重症创伤患者疼痛自动评估系统的初步研发及临床验证

Preliminary development and clinical validation of an automatic pain assessment system for severe trauma patients

摘要

目的 研发监测重症创伤患者的疼痛自动评估系统，并进行临床效能验证。方法 纳入2024年5-8月首都医科大学附属北京积水潭医院重症医学科收治的符合入选标准的重症创伤患者，于静息状态、日常疼痛及拍背等时间节点采集高清面部视频，构建数据库。筛选出6个与疼痛表现高度相关的面部动作单元（AU01、AU02、AU04、AU07、AU17、AU25），以注意力机制为框架构建疼痛评估神经网络。与传统人工重症监护疼痛观察工具（CPOT）评分进行比较，从多维度验证系统临床效能。结果 自动疼痛评估系统与人工CPOT的一致性为71.43%，Kappa值为0.43（P<0.001），误报率为28.57%。在33例真实疼痛事件中，系统早发现率为93.94%；12例新发疼痛报警的平均提前时间为246.5 s，可有效捕捉疼痛早期信号并实时预警。结论 该疼痛自动评估系统可实现自动化、持续性的疼痛监测工作，具有一定的识别准确性和临床预警价值。随着模型进一步优化，有望弥补传统人工评估的不足，为重症创伤患者的疼痛管理提供新思路。

Abstract

Objective To develop an automatic pain assessment system for monitoring severe trauma patients and verify its clinical efficacy. Methods Severe trauma patients who met the inclusion and exclusion criteria and were admitted to the Department of Critical Care Medicine, Beijing Jishuitan Hospital Affiliated to Capital Medical University from May to August 2024 were enrolled. High-definition facial videos were collected at key time points, including rest, during daily painful procedures, and back patting, and a corresponding database was established. Six facial action units (AU01, AU02, AU04, AU07, AU17, AU25) highly correlated with pain expression were selected, and a pain assessment neural network was constructed based on the attention mechanism. By comparing with the traditional manual critical-care pain observation tool (CPOT) score, multiple indicators were used to verify the clinical efficacy of the system from multiple dimensions. Results Compared with the traditional CPOT score, the automatic pain assessment system achieved a consistency rate of 71.43%, with a Kappa value of 0.43 (P<0.001) and a false alarm rate of 28.57%. Among 33 real pain events, the early detection rate of the system was 93.94%. For 12 new pain alarms, the average advance warning time was 246.5 seconds, indicating that the system can effectively capture early pain signals and issue real-time alerts. Conclusion The automatic pain assessment system enables automated and continuous pain monitoring for patients, demonstrating certain recognition accuracy and clinical warning value. With further optimization of the model, this approach may compensate for the limitations of traditional manual assessment and provide new insights for pain management in severe trauma patients.

关键词

重症创伤 / 疼痛评估 / 人工智能 / 面部动作单元 / 重症监护疼痛观察工具

Key words

Severe trauma / Pain assessment / Artificial intelligence / Facial action units / Critical care pain observation tool

参考文献

[1] Puntillo KA, Max A, Timsit JF, et al. Determinants of procedural pain intensity in the intensive care unit. The Europain® study[J]. Am J Respir Crit Care Med, 2014, 189(1): 39-47. DOI: 10.1164/rccm.201306-1174OC.

[2] Ayasrah SM. Pain among non-verbal critically Ⅲ mechanically ventilated patients: prevalence, correlates and predictors[J]. J Crit Care, 2019, 49: 14-20. DOI: 10.1016/j.jcrc.2018.10.002.

[3] Rababa M, Al-Sabbah S, Hayajneh AA. Nurses' perceived barriers to and facilitators of pain assessment and management in critical care patients: a systematic review[J]. J Pain Res, 2021, 14: 3475-3491. DOI: 10.2147/JPR.S332423.

[4] Siddiqui AS, Ahmed A, Rehman A, et al. Pain assessment in intensive care units of a low-middle income country: impact of the basic educational course[J]. BMC Med Educ, 2023, 23(1): 567. DOI: 10.1186/s12909-023-04523-7.

[5] Al-Tekreeti Z, Moreno-Cuesta J, Madrigal Garcia MI, et al. AI-based visual early warning system[J]. Informatics, 2024, 11(3): 59. DOI: 10.3390/informatics11030059.

[6] 蔡会文, 马月兰, 刘永戍, 等. 人工智能在新生儿疼痛评估中应用的研究进展[J]. 中华现代护理杂志, 2023, 29(31): 4325-4330. DOI: 10.3760/cma.j.cn115682-20230502-01698.

[7] Rezaei S, Moturu A, Zhao S, et al. Unobtrusive pain monitoring in older adults with dementia using pairwise and contrastive training[J]. IEEE J Biomed Health Inform, 2021, 25(5): 1450-1462. DOI: 10.1109/JBHI.2020.3045743.

[8] Yuan X, Cui Z, Xu D, et al. Occluded facial pain assessment in the ICU using action units guided network[J]. IEEE J Biomed Health Inform, 2024, 28(1): 438-449. DOI: 10.1109/JBHI.2023.3336157.

[9] Gélinas C, Fillion L, Puntillo KA. Item selection and content validity of the critical-care pain observation tool for non-verbal adults[J]. J Adv Nurs, 2009, 65(1): 203-216. DOI: 10.1111/j.1365-2648.2008.04847.x.

[10] Afenigus AD. Evaluating pain in non-verbal critical care patients: a narrative review of the critical care pain observation tool and its clinical applications[J]. Front Pain Res (Lausanne), 2024, 5: 1481085. DOI: 10.3389/fpain.2024.1481085.

[11] Heiderich TM, Carlini LP, Buzuti LF, et al. Face-based automatic pain assessment: challenges and perspectives in neonatal intensive care units[J]. J Pediatr (Rio J), 2023, 99(6): 546-560. doi: 10.1016/j.jped.2023.05.005.

[12] Haug CJ, Drazen JM. Artificial intelligence and machine learning in clinical medicine[J]. N Engl J Med, 2023, 388(13): 1201-1208. DOI: 10.1056/NEJMra2302038.

[13] Lautenbacher S, Hassan T, Seuss D, et al. Automatic coding of facial expressions of pain: are we there yet?[J] Pain Res Manag, 2022, 11: 6635496. DOI: 10.1155/2022/6635496.

[14] Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[J]. Adv Neural Inf Process Syst, 2017, 30.

[15] Thiam P, Kessler V, Amirian M, et al. Multi-modal pain intensity recognition based on the SenseEmotion database[J]. IEEE Trans Affect Comput, 2019, 12(3): 743-760.

[16] Huang Y, Gopal J, Kakusa B, et al. Naturalistic acute pain states decoded from neural and facial dynamics[J]. Nat commun, 2025, 16(1): 4371. DOI: 10.1038/s41467-025-59756-5.

[17] Olugbade T, Buono RA, Potapov K, et al. The EmoPain@Home dataset: capturing pain level and activity recognition for people with chronic pain in their homes[J]. IEEE Trans Affect Comput, 2024.

[18] 支瑞聪, 周才霞. 疼痛自动识别综述[J]. 计算机系统应用, 2020, 29(2): 9-27. DOI: 10.15888/j.cnki.csa.007262.

[19] 秦运俭, 李颖, 陈剑琴, 等. 基于预防重症患者谵妄发生的最佳疼痛控制目标研究[J]. 中华危重病急救医学, 2021, 33(1): 84-88. DOI: 10.3760/cma.j.cn121430-20200828-00600.

基金资助

首都卫生发展科研专项(2026-2-2077)

北京市属医院科研培育计划项目(PX2024015)