Optimization and validation of a wearable-based method for the ecological estimation of base of support parameters in healthy subjects

Ecological monitoring of gait stability is a key factor in preventing falls in older adults and patients with neurological disorders. The quantification of base of support (BoS) parameters (i.e., stride width and step length) has been observed to provide crucial insights into dynamic stability. Traditionally, BoS parameters are assessed in laboratory settings using optical motion capture systems or instrumented mats, which limit ecological validity and accessibility. In recent years, a few studies proposed wearable-based methods using magneto-inertial measurement units (MIMUs) and, in addition, auxiliary sensors to estimate foot placement and inter-foot distance using machine learning or sensor fusion approaches. However, these methods have been validated exclusively during straight walking, whereas instability often emerges in more complex contexts such as turning or daily-life activities. This thesis aimed at optimizing and validating the method proposed by Rossanigo et al. (2023) which estimates BoS parameters through the integration of IMU and infrared time-of-flight sensors. (M1) on both straight and curvilinear walking), Furthermore, the method M1 was compared with a recent IMU-based machine learning approach proposed by Wang et al. (2024) (M2). A total of 18 healthy participants were recruited to perform an experimental protocol including straight walking at three self-selected speeds (slow, comfortable, and fast), structured tests (a timed up and go test, a 2-minute walking test, a hallway test), and a simulated daily activities scenario. Participants were equipped with MIMUs on feet and pelvis. Furthermore, a single foot was instrumented with a wearable system housing not only the MIMU but also two medial infrared time-of-flight distance sensors. An ad-hoc calibration procedure was implemented to align the foot-mounted wearable system to the foot longitudinal axis. M1 exploited the distance data recorded during swing phases to estimate inter-foot distance, thus being able to describe both feet positions with respect to the same reference system and compute BoS parameters. Reference data were simultaneously collected using a stereophotogrammetric system to compute errors on BoS parameters estimated with methods M1 and M2. More than 3,000 strides were analyzed. M1 outperformed M2 in both straight and curvilinear walking, halving the obtained errors. Across all walking conditions, M1 achieved a root mean square error equal or below 0.05 m on stride width and 0.1 m on step length and demonstrated a high to excellent correlation with reference data (ρ = 0.65-0.91). These results indicate that the optimized sensor fusion method M1 can reliably detect subtle changes in BoS parameters across different gait scenarios, including those simulating real-life activities, thus being suggested for the ecological monitoring of descriptors of dynamic stability.