ESTIMATION OF GROUND REACTION FORCE FEATURES USING INERTIAL SENSORS DATA IN RUNNING: ACCURACY ASSESSMENT AND SENSITIVITY ANALYSIS WITH RESPECT TO GAIT EVENT IDENTIFICATION OF 12 ALGORITHMS

INTRODUCTION: Running is a widely practiced activity with significant health benefits but also a high risk of lower limb injuries. The study of Ground Reaction Force (GRF) is essential for performance optimization and injury prevention. Traditional GRF measurement instruments ensure high estimation accuracy, but are restricted to laboratory settings, while wearable alternatives like force insoles represent a more practical tool but face durability and foot-shoe interaction issues. Inertial Measurement Units (IMUs) provide a portable solution for in real-world applications, yet their accuracy and robustness in relation to gait event identification relying exclusively on IMU data has not been investigated. This study evaluates the performance of 12 IMU-based algorithms for GRF estimation using Loadsol® insoles as gold standard, aiming to refine estimation techniques for a portable and fully independent alternative to traditional measurement tools. METHODS: Eleven amateur runners (9M2F, age: 26±4 years) performed sprints at 15, 20 and 25 km/h. Two IMUs (OPAL, ±200g, ±2000deg/s, 800Hz) were attached on the participant’s right shank and lower back (L5) and secured with adhesive tape and a belt for the lower back sensor. Loadsol® insoles (Novel, 200Hz) served as gold standard. A total of 1177 steps were analysed separately for each speed. The GRF features examined were the first and second peak, loading rate, average load, impulse, and GRF time series. Algorithms were classified based on sensor placement (shank: 4, sacrum: 8) and methodology (Neural Networks (NN: 2, traditional approaches: 10) A linear mixed-effects model was used for statistical comparison. Additionally, sensitivity analysis was performed by varying initial contact (IC), final contact (FC) timings and stance window, with starting time points identified by selected IMU-based algorithm, assessing their impact on estimation errors. RESULTS: While no single algorithm consistently outperformed the others across all metrics, NN-based methods generally showed better performance, with lower bias and dispersion (e.g. -0.3kN/s bias and limits of agreement 48.9kN/s for the loading rate metric with Kim method at 20km/h). The sensitivity analysis highlighted that estimation errors depended on both the analyzed GRF metric and the specific algorithm. NN-based approaches were more affected by FC misidentifications, while traditional methods showed higher sensitivity to IC and stance window variations. An overall worsening of algorithms’ performances was observed at increasing speeds. CONCLUSION: NN-based algorithms demonstrated the highest accuracy in GRF estimation. Traditional methods proved to be less reliable, particularly for peak force and loading rate measurements. The study highlights the importance of accounting for test speed and ensuring accurate gait event detection, as variations in IC and FC timing significantly affected GRF estimates.