AUTHOR=Van Andel Steven , Mohr Maurice , Schmidt Andreas , Werner Inge , Federolf Peter TITLE=Whole-body movement analysis using principal component analysis: What is the internal consistency between outcomes originating from the same movement simultaneously recorded with different measurement devices? JOURNAL=Frontiers in Bioengineering and Biotechnology VOLUME=10 YEAR=2022 URL=https://www.frontiersin.org/journals/bioengineering-and-biotechnology/articles/10.3389/fbioe.2022.1006670 DOI=10.3389/fbioe.2022.1006670 ISSN=2296-4185 ABSTRACT=

A growing number of studies apply Principal Component Analysis (PCA) on whole-body kinematic data to facilitate an analysis of posture changes in human movement. An unanswered question is, how much the PCA outcomes depend on the chosen measurement device. This study aimed to assess the internal consistency of PCA outcomes from treadmill walking motion capture data simultaneously collected through laboratory-grade optical motion capture and field-suitable inertial-based motion tracking. Data was simultaneously collected using VICON (whole-body plug-in gait marker positions) and Xsens (body segment positions) from 20 participants during 2-min treadmill walking. Using PCA, Principal Movements (PMs) were determined using two commonly used practices: on an individual and a grouped basis. For both, correlation matrices were used to determine internal consistency between outcomes from either measurement system for each PM. Both individual and grouped approach showed excellent internal consistency between outcomes from the two systems among the lower order PMs. For the individual analysis, high correlations were only found along the diagonal of the correlation matrix while the grouped analysis also showed high off-diagonal correlations. These results have important implications for future application of PCA in terms of the independence of the resulting PM data, the way group-differences are expressed in higher-order PMs and the interpretation of movement complexity. Concluding, while PCA-outcomes from the two systems start to deviate in the higher order PMs, excellent internal consistency was found in the lower order PMs which already represent about 98% of the variance in the dataset.