Assessing two IMU-based gait event detection methods and their effect on spatiotemporal gait parameters across young and elderly populations.

Gait analysis is crucial for assessing mobility and health in various populations [1]. Accurate gait event (GE) detection – like initial contact (IC) and final contact (FC) - from inertial measurement units (IMUs) is challenging due to movement variability, impacting the computation of spatiotemporal gait parameters (SGP) and insights into balance and mobility. This study compares two different GE detection methods and assesses their impact on SGP computation in young and healthy controls, in healthy elderly controls and in elderly fallers. Research Question: (1) Does the mean absolute error (MAE) in detecting GE vary significantly between the two detection methods?
(2) (How) Do the GE detection methods impact the computation of SGP in the considered groups? The study involved 40 young controls, 41 healthy elderly controls, and 37 elderly fallers. Each participant wore an IMU sensor on their sacrum and completed six trials of back-and-forth walking. IMU acceleration was adjusted for rotation/tilt, and GE were detected with the SensorMotion (based on Butterworth filter) [2] and ScikitDigitalHealth (based on continuous wavelet transform filter) [3] open-source Python libraries. The MAE was calculated to measure the IC and FC difference between the two methods. SGP features were computed considering (a) all GE detections from each method and (b) only detections that were closely aligned between the two methods. SGP per subject were reported as the mean over the six trials. The Shapiro–Wilk test was used to check for normality, T-test or Mann-Whitney U test to determine significance based on data distribution. The average MAE for IC and FC across all trials were 0.0391s and 0.0691s respectively. When considering all the GEs of each method, significant differences between both methods were found in all groups, except for stride regularity and step symmetry in young controls, and step regularity, symmetry, and stride regularity in both healthy and faller elderly groups (see Table 1). When comparing both libraries for closely matching GE detections, significant results were observed only in healthy elderly controls for both the standard deviation and the coefficient of variation of step time.