intelligent Behavior-Evolving Medical framework).

“VIBEMed uses multi-agent collaboration to break complex clinical decisions into three specialist roles: the Clinical Diagnostic Agent (CDA) for diagnostic reasoning and hypothesis generation, the Therapeutic Execution Agent (TEA) for treatment planning, and the Clinical Evolution Manager Agent (CEMA) for integrating longitudinal feedback and driving continuous optimization,” shares co-corresponding author Lian Zhang.

Unlike conventional single-model approaches, VIBEMed further implements a three-level self-evolution mechanism spanning memory, model, and code, improving the performance of the backbone LLM and system over time. It also employs an architecture-level safety sandbox to constrain model updates and data access, ensuring that continuous evolution remains safe, controllable, and traceable.

The team then validated VIBEMed in complex clinical scenarios. Compared with traditional single-model pipelines. “The framework demonstrated superior performance in complex medical reasoning and treatment planning tasks,” says Zhang. “VIBEMed presents an experience-driven medical AI system with strong adaptive capabilities, offering a practical direction for clinical decision-support systems that can learn continuously and evolve safely.”

Contact the author:

Lian Zhang, the First Hospital of Hebei Medical University, lianzhang@hebmu.edu.cn

Funder:

This work was supported by the National Foreign Experts Program (S20250235), Chronic Disease Management Research Program of the Center for Capacity Building and Continuing Education, National Health Commission (GWJJMB202510022202), Hebei Provincial “Yanzhao Golden Terrace Talent Recruitment Program” - Top-Tier Talent Program (HY2025050008), The Innovation & Development Medical Cooperation Program of Hengrui-Hebei (HR202502085), Hebei Provincial Health Commission Medical Research and Enterprise Joint Innovation Program (LH20250086), Hebei Provincial Health Commission Medical Research Program (20250011, 20260208, 20260195), Spark Scientific Research Program of The First Hospital of Hebei Medical University (XH202515).

Conflict of interest:

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

See the article:

Zhang, Qianxue, Ren, Yiming, et al. Toward Vibe Medicine: A Self-Evolving Multi-Agent Framework for Clinical Decision Support. Meta-Radiology, 100223. https://doi.org/10.1016/j.metrad.2026.100223