Artificial intelligence-enabled full-course diagnosis and treatment for patients with severe burns and wounds requiring repair: from risk prediction to causal inference
-
摘要: 严重烧伤与复杂创面的诊疗需要统筹局部组织损伤、全身病理生理变化及连续临床干预。现有诊疗路径能够支持医师基于床旁证据进行判断并随患者病情变化调整治疗方案,但在院前与基层评估的一致性、多模态纵向信息整合以及风险预测结果对临床决策的支持等方面仍存在不足。该文结合严重烧伤与复杂创面的诊疗需求及相关人工智能研究证据,分析现有诊疗路径的能力边界、不同使用者可能获得的临床增益,以及风险预测进一步用于受约束因果推断所需的条件。在此基础上,提出“创面评估—全身监测—临床干预”的全程动态诊疗框架：通过标准化图像分析评估创面面积、深度、坏死、感染及愈合趋势；通过时间序列模型整合生命体征、实验室检验、炎症与感染指标、代谢与凝血状态以及器官功能等信息,连续监测患者全身状态；同时将液体复苏、抗感染、清创与创面覆盖、营养支持和康复等临床干预映射至统一时间轴,以保留患者状态变化、干预时机与临床结局之间的时间序列关系。风险预测用于估计现有诊疗路径下可能发生的临床结局,而反事实推演必须建立在明确的因果假设、适用条件、不确定性估计和责任边界之上,其结果不能直接作为治疗处方。人工智能的临床价值应通过前瞻性研究直接比较医师独立判断与医师结合人工智能判断的效果得出；评价指标除区分度外,还应包括评估耗时、医师间一致性、校准度、预警提前时间、警报与工作流程负担、决策改变情况、患者安全及临床结局。该文围绕现有诊疗路径的能力边界、智能创面评估、全身风险监测、因果推断以及转化应用与治理,讨论人工智能参与严重烧伤与创面修复患者全程诊疗的路径及边界。Abstract: The diagnosis and treatment of severe burns and complex wounds require the coordinated management of local tissue injury, systemic pathophysiological changes, and sequential clinical interventions. Existing diagnostic and treatment pathways support bedside evidence-based decision-making and allow clinicians to adjust treatment as a patient's condition evolves. However, limitations remain in the consistency of prehospital and primary-care assessments, the longitudinal integration of multimodal information, and the use of risk predictions to support clinical decisions. Based on the clinical requirements of severe burn and complex wound care and the available evidence on artificial intelligence, this expert commentary examines the capability boundaries of current care pathways, the potential clinical benefits for different users, and the conditions required to extend risk prediction to constrained causal inference. We propose a full-course dynamic diagnostic and treatment framework comprising wound assessment, systemic monitoring, and clinical intervention. Standardized image analysis is used to assess wound area, depth, necrosis, infection, and healing trajectories. Time-series models integrate vital signs, laboratory measurements, inflammatory and infectious markers, metabolic and coagulation status, and organ function to continuously monitor the patient's systemic condition. Clinical interventions including fluid resuscitation, antimicrobial treatment, debridement and wound coverage, nutritional support, and rehabilitation, are aligned along a unified timeline to preserve the temporal relationships among changes in patient status, intervention timing, and clinical outcomes. Risk prediction estimates outcomes that may occur under the current care pathway, whereas counterfactual simulation requires explicit causal assumptions, clearly defined applicability conditions, uncertainty estimates, and accountability boundaries, and should not be directly interpreted as a treatment prescription. The clinical value of artificial intelligence should be established prospectively by comparing clinicians working independently with clinicians assisted by artificial intelligence. In addition to discrimination, evaluation should include assessment time, inter-clinician agreement, calibration, warning lead time, alert and workflow burden, changes in clinical decisions, patient safety, and clinical outcomes. The article discusses the pathway and boundaries of artificial intelligence-enabled full-course care from the perspectives of current pathway capabilities, intelligent wound assessment, systemic risk monitoring, causal inference, and translational implementation and governance.
-
Key words:
- Artificial intelligence /
- Burns /
- Wounds and injuries /
- Wound healing /
- Critical care /
- Risk assessment /
- Causal inference
-
参考文献
(38) [1] JeschkeMG,van BaarME,ChoudhryMA,et al.Burn injury[J].Nat Rev Dis Primers,2020,6(1):11.DOI: 10.1038/s41572-020-0145-5. [2] GreenhalghDG.Management of burns[J].N Engl J Med,2019,380(24):2349-2359.DOI: 10.1056/NEJMra1807442. [3] ISBI Practice Guidelines Committee,SubcommitteeSteering,SubcommitteeAdvisory.ISBI practice guidelines for burn care[J].Burns,2016,42(5):953-1021.DOI: 10.1016/j.burns.2016.05.013. [4] RowanMP,CancioLC,ElsterEA,et al.Burn wound healing and treatment: review and advancements[J].Crit Care,2015,19:243.DOI: 10.1186/s13054-015-0961-2. [5] ZhangP,ZouB,LiouYC,et al.The pathogenesis and diagnosis of sepsis post burn injury[J/OL].Burns Trauma,2021,9:tkaa047[2026-07-18].https://pubmed.ncbi.nlm.nih.gov/33654698/.DOI: 10.1093/burnst/tkaa047. [6] DvorakJE,LadhaniHA,ClaridgeJA.Review of sepsis in burn patients in 2020[J].Surg Infect (Larchmt),2021,22(1):37-43.DOI: 10.1089/sur.2020.367. [7] SingerM,DeutschmanCS,SeymourCW,et al.The third international consensus definitions for sepsis and septic shock (Sepsis-3)[J].JAMA,2016,315(8):801-810.DOI: 10.1001/jama.2016.0287. [8] LeeKS,YoungA,KingH,et al.Variation in definitions of burn wound infection limits the validity of systematic review findings in burn care: a systematic review of systematic reviews[J].Burns,2022,48(1):1-12.DOI: 10.1016/j.burns.2021.05.006. [9] GreenhalghDG,KileyJL.Diagnosis and treatment of infections in the burn patient[J].Eur Burn J,2024,5(3):296-308.DOI: 10.3390/ebj5030028. [10] HuangS,DangJ,SheckterCC,et al.A systematic review of machine learning and automation in burn wound evaluation: a promising but developing frontier[J].Burns,2021,47(8):1691-1704.DOI: 10.1016/j.burns.2021.07.007. [11] TaibBG,KarwathA,WensleyK,et al.Artificial intelligence in the management and treatment of burns: a systematic review and meta-analyses[J].J Plast Reconstr Aesthet Surg,2023,77:133-161.DOI: 10.1016/j.bjps.2022.11.049. [12] AnisuzzamanDM,WangC,RostamiB,et al.Image-based artificial intelligence in wound assessment: a systematic review[J].Adv Wound Care (New Rochelle),2022,11(12):687-709.DOI: 10.1089/wound.2021.0091. [13] 何志友,王元,张丕红,等.基于卷积神经网络的人工智能烧伤深度识别模型的建立及测试效果[J].中华烧伤杂志,2020,36(11):1070-1074.DOI: 10.3760/cma.j.cn501120-20190926-00385. [14] ChangCW,LaiF,ChristianM,et al.Deep learning-assisted burn wound diagnosis: diagnostic model development study[J].JMIR Med Inform,2021,9(12):e22798.DOI: 10.2196/22798. [15] Nie W Zhang C Song D et al Chest X-ray image classification: a causal perspective Medical Image Computing and Computer Assisted Intervention-MICCAI 2023 Cham Springer 2023 25 35 DOI: 10.1007/978-3-031-43898-1_3 NieW,ZhangC,SongD,et al.Chest X-ray image classification: a causal perspective[C]//Greenspan H, Madabhushi A, Mousavi P, et al. Medical Image Computing and Computer Assisted Intervention-MICCAI 2023. Cham: Springer,2023:25-35.DOI:10.1007/978-3-031-43898-1_3.
[16] Nie W Zhang C Song D et al Instrumental variable learning for chest X-ray classification 2023 IEEE International Conference on Systems, Man, and Cybernetics Piscataway IEEE 2023 4506 4512 DOI: 10.1109/SMC53992.2023.10394601 NieW,ZhangC,SongD,et al.Instrumental variable learning for chest X-ray classification[C]//2023 IEEE International Conference on Systems, Man, and Cybernetics. Piscataway: IEEE,2023:4506-4512.DOI:10.1109/SMC53992.2023.10394601.
[17] NieW,ZhangC,SongD,et al.Deep reinforcement learning framework for thoracic diseases classification via prior knowledge guidance[J].Comput Med Imaging Graph,2023,108:102277.DOI: 10.1016/j.compmedimag.2023.102277. [18] SongJ,ChenH,LyuY,et al.Causality-inspired unsupervised domain adaptation with target style imitation for medical image segmentation[J].IEEE Trans Circuits Syst Video Technol,2025,35(10):10175-10187.DOI: 10.1109/TCSVT.2025.3562650. [19] JohnsonAEW,BulgarelliL,ShenL,et al.MIMIC-IV, a freely accessible electronic health record dataset[J].Sci Data,2023,10(1):1.DOI: 10.1038/s41597-022-01899-x. [20] NieW,YuY,ZhangC,et al.Temporal-spatial correlation attention network for clinical data analysis in intensive care unit[J].IEEE Trans Biomed Eng,2024,71(2):583-595.DOI: 10.1109/TBME.2023.3309956. [21] MoorM,RieckB,HornM,et al.Early prediction of sepsis in the ICU using machine learning: a systematic review[J].Front Med (Lausanne),2021,8:607952.DOI: 10.3389/fmed.2021.607952. [22] LiQ,LiD,JiaoH,et al.CISepsis: a causal inference framework for early sepsis detection[J].Front Cell Infect Microbiol,2024,14:1488130.DOI: 10.3389/fcimb.2024.1488130. [23] LiQ,LiD,NieW,et al.Temporal and spatial analysis in early sepsis prediction via causal disentanglements[J].IEEE Trans Knowl Data Eng,2025,37(8):4860-4872.DOI: 10.1109/TKDE.2025.3569584. [24] LiAT,MoussaA,GusE,et al.Biomarkers for the early diagnosis of sepsis in burns: systematic review and meta-analysis[J].Ann Surg,2022,275(4):654-662.DOI: 10.1097/SLA.0000000000005198. [25] SchultL,HalbgebauerR,KarasuE,et al.Glomerular injury after trauma, burn, and sepsis[J].J Nephrol,2023,36(9):2417-2429.DOI: 10.1007/s40620-023-01718-5. [26] ZhangQ,ZhangW,LiQ,et al.Causal inference model for accurate medical diagnosis in coronary artery bypass graft operation[J].Artif Intell Med,2025,167:103150.DOI: 10.1016/j.artmed.2025.103150. [27] ProsperiM,GuoY,SperrinM,et al.Causal inference and counterfactual prediction in machine learning for actionable healthcare[J].Nat Mach Intell,2020,2(7):369-375.DOI: 10.1038/s42256-020-0197-y. [28] Chang R Jiao H Nie W et al Organ-Agents: virtual human physiology simulator via LLMs 2025-08-20 2026-07-18 https://doi.org/10.48550/arXiv.2508.14357.DOI:10.48550/arXiv.2508.14357 ChangR,JiaoH,NieW,et al.Organ-Agents: virtual human physiology simulator via LLMs[EB/OL].(2025-08-20)[2026-07-18].https://doi.org/10.48550/arXiv.2508.14357.DOI:10.48550/arXiv.2508.14357.
[29] BjörnssonB,BorrebaeckC,ElanderN,et al.Digital twins to personalize medicine[J].Genome Med,2019,12(1):4.DOI: 10.1186/s13073-019-0701-3. [30] Corral-AceroJ,MargaraF,MarciniakM,et al.The 'Digital Twin' to enable the vision of precision cardiology[J].Eur Heart J,2020,41(48):4556-4564.DOI: 10.1093/eurheartj/ehaa159. [31] LiX,LoscalzoJ,MahmudAKMF,et al.Digital twins as global learning health and disease models for preventive and personalized medicine[J].Genome Med,2025,17(1):11.DOI: 10.1186/s13073-025-01435-7. [32] KellyCJ,KarthikesalingamA,SuleymanM,et al.Key challenges for delivering clinical impact with artificial intelligence[J].BMC Med,2019,17(1):195.DOI: 10.1186/s12916-019-1426-2. [33] van der VegtAH,ScottIA,DermawanK,et al.Deployment of machine learning algorithms to predict sepsis: systematic review and application of the SALIENT clinical AI implementation framework[J].J Am Med Inform Assoc,2023,30(7):1349-1361.DOI: 10.1093/jamia/ocad075. [34] CollinsGS,MoonsKGM,DhimanP,et al.TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods[J].BMJ,2024,385:e078378.DOI: 10.1136/bmj-2023-078378. [35] LiuX,Cruz RiveraS,MoherD,et al.Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension[J].Lancet Digit Health,2020,2(10):e537-e548.DOI: 10.1016/S2589-7500(20)30218-1. [36] Cruz RiveraS,LiuX,ChanAW,et al.Guidelines for clinical trial protocols for interventions involving artificial intelligence: the SPIRIT-AI extension[J].Lancet Digit Health,2020,2(10):e549-e560.DOI: 10.1016/S2589-7500(20)30219-3. [37] World Health Organization.Ethics and governance of artificial intelligence for health: WHO guidance[M].Geneva: World Health Organization,2021. [38] CharDS,ShahNH,MagnusD.Implementing machine learning in health care - addressing ethical challenges[J].N Engl J Med,2018,378(11):981-983.DOI: 10.1056/NEJMp1714229. -
Table 1. 严重烧伤与创面诊疗中现有临床方法与人工智能辅助方法的比较
诊疗任务 现有临床方法 现有临床方法主要优势 现有临床方法主要瓶颈 人工智能合理增益 人工智能主要使用者 应比较的临床终点 院前或急诊伤情分层与转诊 致伤史、生命体征、烧伤总面积与部位评估 快速、可解释、便于转诊沟通 评估较依赖临床经验；在非烧伤专科人员参与或大批量伤员场景下,评估结果的一致性不足 辅助开展烧伤面积与深度的标准化量化评估,并提示关键高风险部位 院前人员、非烧伤专科医护人员 评估耗时、评估结果一致性、伤情漏判率、转诊及时性 局部创面动态评估 连续体格检查、带标尺的标准化摄影、组织灌注评估及微生物培养 可结合触感、气味及床旁背景即时纠偏 记录格式和采集条件不一致,观察者差异明显,不同时间点的变化难以量化 自动分割、配准、变化量化及图像质量提示 青年烧伤专科医师、烧伤专科护士及负责复核的专家 测量误差、重复性、复核时间、错误提示率及治疗时机改变 全身严重程度与风险判断 生命体征、实验室检验、器官功能、严重程度评分及临床查房 机制较明确,可随病情即时修正 静态阈值难以捕捉异步多变量轨迹,整合负担高 融合纵向数据并显示风险趋势和不确定性 包含烧伤、重症、感染和康复等专科医师在内的多学科团队 校准度、预警提前时间、警报负担、响应率及器官结局 治疗与修复决策 指南、专科经验及多学科团队讨论 可纳入解剖条件、可用医疗资源、患者意愿及未结构化信息 复杂方案比较依赖经验,群体层面的证据难以直接迁移到个体患者 汇总证据、识别异常轨迹并给出可核查候选路径 负责最终决策的经验丰富的烧伤专科医师 决策改变情况、临床净获益、并发症发生及功能结局 因果推断和反事实比较 机制知识、临床研究及专家判断 临床责任主体明确,可结合具体临床情境解释干预与结局之间的关系 观察性数据易受混杂因素影响,且个体在未接受干预下的反事实结局无法直接观测 仅在明确因果图、适用条件及不确定性前提下开展研究性模拟 包含烧伤专科医师在内的多学科专家团队 外部验证、前瞻性验证、敏感性分析、心理伤害、知情同意及责任追溯 注：人工智能的对照对象应是现行诊疗路径,而非另一种算法；高风险决策须保留人工复核步骤和医师否决通道 
下载: 导出CSV
-
下载:
图(1) / 表(1)
计量
- 文章访问数: 3
- HTML全文浏览量: 4
- PDF下载量: 0
- 被引次数: 0
