《中国康复理论与实践》

《中国康复理论与实践》 ›› 2025, Vol. 31 ›› Issue (3): 249-253.doi: 10.3969/j.issn.1006-9771.2025.03.001

• 专题 运动功能康复 • 上一篇    下一篇

人工智能人体关键点检测技术在步态运动学分析中的效度

刘芳超1, 张元鸣飞1, 武美奇2, 周谋望1, 李涛1,3()   

  1. 1.北京大学第三医院，北京市 100191
    2.中国科学院大学，北京市 101408
    3.北京大学人民医院，北京市 100044
  • 收稿日期:2024-09-21 修回日期:2024-12-10 出版日期:2025-03-25 发布日期:2025-03-25
  • 通讯作者: 李涛，E-mail：tobetheone@126.com
  • 作者简介:刘芳超(1994-)，女，汉族，山东沂水县人，硕士研究生，主要研究方向：骨科与运动损伤康复，国际功能、残疾和健康分类组合。
  • 基金资助:
    1.中国自然科学基金青年基金项目(82202817);2.国家重点研发计划项目(2018YFF0301104)

Validity of key points detection technology of artificial intelligence in gait kinematics analysis

LIU Fangchao1, ZHANG Yuanmingfei1, WU Meiqi2, ZHOU Mouwang1, LI Tao1,3()   

  1. 1. Peking University Third Hospital, Beijing 100191, China
    2. University of Chinese Academy of Sciences, Beijing 101408, China
    3. Peking University People's Hospital, Beijing 100044, China
  • Received:2024-09-21 Revised:2024-12-10 Published:2025-03-25 Online:2025-03-25
  • Contact: LI Tao, E-mail: tobetheone@126.com
  • Supported by:
    National Natural Science Foundation of China (Youth)(82202817);National Key Research and Development Program of China(2018YFF0301104)

RichHTML

PDF (PC)

摘要：

目的 基于人工智能人体关键点检测技术开发一种多参数步态运动学分析测量系统(AI系统)，并验证其测量的有效性。

方法 2022年4月至10月，北京大学第三医院住院的步态异常患者10例，同时招募10例健康受试者，采用标准步态分析系统GaitWatch和AI系统同时采集步态数据。

结果 以GaitWatch为标准，AI系统测得的步频、步态周期、步幅、步速平均绝对误差较小，精确度较高(> 95%)，一致性良好(ICC > 0.75)；髋关节、膝关节、踝关节角度平均绝对误差较大，但相关性良好(r > 0.60)。

结论 开发的步态分析系统对时空步态参数的测量有较高的准确性和一致性，对关节角度的测量误差较大，有待进一步完善。

关键词: 步态分析系统, 运动学, 人工智能, 关键点, 效度

Abstract:

Objective To develop a multi-parameter gait kinematics analysis and measurement system using artificial intelligence key point detection technology (AI system), and verify the effectiveness of the measurement.

Methods A total of ten inpatients with abnormal gait were selected in Peking University Third Hospital from April to October, 2022, while ten normal subjects were recruited. Their gait data were measured using GaitWatch and AI system simultaneously.

Results Compared with those from GaitWatch, the mean absolute error (MAE) of stride frequency, gait period, stride length and stride speed were little from AI system, with high accuracy (> 95%) and good consistency (ICC > 0.75); however, MAE of hip, knee and ankle angles was larger, and correlation was satisfactory (r> 0.60).

Conclusion The novel gait analysis system is accurate and consistent in measuring spatiotemporal gait parameters, but not well in measuring joint angles, which need to be modified.

Key words: gait analysis system, kinematics, artificial intelligence, key points, validity

中图分类号: