文献阅读规划(12篇)
这一组决定你对”多智能体””MDT””综述定位”的基本理解。读的时候不要追求记住每个细节,要追求”能用自己的话讲清楚一个完整故事”。
目录
| # | 文献(按参考编号) | 论文原名称 | 对应综述章节 | 要学到的核心内容 | 流程图 |
|---|---|---|---|---|---|
| 1 | 44. Konzmann et al. (2026) LLM in tumor boards: systematic review | Applications of large language models in tumor boards: a systematic review | 引言、4.3 | 现有 MDT+LLM 综述怎么定义范围、怎么组织证据、结论是什么;这是你的”锚点综述” | |
| 2 | 33. Guo et al. (2024) LLM-based multi-agents survey | Large language model based multi-agents: a survey of progress and challenges | 3.1、3.2 | LLM 智能体的定义(工具、RAG、记忆、规划、反思);多智能体的通用分类 | |
| 3✅ | 54. Li, X. et al. (2024) LLM-based MAS: workflow, infrastructure, challenges | A survey on LLM-based multi-agent systems: workflow, infrastructure, and challenges | 3.1、4.5 | 多智能体从工作流、基础设施、挑战三个角度的全景框架 | 认识 Agent 的零件:agent可以被分类成什么?(定义) |
| 4✅ | 96. Yao et al. (2026) AI hospitals survey | LLM-based multi-agent systems for clinical workflows: a survey of AI hospitals | 3.1、3.3、3.5 | 医疗多智能体如何组织临床工作流;AI 医院类模拟环境如何评测 | 提出 AI hospital概念:将AI医院定义为工作流级临床系统(流程落地) |
| 5✅⭐ | 92. Xiong et al. (2026) MAS for medicine: accountable orchestration | Not just one agent: LLM-based multi-agent systems for medicine from answer generation to accountable workflow orchestration | 3.1、4.1 | 医疗多智能体从”生成答案”到”工作流编排”的演变;可问责性 | 把医学 MAS 从“能跑医院流程”推进到“能被审计、被追责、被治理”。(落地审查) |
| 6✅ | 84. Tang et al. (2024) MedAgents | MedAgents: large language models as collaborators for zero-shot medical reasoning | 3.2.1、3.4 | 最早的医疗多智能体框架之一:角色化、协作讨论、零样本推理 | 医疗多智能体框架(早期agent落地) |
| 7✅(实现代码) | 43. Kim et al. (2024) MDAgents | MDAgents: an adaptive collaboration of LLMs for medical decision-making | 3.2.1、3.2.3、3.4 | 自适应协作:根据任务难度选择独诊/团队;这是”协作+层级”的交界 | 自适应协作解决了什么问题?它和固定团队的区别? |
| 8 | 13/15. Chen, K. et al. MDTeamGPT (2025/2026) | MDTeamGPT: a self-evolving LLM-based multi-agent framework for multi-disciplinary team medical consultation;MDTeamGPT: mitigating context collapse and enabling self-evolution in medical multi-agent reasoning | 3.2.4、3.4、3.6.2 | 共识聚合+残余讨论+知识库;”上下文崩塌”问题 | 上下文崩塌是什么?系统怎么缓解? |
| 9 | 58. Liu, Q. et al. (2026) EvoMDT | EvoMDT: a self-evolving multi-agent system for structured clinical decision-making in multi-cancer | 3.2.4、3.4、4.2 | 自进化 MDT 系统:根据专家反馈和结局信号更新提示词、权重、检索范围 | 自进化在 MDT 中如何实现?它和普通多智能体的差异? |
| 10 | 85. Vasilev et al. (2025) MTBBench | MTBBench: a multimodal sequential clinical decision-making benchmark in oncology | 3.5、4.4 | 肿瘤会诊专用基准:多模态、纵向、证据调和 | MTBBench 为什么比通用基准更适合 MDT?它测出了什么失败? |
| 11 | 104. Zhu, Y. et al. (2025) MedAgentAudit | Auditing medical multi-agent AI reveals risks of false consensus | 3.2.2、3.6.1、4.5 | 审计多智能体执行日志;虚假一致、沉默同意、未激活专科推理 | 审计发现了哪几类协作失败?为什么”辩论”不一定可靠? |
| 12 | 27. Gallifant et al. (2025) TRIPOD-LLM | The TRIPOD-LLM reporting guideline for studies using large language models | 3.6.4、4.5 | LLM 研究报告的最低报告要求 | TRIPOD-LLM 有几个条目?它如何支撑你综述的”评测不足”论点? |
本博客所有文章除特别声明外,均采用 CC BY-NC-SA 4.0 许可协议。转载请注明来源 ctyyy的博客!









/cover.jpg)
