Reproduce “SurgBox: Agent-Driven Operating Room Sandbox with Surgery Copilot” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jinlin Wu, Xusheng Liang, Xuexue Bai, Zhen Chen

SurgBox introduces an agent-driven operating room simulation framework featuring a "Surgery Copilot" to enhance surgical training and reduce cognitive workload during procedures. The system achieved 88.00% accuracy in surgical route selection and 88.02% in surgical plan accuracy, outperforming leading LLMs like GPT-4 in experimental validations.

Key insights

  • The integration of domain-specific Retrieval-Augmented Generation (RAG) dramatically improves LLM performance in highly specialized medical contexts, reducing hallucinations.
  • A novel long-short memory mechanism allows the AI Surgery Copilot to accumulate surgical experience and adaptively improve planning and real-time support over time.
  • Comprehensive simulation of distinct surgical roles and their interactions, covering the entire surgical closed loop, creates a robust platform for cognitive skill enhancement and operational assistance.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.