[1]王静,申乐,林飞,等.MorpheusAPI：基于大语言模型Agent的智能麻醉平台[J].智能系统学报,2026,21(1):156-166.[doi:10.11992/tis.202505004]
WANG Jing,SHEN Le,LIN Fei,et al.MorpheusAPI: an LLM Agent for intelligent anesthesia platform[J].CAAI Transactions on Intelligent Systems,2026,21(1):156-166.[doi:10.11992/tis.202505004]
点击复制
WANG Jing,SHEN Le,LIN Fei,et al.MorpheusAPI: an LLM Agent for intelligent anesthesia platform[J].CAAI Transactions on Intelligent Systems,2026,21(1):156-166.[doi:10.11992/tis.202505004]
MorpheusAPI：基于大语言模型Agent的智能麻醉平台
《智能系统学报》[ISSN 1673-4785/CN 23-1538/TP] 卷:
21
期数:
2026年第1期
页码:
156-166
栏目:
学术论文—智能系统
出版日期:
2026-03-05
- Title:
- MorpheusAPI: an LLM Agent for intelligent anesthesia platform
- Keywords:
- anesthetics; multi-agent systems; large language models; artificial intelligence; intelligent agents; decision support systems; medical computing; risk assessment
- 分类号:
- TP391
- 摘要:
- 针对传统围术期麻醉管理模式主要依赖临床指南和麻醉医生临床判断，使得麻醉医生在面对海量的实时生理数据、复杂的患者个体化情况以及瞬息万变的高风险场景时需要承担巨大的工作负担和决策压力的问题，本文提出了 MorpheusAPI框架，一种基于大语言模型(large language models, LLMs)的多Agents智能麻醉平台。该平台包含执行大模型和影子大模型，执行大模型整合了感知、预测、决策、验证和中央协调5个Agents，通过模型上下文协议(model context protocol, MCP)实现多模态数据高效整合，思维链(chain-of-thought, CoT)提示增强风险推理，检索增强生成(retrieval-augmented generation, RAG)确保临床决策可靠性，麻醉影子大模型通过持续优化执行模型性能形成闭环系统。案例研究显示MorpheusAPI系统风险预测响应时间0.4 s，核心推理延迟10~15 ms，成功将丙泊酚诱导剂量优化至2.0 mg/(kg·h)，维持平均动脉压不小于65 mmHg，结果验证了其在提升麻醉安全性和效率的巨大潜力，可为智能麻醉系统设计与应用提供新思路。
- Abstract:
- To address the reliance of traditional perioperative anesthesia management models on clinical guidelines and the clinical judgment of anesthesiologists, which causes anesthesiologists to bear a huge workload and decision-making pressure when faced with massive real-time physiological data, complex individualized patient conditions, and rapidly changing high-risk scenarios, MorpheusAPI, a multi-agent, intelligent anesthesia platform based on large language models(LLMs), was proposed. The platform includes an execution model and a shadow model. The execution model integrates five agents for perception, prediction, decision-making, verification and central coordination. It enables efficient integration of multi-modal data through the model context protocol, the chain-of-thought prompts to enhance risk reasoning, and retrieval-augmented generation(RAG) to ensure the reliability of clinical decisions. The anesthesia shadow model forms a closed loop that continuously optimizes the performance of the execution model. Case studies demonstrated that the risk prediction response time for the MorpheusAPI system is 0.4 s, the core reasoning delay is 10~15 ms, the propofol induction dose is successfully optimized to 2.0 mg/(kg·h), and the mean arterial pressure is maintained not less than 65 mmHg. The results verify the great potential of the model to improve anesthesia safety and efficiency and provide insights into the design and application of intelligent anesthesia systems.
- 参考文献/References:
-
[1] ZENG Shiyue, QING Qi, XU Wei, et al. Personalized anesthesia and precision medicine: a comprehensive review of genetic factors, artificial intelligence, and patient-specific factors[J]. Frontiers in medicine, 2024, 11: 1365524.
[2] CANNESSON M, ANI F, MYTHEN M M, et al. Anaesthesiology and perioperative medicine around the world: different names, same goals[J]. British journal of anaesthesia, 2015, 114(1): 8-9.
[3] SESSLER D I, MEYHOFF C S, ZIMMERMAN N M, et al. Period-dependent associations between hypotension during and for four days after noncardiac surgery and a composite of myocardial infarction and death: a substudy of the POISE-2 trial[J]. Anesthesiology, 2018, 128(2): 317-327.
[4] FUTIER E, LEFRANT J Y, GUINOT P G, et al. Effect of individualized vs standard blood pressure management strategies on postoperative organ dysfunction among high-risk patients undergoing major surgery: a randomized clinical trial[J]. Journal of the American medical association, 2017, 318(14): 1346-1357
[5] NAIR B G, GABEL E, HOFER I, et al. Intraoperative clinical decision support for anesthesia: a narrative review of available systems[J]. Anesthesia & analgesia, 2017, 124(2): 603-617.
[6] SHICKEL B, TIGHE P J, BIHORAC A, et al. Deep EHR: a survey of recent advances in deep learning techniques for electronic health record(EHR) analysis[J]. IEEE journal of biomedical and health informatics, 2018, 22(5): 1589-1604
[7] SHORTLIFFE E H, SEP?LVEDA M J. Clinical decision support in the era of artificial intelligence[J]. Journal of the American medical association, 2018, 320(21): 2199
[8] THIRUNAVUKARASU A J, TING D S J, ELANGOVAN K, et al. Large language models in medicine[J]. Nature medicine, 2023, 29(8): 1930-1940.
[9] SAFDARI R, MALAK J S, MOHAMMADZADEH N, et al. A multi agent based approach for prehospital emergency management[J]. Bulletin of emergency and trauma, 2017, 5(3): 171-178.
[10] 赵毅飞, 申乐, 叶佩军, 等. 平行麻醉: 从麻醉自动化走向智慧型全周期麻醉平台[J]. 智能科学与技术学报, 2024, 5(2): 234-246. ZHAO Yifei, SHEN Le, YE Peijun, et al. Parallel anesthesia: from anesthesia automation to intelligent full-cycle anesthesia platform[J]. Chinese journal of intelligent science and technology, 2023, 5(2): 234-246.
[11] 王惠珍, 张捷, 俞怡, 等. 平行手术室: 围术期护理流程与智慧手术平台管理的新模式[J]. 模式识别与人工智能, 2023, 36(10): 867-876. WANG Huizhen, ZHANG Jie, YU Yi, et al. Parallel operating rooms: a new model of perioperative nursing process and smart surgical platform management[J]. Pattern recognition and artificial intelligence, 2023, 36(10): 867-876.
[12] LIN Fei, GAO T, SUN Dali, et al. Parallel medical devices and instruments: integrating edge and cloud intelligence for smart treatment and health systems[J]. IEEE/CAA journal of automatica sinica, 2025, 12(4): 651-654.
[13] PATNAIK S. S, HOFFMANN U. Comparison of ChatGPT vs. Bard to anesthesia-related queries[EB/OL]. (2023-06-30)[2025-04-20]. https://www.medrxiv.org/content/10.1101/2023.06.29.23292057v1.
[14] WANG Zhonghai, JIANG Jie, ZHAN Yibing, et al. Hypnos: a domain-specific large language model for anesthesiology[J]. Neurocomputing, 2025, 624: 129389.
[15] ZHOU Bohao, ZHAN Yibing, WANG Zhonghai, et al. Benchmarking medical LLMs on anesthesiology: a comprehensive dataset in Chinese[J]. IEEE transactions on emerging topics in computational intelligence, 2025, 9(4): 3057-3071.
[16] GUTHRIE E, LEVY D, DEL CARMEN G. The operating and anesthetic reference assistant(OARA): a fine-tuned large language model for resident teaching[J]. The American journal of surgery, 2024, 234: 28-34
[17] Anthropic. 2024. Introducing the Model Context Protocol[EB/OL]. (2024-11-25)[2025-04-06]. https://www.anthropic.com/news/model-context-protocol.
[18] WEI J, WANG Xuezhi, SCHUURMANS D, et al. Chain-of-thought prompting elicits reasoning in large language models[EB/OL]. (2023-01-10)[2025-04-06]. https://arxiv.org/abs/2201.11903.
[19] LEWIS P, ETHAN P, ALEKSANDRA P, et al. Retrieval-augmented generation for knowledge-intensive NLP tasks[EB/OL]. (2020-05-22)[2025-03-20]. https://arxiv.org/abs/2005.11401.
[20] YANG Jing, TIAN Yonglin, LIN Fei, et al. From RAG to ARG: agent reinforced generation for agentic intelligence[J]. The international journal of intelligent control and systems, 2025, 30(1): 76-82.
[21] ZHANG Jiayi, XIANG Jinyu, YU Zhaoyang, et al. AFlow: automating agentic workflow generation[EB/OL]. (2025-04-15)[2025-06-15]. https://arxiv.org/abs/2410.10762.
[22] CONNOR C W. Artificial intelligence and machine learning in anesthesiology[J]. Anesthesiology, 2019, 131(6): 1346-1359.
[23] JOHN DOYLE D, DAHABA A A, LEMANACH Y. Advances in anesthesia technology are improving patient care, but many challenges remain[J]. BMC anesthesiology, 2018, 18(1): 39.
[24] HASHIMOTO D A, WITKOWSKI E, GAO Lei, et al. Artificial intelligence in anesthesiology[J]. Anesthesiology, 2020, 132(2): 379-394.
[25] SCHMIERER T, LI Tianning, LI Yan. Harnessing machine learning for EEG signal analysis: innovations in depth of anaesthesia assessment[J]. Artificial intelligence in medicine, 2024, 151: 102869
[26] HASHEMI S, YOUSEFZADEH Z, ALI ABIN A, et al. Machine learning-guided anesthesiology: a review of recent advances and clinical applications[J]. Journal of cellular & molecular anesthesia, 2024, 9(1): 1-29.
[27] SCHAMBERG G, BADGELEY M, MESCHEDE-KRASA B, et al. Continuous action deep reinforcement learning for propofol dosing during general anesthesia[J]. Artificial intelligence in medicine, 2022, 123: 102227.
[28] LEE H, YOON H K, KIM J, et al. Development and validation of a reinforcement learning model for ventilation control during emergence from general anesthesia[J]. NPJ digital medicine, 2023, 6(1): 145.
[29] OpenAI, ACHIAM J, ADLER S, et al. GPT-4 Technical Report[EB/OL]. (2024-03-04)[2025-04-20]. https://arxiv.org/abs/2303.08774
[30] SINGHAL K, AZIZI S, TU Tao, et al. Large language models encode clinical knowledge[J]. Nature, 2023, 620(7972): 172-180.
[31] LUO Renqian, SUN Liai, XIA Yingce, et al. BioGPT: generative pre-trained transformer for biomedical text generation and mining[J]. Briefings in bioinformatics, 2022, 23(6): bbac409.
[32] 张志成, 王静, 张阳, 等. OrthoGPT: 面向精准诊疗的多模态骨科大模型[J]. 智能科学与技术学报, 2024, 6(3): 338-346. ZHANG Zhicheng, WANG Jing, ZHANG Yang, et al. OrthoGPT: multimodal generative pre-trained transformer models for precise diagnosis and treatment of orthopedics[J]. Chinese journal of intelligent science and technology, 2024, 6(3): 338-346.
[33] 肖建力, 许东舟, 王浩, 等. 医疗领域的大型语言模型综述[J]. 智能系统学报, 2025, 20(3): 530-547. XIAO Jianli, XU Dongzhou, WANG Hao, et al. Survey of large language models in healthcare[J]. CAAI transactions on intelligent systems, 2025, 20(3): 530-547.
[34] 刘学博, 户保田, 陈科海, 等. 大模型关键技术与未来发展方向——从ChatGPT谈起[J]. 中国科学基金, 2023, 37(5): 758-766 LIU Xuebo, HU Baotian, CHEN Kehai, et al. Key technologies and future development directions of large models — starting from ChatGPT[J]. Science foundation in China, 2023, 37(5): 758-766
[35] 马武仁, 弓孟春, 戴辉, 等. 以ChatGPT为代表的大语言模型在临床医学中的应用综述[J]. 医学信息学杂志, 2023, 44(7): 9-17. MA Wuren, GONG Mengchun, DAI Hui, et al. A comprehensive review of the applications of large language models in clinical medicine with ChatGPT as a representative[J]. Journal of medical informatics, 2023, 44(7): 9-17.
[36] ANGEL M C, RINEHART J B, CANNESSON M P, et al. Clinical knowledge and reasoning abilities of AI large language models in anesthesiology: a comparative study on the American board of anesthesiology examination[J]. Anesthesia & analgesia, 2024, 139(2): 349-356.
[37] KHAN A A, YUNUS R, SOHAIL M, et al. Artificial intelligence for anesthesiology board-style examination questions: role of large language models[J]. Journal of cardiothoracic and vascular anesthesia, 2024, 38(5): 1251-1259.
[38] RATNAGANDHI J A, GODAVARTHY P, GNANESWARAN M, et al. Enhancing anesthetic patient education through the utilization of large language models for improved communication and understanding[J]. Anesthesia research, 2025, 2(1): 4.
[39] KUO F H, FIERSTEIN J L, TUDOR B H, et al. Comparing ChatGPT and a single anesthesiologist’s responses to common patient questions: an exploratory cross-sectional survey of a panel of anesthesiologists[J]. Journal of medical systems, 2024, 48(1): 77.
[40] FENG Xiang, JIANG Wentao, WANG Zhenmao, et al. AnesBench: multi-dimensional evaluation of LLM reasoning in anesthesiology[EB/OL]. (2025-04-03)[2025-04-20]. https://arxiv.org/abs/2504.02404.
[41] WANG Zhonghai, ZHOU Bohao, LIU Weifeng. ATSCOT:chain of thought for structuring anesthesia medical records[C]//2024 International Conference on New Trends in Computational Intelligence. Qingdao: IEEE, 2024: 145-149.
[42] WANG Zhonghai, JIANG Jie, ZHAN Yibin, et al. Towards training a Chinese large language model for anesthesiology[EB/OL]. (2024-03-05)[2025-02-16]. https://arxiv.org/abs/2403.02742
[43] QIU Jianing, LAM K, LI Guohao, et al. LLM-based agentic systems in medicine and healthcare[J]. Nature machine intelligence, 2024, 6(12): 1418-1420.
[44] KARUNANAYAKE N. Next-generation agentic AI for transforming healthcare[J]. Informatics and health, 2025, 2(2): 73-83.
[45] SUDARSHAN M, SHIH S, YEE E, et al. Agentic LLM workflows for generating patient-friendly medical reports[EB/OL]. (2024-08-05)[2025-04-20]. https://arxiv.org/abs/2408.01112.
[46] YU Huizhi, ZHOU Jiayan, LI Lingyao, et al. AIPatient: simulating patients with EHRs and LLM powered agentic workflow[EB/OL]. (2024-09-27)[2025-02-15]. https://arxiv.org/abs/2409.18924.
[47] ZOU J, TOPOL E J. The rise of agentic AI teammates in medicine[J]. The lancet, 2025, 405(10477): 457.
[48] YUAN Han. Agentic large language models for healthcare: current progress and future opportunities[J]. Medicine advances, 2025, 3(1): 37-41.
[49] DOLANT A, KUMAR P. Agentic LLM framework for adaptive decision discourse[EB/OL]. (2025-02-16)[2025-04-20]. https://arxiv.org/abs/2502.10978.
[50] PAGLIERI D, CUPIAL B, COWARD S, et al. BALROG: benchmarking agentic LLM and VLM reasoning on games[EB/OL]. (2025-04-14)[2025-05-05]. https://arxiv.org/abs/2411.13543v2.
[51] TIAN Yonglin, LIN Fei, WANG Cong, et al. Parallel deep foundation model: a co-evolution framework for analogical imagination and embodied cognition of parallel intelligence[J]. The international journal of intelligent control and systems, 2025, 30(1): 83-90.
[52] 陈小平. 大模型关联度预测的形式化和语义解释研究[J]. 智能系统学报, 2023, 18(4): 894-900 CHEN Xiaoping. Research on formalization and semantic interpretations of correlation degree prediction in large language models[J]. CAAI transactions on intelligent systems, 2023, 18(4): 894-900
[53] 陈小平. 人工智能中的封闭性和强封闭性: 现有成果的能力边界、应用条件和伦理风险[J]. 智能系统学报, 2020, 15(1): 114-120. CHEN Xiaoping. Criteria of closeness and strong closeness in artificial intelligence—limits, application conditions and ethical risks of existing technologies[J]. CAAI transactions on intelligent systems, 2020, 15(1): 114-120.
[54] 古天龙, 马露, 李龙, 等. 符合伦理的人工智能应用的价值敏感设计: 现状与展望[J]. 智能系统学报, 2022, 17(1): 2-15. GU Tianlong, MA Lu, LI Long, et al. Value sensitive design of ethical-aligned AI applications: current situation and prospect[J]. CAAI transactions on intelligent systems, 2022, 17(1): 2-15.
[55] 黄河燕, 李思霖, 兰天伟, 等. 大语言模型安全性: 分类、评估、归因、缓解、展望[J]. 智能系统学报, 2025, 20(1): 2-32. HUANG Heyan, LI Silin, LAN Tianwei al. A survey on the safety of large language model: classification, evaluation, attribution, mitigation and prospect[J]. CAAI transactions on intelligent systems, 2025, 20(1): 2-32.
备注/Memo
收稿日期:2025-5-8。
基金项目:澳门特别行政区科学与技术发展基金项目(0093/2023/RIA2, 0145/2023/RIA3, 0157/2024/RIA2)；北京市自然科学基金海淀联合基金项目(L222099)；首都卫生发展科研专项(2024-2-4015)；四川科技厅重点研发计划项目(2024YFHZ0011).
作者简介:王静，博士研究生，主要研究方向为平行医疗理论与方法、智慧医疗系统的构建与应用。E-mail：wangjing2014@ia.ac.cn。;申乐，教授，博士，中国医学科学院北京协和医院麻醉科主任，主要研究方向为基于麻醉大数据和人工智能的医疗质量提升与患者安全，疼痛与瘙痒的机制与生物干预研究。主持国家级省部级课题10项，发表学术论文60余篇。E-mail：pumchshenle@163.com。;王飞跃，教授，博士，中国科学院自动化研究所复杂系统管理与控制国家重点实验室主任，主要研究方向为平行系统的方法与应用、社会计算、平行智能以及知识自动化。先后当选IEEE、IFAC、AAAS、ASME等多项国际会士，获得国家自然科学二等奖 1项。E-mail：feiyue.wang@ia.ac.cn。
通讯作者:王飞跃. E-mail：feiyue.wang@ia.ac.cn
更新日期/Last Update:
2026-01-05