Gait smoothness using wearable sensors in patients with neurological disorders: a comparison of different metrics

Introduction: Smoothness (i.e. non-intermittency) of movement [1] is a clinically relevant parameter, proven to decrease in neurological populations, like Multiple Sclerosis (MS) or Traumatic Brain Injury (TBI). Many metrics have been proposed to quantify smoothness, differing in the considered data source and mathematical approach, and mainly focused on upper limbs movements [2]. Although locomotion is a major component of daily physical activities and a key to functional independence, it is unclear whether and how these metrics can be applied to locomotion [3]. In addition, inertial measurement units (IMUs) are increasingly used to estimate gait smoothness [1, 4], however no consensus exists about which IMU signal and metrics should be considered. The aim of this study is to compare two metrics commonly used to quantify movement smoothness, considering as data source both acceleration and angular velocity signals measured during locomotion in MS and TBI populations as well as in healthy subjects. Methods: Thirty-three patients suffering from MS (24F; 49.8±9.3y; EDSS 4.0±1.6), 30 patients suffering from TBI (11F; 35.8±12.8y; DGI 17.3±4.7), and 43 healthy adults (HC) (12F; 33.0±10.8y) were enrolled in this study (CE/PROG.700). Three IMUs (APDM Opal, 128Hz) were located on the lower trunk and on both legs. Participants performed a 10m walk Test (10mWT) for three times. Stride segmentation was performed using leg-mounted IMUs, and two smoothness parameters were calculated for each stride and each direction (antero-posterior AP, medio-lateral ML, cranio-caudal CC) using trunk linear accelerations (a) and angular velocities (w) as source data: the Spectral Arc Length (SPARC) and the log dimensionless jerk (LDLJ) [1]. Median and inter quartile range (IQR) were calculated over all strides for each participant and the IQR/median was obtained for each metrics. To test the effect of the considered populations (MS, TBI, HC) and of the considered metrics (SPARCa, SPACRw, LDLJa, LDLJw) on the results, a mixed model ANOVA was run after verifying for normal data distribution.Post- hocanalysis was also performed using Bonferroni-Holmes correction for multiple comparisons. Results: Both population and metrics significantly influenced the smoothness values (p