摘要：

目的: 开发基于双Kinect V2传感器和通用开源软件平台的低成本数智步态分析系统（DKS）, 评估其用于步态分析的可行性和准确性。方法: 选择2022年12月至2023年12月解放军总医院第一医学中心及新疆维吾尔自治区和田县人民医院确诊的15例步态障碍患者（步态障碍组）及18例健康志愿者（健康组）, 以Right Gait & Posture步态分析系统（RGP）为“金标准”, 采用Pearson相关分析探讨DKS与RGP系统所获步态参数的相关性, Bland-Altman法计算平均差和95% 一致性界限、一致性相关系数（CCC）行一致性检验。结果: 健康组和步态障碍组受试者通过DKS系统所获总步速、左侧步速、右侧步速、左侧步频、右侧步频、步幅、左侧步长、右侧步长、双支撑相、摆动相、支撑相与RGP系统所获各相应步态参数均呈正相关（r > 0, P < 0.05）。健康组通过两个系统获取的步态参数平均差的偏差均为小; 除双支撑相（CCC = 0.572）、摆动相（CCC = 0.603）、支撑相（CCC = 0.569）外, 其余步态参数均显示出较强的一致性（0.712 ≤ CCC ≤ 0.882）; 步态障碍组除总步速、右侧步速、步幅和双支撑相外, 其余步态参数平均差的偏差均为小, 除双支撑相（CCC = 0.524）、摆动相（CCC = 0.352）和支撑相（CCC = 0.421）外, 其余步态参数均显示出较强的一致性（0.716 ≤ CCC ≤ 0.943）。结论: 基于双Kinect V2的数智步态分析系统的准确性可基本满足临床需求且成本低、软件易获取、易携带, 可协助临床医师在门诊、病房甚至家庭环境下进行步态测试和分析。

关键词: 步态障碍，神经性, Kinect V2（非MeSH词）, 数字技术, 人工智能, 步态分析, 可行性研究

Abstract:

Objective: To develop a low-cost digit-intelligent gait analysis system based on dual Kinect V2 sensors and universal open-source software platforms (DKS), and to evaluate its feasibility and accuracy for gait analysis. Methods: A total of 15 patients with gait disorders (gait disorder group)admitted to The First Medical Center of Chinese PLA General Hospital and People's Hospital of Hotan County in Xinjiang and 18 healthy volunteers (healthy group) between December 2022 and December 2023 were included. With the Right Gait & Posture gait analysis system (RGP) serving as the "gold standard", Pearson correlation analysis was employed to assess the correlation between gait parameters obtained from the DKS system and the RGP system. Bland-Altman analysis was used to calculate mean difference and 95% limits of agreement (95%LoA), while the concordance correlation coefficient (CCC) was applied for concordance evaluation. Results: Pearson correlation analysis revealed positive correlations between the DKS and RGP systems in both gait disorder group and healthy group for the following gait parameters: velocity, left/right velocity, left/right cadence, stride length, left/right step length, double support phase, swing phase, and stance phase (r > 0, P < 0.05; for all). Consistency analysis demonstrated that in the healthy group, all gait parameters exhibited mild mean differences between the two systems. Except for the double support phase (CCC = 0.572), swing phase (CCC = 0.603), and stance phase (CCC = 0.569), the remaining parameters showed strong consistency (0.712 ≤ CCC ≤ 0.882). In the gait disorder group, most parameters (excluding velocity, right velocity, stride length, and double support phase) displayed mild mean differences. Except for the double support phase (CCC = 0.524), swing phase (CCC = 0.352) and stance phase (CCC = 0.421), other parameters demonstrated strong consistency (0.716 ≤ CCC ≤ 0.943). Conclusions: The digit-intelligent gait analysis system based on dual Kinect V2 developed in this study can satisfy the clinical accuracy requirements. With its advantages of low cost, easily accessible software, and portability, it can assist clinicians in performing gait testing and analysis in outpatient clinics, hospital wards, and even home settings.

Key words: Gait disorders, neurologic, Kinect V2 (not in MeSH), Digital technology, Artificial intelligence, Gait analysis, Feasibility studies