Department of Exercise Sciences, Faculty of Science, University of Auckland, Auckland 1023, New Zealand.
Auckland Bioengineering Institute, University of Auckland, Auckland 1010, New Zealand.
Sensors (Basel). 2019 Apr 7;19(7):1660. doi: 10.3390/s19071660.
The aim of this study is to evaluate if Kinect is a valid and reliable clinical gait analysis tool for children with cerebral palsy (CP), and whether linear regression and long short-term memory (LSTM) recurrent neural network methods can improve its performance. A gait analysis was conducted on ten children with CP, on two occasions. Lower limb joint kinematics computed from the Kinect and a traditional marker-based Motion Analysis system were investigated by calculating the root mean square errors (RMSE), the coefficients of multiple correlation (CMC), and the intra-class correlation coefficients (ICC). Results showed that the Kinect-based kinematics had an overall modest to poor correlation (CMC-less than 0.001 to 0.70) and an angle pattern similarity with Motion Analysis. After the calibration, RMSE on every degree of freedom decreased. The two calibration methods indicated similar levels of improvement in hip sagittal (CMC-0.81 ± 0.10 vs. 0.75 ± 0.22)/frontal (CMC-0.41 ± 0.35 vs. 0.42 ± 0.37) and knee sagittal kinematics (CMC-0.85±0.07 vs. 0.87 ± 0.12). The hip sagittal (CMC-0.97±0.05) and knee sagittal (CMC-0.88 ± 0.12) angle patterns showed a very good agreement over two days. Modest to excellent reliability (ICC-0.45 to 0.93) for most parameters renders it feasible for observing ongoing changes in gait kinematics.
本研究旨在评估 Kinect 是否为一种有效的、可靠的儿童脑瘫临床步态分析工具，以及线性回归和长短期记忆（LSTM）递归神经网络方法是否可以改善其性能。对 10 名脑瘫儿童进行了两次步态分析。通过计算均方根误差（RMSE）、多重相关系数（CMC）和组内相关系数（ICC），研究了从 Kinect 和传统基于标记的运动分析系统计算得出的下肢关节运动学。结果表明，基于 Kinect 的运动学具有整体中等至较差的相关性（CMC 小于 0.001 至 0.70），与运动分析具有相似的角度模式。校准后，每个自由度的 RMSE 都有所降低。两种校准方法都表明髋关节矢状面（CMC-0.81 ± 0.10 与 0.75 ± 0.22）/额状面（CMC-0.41 ± 0.35 与 0.42 ± 0.37）和膝关节矢状面运动学的改善程度相似（CMC-0.85±0.07 与 0.87 ± 0.12）。髋关节矢状面（CMC-0.97±0.05）和膝关节矢状面（CMC-0.88 ± 0.12）的角度模式在两天内具有非常好的一致性。大多数参数的中等至极好的可靠性（ICC-0.45 至 0.93）使得观察步态运动学的持续变化成为可能。
