床旁计算机视觉用于脑损伤康复姿势评估
该研究由临床医生主导开发并初步评估基于计算机视觉的床旁姿势评估系统,用于住院康复的获得性脑损伤患者。系统可用性优秀(SUS 85.3 ± 12.3),所有患者完成测试且无不良事件;坐位颈椎侧屈和站立躯干侧屈随时间显著改善,且与临床医生评定的姿势异常相关。
要点
- 临床医生主导的无标记计算机视觉流程可用性优秀(SUS 85.3 ± 12.3)。
- 所有获得性脑损伤患者完成测试且无不良事件,姿势指标均成功获取。
- 坐位颈椎侧屈和站立躯干侧屈随时间显著改善,与临床评定相关。
核心结论
- 临床医生主导的共同创新产出了可用、可行且具有临床意义的计算机视觉工具,能捕捉治疗相关的姿势变化并与临床医生报告的损伤一致。
研究设计与样本
- 横断面和纵向观察性研究,包含临床医生主导的共同创新、可用性测试、可行性评估、临床效用分析以及与三维运动捕捉的解析验证。
- 研究对象包括康复临床医生10名、获得性脑损伤患者20名(60%女性,年龄45.0 ± 17.7岁)和神经功能完好对照12名。
主要发现
- 临床医生报告可用性优秀(系统可用性量表:85.3 ± 12.3)。
- 所有获得性脑损伤参与者完成测试且无不良事件,所有姿势指标均成功获取。
- 住院神经康复期间,坐位颈椎侧屈(β = -0.084°/天,95%CI:-0.163 至 -0.005,p = 0.037)和站立躯干侧屈(β = -0.096°/天,95%CI:-0.152 至 -0.040,p = 0.0008)随时间显著下降,表明姿势对齐改善。
- 计算机视觉指标与临床医生评定的姿势异常相关,包括无支撑坐位躯干位移(ρ = 0.531,p = 0.023)、骨盆位移(ρ = 0.501,p = 0.034)和头/颈位移(ρ = 0.482,p = 0.043)。
- 与运动捕捉的一致性在颈椎和躯干侧屈方面均强(ICC[A,1] = 0.83)。
关键数值与推荐
- 坐位颈椎侧屈变化:β = -0.084°/天,95%CI:-0.163 至 -0.005,p = 0.037。
- 站立躯干侧屈变化:β = -0.096°/天,95%CI:-0.152 至 -0.040,p = 0.0008。
- 无支撑坐位躯干位移与临床评定相关:ρ = 0.531,p = 0.023。
- 骨盆位移与临床评定相关:ρ = 0.501,p = 0.034。
- 头/颈位移与临床评定相关:ρ = 0.482,p = 0.043。
临床怎么用
- 该计算机视觉工具可用于住院脑损伤康复的床旁坐位和站立姿势评估,捕捉治疗相关变化。
- 工具可用性优秀且可行,所有患者完成测试无不良事件,姿势指标均成功获取。
- 计算机视觉指标与临床医生评定的姿势异常相关,可作为临床评估的辅助参考。
局限性
- 原文未报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
OBJECTIVE: To develop and preliminarily evaluate a clinician-led, computer vision (CV)-based system for point-of-care postural assessment in people with acquired brain injury (ABI) undergoing inpatient rehabilitation. DESIGN: Cross-sectional and longitudinal observational study incorporating clinician-led co-innovation, usability testing, feasibility assessment, clinical utility analysis, and analytical validation against three-dimensional motion capture. SETTING: Tertiary inpatient neurorehabilitation hospital. PARTICIPANTS: Rehabilitation clinicians (n=10), individuals with ABI (n=20; 60% female; age 45.0 ± 17.7 years), and neurologically intact controls (n=12). METHODS: Occupational therapists, physical therapists, engineers, and rehabilitation scientists co-developed a markerless CV workflow for seated and standing postural assessment. Usability, feasibility, therapy-related change, analytical validity, and associations with clinician-reported measures were evaluated. RESULTS: Clinicians reported excellent usability (System Usability Scale: 85.3 ± 12.3). All participants with ABI completed testing without adverse events, and all posture metrics were successfully acquired. During inpatient neurorehabilitation, seated cervical lateral flexion (β = -0.084°/day, 95% CI: -0.163 to -0.005, p = 0.037) and standing trunk lateral flexion (β = -0.096°/day, 95% CI: -0.152 to -0.040, p = 0.0008) decreased significantly over time, indicating improvements in postural alignment. CV metrics correlated with clinician-rated postural abnormalities, including unsupported sitting trunk displacement (ρ = 0.531, p = 0.023), pelvic displacement (ρ = 0.501, p = 0.034), and head/neck displacement (ρ = 0.482, p = 0.043). Agreement with motion capture was strong for cervical and trunk lateral flexion (ICC[A,1] = 0.83 for both). CONCLUSIONS: Clinician-led co-innovation produced a usable, feasible, and clinically meaningful CV tool that captured therapy-related postural change and aligned with clinician-reported impairments in ABI.
引用
- Kirkby E, Burke H, Cline S, et al. Point-of-Care Computer Vision for Postural Assessment in Brain Injury Rehabilitation. Arch Phys Med Rehabil. 2026. PMID: 42927784. doi:10.1016/j.apmr.2026.09.026
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。