Validation of inertial measurement units based on waveform similarity assessment against a photogrammetry system for gait kinematic analysis

When assessing gait analysis outcomes for clinical use, it is indispensable to use an accurate system ensuring a minimal measurement error. Inertial Measurement Units (IMUs) are a versatile motion capture system to evaluate gait kinematics during out-of-lab activities and technology-assisted rehabilitation therapies. However, IMUs are susceptible to distortions, offset and drifting. Therefore, it is important to have a validated instrumentation and recording protocol to ensure the reliability of the measurements, to differentiate therapy effects from system-induced errors. A protocol was carried out to validate the accuracy of gait kinematic assessment with IMUs based on the similarity of the waveform of concurrent signals captured by this system and by a photogrammetry reference system. A gait database of 32 healthy subjects was registered synchronously with both devices. The validation process involved two steps: 1) a preliminary similarity assessment using the Pearson correlation coefficient, and 2) a similarity assessment in terms of correlation, displacement and gain by estimating the offset between signals, the difference between the registered range of motion (∆ROM), the root mean square error (RMSE) and the interprotocol coefficient of multiple correlation (CMC_P). Besides, the CMC_P was recomputed after removing the offset between signals (CMC_Poff). The correlation was strong (r > 0.75) for both limbs for hip flexion/extension, hip adduction/abduction, knee flexion/extension and ankle dorsal/plantar flexion. These joint movements were studied in the second part of the analysis. The ∆ROM values obtained were smaller than 6°, being negligible relative to the minimally clinically important difference (MCID) estimated for unaffected limbs, and the RMSE values were under 10°. The offset for hips and ankles in the sagittal plane reached -9° and -8°, respectively, whereas hips adduction/abduction and knees flexion/extension were around 1°. According to the CMC_P, the kinematic pattern of hip flexion/extension (CMC_P > 0.90) and adduction/abduction (CMC_P > 0.75), knee flexion/extension (CMC_P > 0.95) and ankle dorsi/plantar flexion (CMC_P > 0.90) were equivalent when captured by each system synchronously. However, after offset correction, only hip flexion/extension (CMC_Poff = 1), hip adduction/abduction (CMC_Poff > 0.85) and knee flexion/extension (CMC_Poff > 0.95) satisfied the conditions to be considered similar.