ABSTRACT
BACKGROUND: There is a range of magneto-inertial measurement unit (MIMU) systems commercially available, however sensor specifications and fusion methods vary considerably between manufacturers. Such variability can influence the concurrent validity of MIMUs relative to reference standard measurement devices. Different MIMUs have been compared during static or low-velocity conditions, with higher-velocity movements assessed in robotic-based studies. However, there is a need for the concurrent validity of higher-velocity movements to be established in human-based studies. RESEARCH QUESTION: This study aimed to assess the concurrent validity of two commercial MIMU systems (Noraxon and Xsens), relative to a 'gold-standard' retro-reflective motion capture system, when measuring trunk angles during uni-planar range of motion (ROM) and cricket bowling, which involves high-speed, multi-planar movements. METHODS: For this criterion-based validity study, both MIMU systems incorporated comparable sensor specifications and employed Kalman filter sensor fusion algorithms. The MIMU based angles were compared with angles derived from concurrently captured three-dimensional retro-reflective data for 10 fast-medium bowlers. Statistical parametric mapping and root mean squared differences (RMSD) were computed for both MIMU systems. RESULTS: One-dimensional statistical parametric mapping showed no significant differences for angles from both MIMU systems when compared with retro-reflective based angle outputs. The MIMU systems produced ROM RMSDs between 1.4 ± 1.0° and 2.6 ± 1.5°. One system displayed RMSDs between 4.6 ± 1.4° and 7.4 ± 1.9° during bowling, indicating functionally relevant differences to retro-reflective derived angles. There were some small but statistically significant differences in RMSDs between the MIMU systems. SIGNIFICANCE: MIMU-based angle accuracy is poorer during high-speed, multi-planar movement than uni-planar tasks. Comparable MIMU systems can produce varying measurements during ROM and bowling tasks. It is likely that varying sample rates and sensor fusion algorithm parameters contributed to the differences.
