Estimation of running biomechanical parameters using magneto-inertial sensors: a preliminary investigation

Abstract

Many runners perform training sessions outdoors and on varying surfaces (e.g., track, paved path, trail). Training surface has been shown to influence running spatiotemporal measures and stiffness [1]. However, applied research in understanding training surface influence has been limited by lack of robust methodology to capture biomechanics in-the-field. The increasing availability of wearable magneto- inertial sensors (MIMUs) is now making it possible to perform in-field assessment of running biomechanics [2]. Validation of new inertial-based methods is vital to widespread field use. In this study, we compared inertial-based methods for running biomechanical analysis and against gold-standard stereo-photogrammetric system (SP). Ten recreational male runners (age: 32.3 ± 10.0 years) were enrolled and asked to run on a treadmill at 14 km/hr for 8 trials lasting 6 minutes. We simulated changes in surface in a controlled lab setting by using 8 different footwear conditions. Participants were instrumented with MIMUs (mod: Opal v2, APDM, Portland, USA; fs=200 sample/s) attached on the pelvis (L5) and the instep of both feet, and retroreflective markers on heels, toes, and each MIMU (recorded with a 9-camera Vero system, Vicon, Oxford, UK; fs=200 sample/s). The estimated relevant biomechanical parameters included: stride/stance/swing duration, stride rate, vertical oscillations of the center of mass (CoM), and vertical/leg stiffness. Gait events (initial and final contacts) were estimated using a modified version of the MIMU-based method proposed by Benson et al. [2]. CoM vertical displacement was estimated as the vertical component of the L5 MIMU trajectory. Vertical and leg stiffness were computed starting from CoM displacement, stance interval, and swing interval [3]. For each participant and trial condition, errors with respect to SP were computed (80 values for each parameter) and root mean squared errors (RMSE) across participants and conditions were then calculated. In addition, a Bland-Altman analysis was used to quantify the agreement between MIMU and SP outcomes in terms of 50% limits of agreement interval (LoAI). Differences between MIMU-based and SP-based parameters over all the runners are reported in Table 1. Results showed high accuracy especially in the estimation of stride duration, stride rate and CoM vertical displacement. The level of uncertainty associated to each biomechanical variable, as provided by the LoAIs, can be used to detect differences in surface conditions (footwear conditions). Future analysis and development of field methods will investigate the influence of unconstrained movements (outdoor, no treadmill) in addition to varying surface conditions and the influence of running speed on inertial-based method accuracy.