Asaeda Makoto, Onishi Tomoya, Ito Hideyuki, Miyahara So, Mikami Yukio
Faculty of Wakayama Health Care Sciences, Takarazuka University of Medical and Health, 2252 Nakanoshima, Wakayama, 640-8392, Japan.
Division of Rehabilitation, Department of Clinical Practice and Support, Hiroshima University Hospital, 1-2-3, Kasumi, Minami-ku, Hiroshima, 734-8551, Japan.
Heliyon. 2024 Aug 23;10(17):e36338. doi: 10.1016/j.heliyon.2024.e36338. eCollection 2024 Sep 15.
The consensus on anterior cruciate ligament (ACL) injury prevention involves the suppression of dynamic knee valgus (DKV). The gold standard for evaluating the DKV includes three-dimensional motion analysis systems; however, these are expensive and cannot be used to evaluate all athletes. Markerless motion-capture systems and joint angle calculations using posture estimation have been reported. However, there have been no reports on the reliability and validity of DKV calculations using posture estimation.
This study aimed to clarify the reliability and validity of DKV calculation using posture estimation.
Fifteen participants performed 10 single-leg jump landings from a height of 20 cm, and the knee joint angle was calculated using joint points measured using machine learning (MediaPipe Pose) and motion-capture systems (VICON MX). Two types of angle calculation methods were used: absolute value and change from the initial ground contact (IC). Intra- and inter-rater reliabilities were examined using intraclass correlation coefficients, and concurrent validity was examined using Pearson's correlation coefficients. To examine intra-examiner reliability, we performed single-leg jump landings at intervals of ≥3 days.
The calculation by MediaPipe Pose was significantly higher than that by the 3-D motion analysis systems (p < 0.05, error range 18.83-19.68°), and there was no main effect of knee valgus angle or time on the excursion angle from IC (p > 0.05). No significant concurrent validity was found in the absolute value, which was significantly correlated with the change in IC. Although the inter-rater reliability of the absolute value was low, the change in IC showed good reliability and concurrent validity.
The results of this research suggest that the DKV calculation by pose estimation using machine learning is practical, with normalization by the angle at IC.
前交叉韧带（ACL）损伤预防的共识包括抑制动态膝外翻（DKV）。评估DKV的金标准包括三维运动分析系统；然而，这些系统价格昂贵，且不能用于评估所有运动员。已有关于无标记运动捕捉系统和使用姿势估计进行关节角度计算的报道。然而，尚未有关于使用姿势估计进行DKV计算的可靠性和有效性的报道。
本研究旨在阐明使用姿势估计进行DKV计算的可靠性和有效性。
15名参与者从20厘米的高度进行10次单腿跳跃落地，使用机器学习（MediaPipe Pose）测量的关节点和运动捕捉系统（VICON MX）计算膝关节角度。使用两种角度计算方法：绝对值和从初始地面接触（IC）的变化。使用组内相关系数检查评分者内和评分者间的可靠性，使用Pearson相关系数检查同时效度。为了检查检查者内可靠性，我们以≥3天的间隔进行单腿跳跃落地。
MediaPipe Pose的计算结果显著高于三维运动分析系统（p<0.05，误差范围18.83 - 19.68°），并且膝外翻角度或时间对从IC的偏移角度没有主要影响（p>0.05）。在绝对值中未发现显著的同时效度，其与IC的变化显著相关。虽然绝对值的评分者间可靠性较低，但IC的变化显示出良好的可靠性和同时效度。
本研究结果表明，使用机器学习通过姿势估计进行DKV计算是可行的，并通过IC时的角度进行标准化。
