Effects of different wearable sensors and locomotion tasks on machine learning-based joint moment prediction

Joint kinetics play an important role in assessing the mechanical joint load, and their analysis is crucial to understand injury mechanisms and disease progression. Joint kinetics analysis (e.g., joint moments) is commonly conducted using human motion data recorded in a laboratory and biomechanical modeling. The combination of wearable sensors, such as inertial measurement units (IMU) or electromyography (EMG), and machine learning (ML) is increasingly popular to overcome laboratory-based limitations (Gurchiek et al., 2019). However, comparing different studies investigating various sensors and locomotion tasks can be challenging due to variations in ML algorithms, model evaluation techniques, and reported performance metrics (Gurchiek et al., 2019). Therefore, it is currently unclear which type of wearable sensor yields the highest joint moment prediction accuracies using ML. Additionally, comparing joint moment prediction accuracies for different locomotion tasks could provide valuable insight for developing assistive tools (Lee Figure 1.A), and all correlation coefficients between predicted and reference joint moments were greater than 0.91. Different locomotion tasks revealed small differences in mean joint moment prediction accuracy (0.15 Nm/kg ≤ RMSE ≤ 0.17 Nm/kg, 12.7% ≤ relRMSE ≤ 16.3%, 0.93 ≤ r ≤ 0.96; Figure 1.B). However, predicting the hip flexion moment for stair descent revealed reduced accuracy (relRMSE = 22.3 ± 6.9 %, r = 0.78 ± 0.30). These findings demonstrate, consistent with related studies (Lee Moghadam et al., 2023), the potential of predicting lower-limb joint moments for various locomotion tasks by combining wearable sensors and ML. Other factors, such as task diversity, sample size, ML algorithms, and model evaluation, may have a greater impact on joint moment predictions than different wearable sensor-based model inputs. The provided systematic comparison of different wearable sensors and locomotion tasks can assist the advancement of wearable measurement technology for clinical and sports biomechanics. References Camargo, J., Ramanathan, A., Flanagan, W., & Young, A. (2021). A comprehensive, open-source dataset of lower limb biomechanics in multiple conditions of stairs, ramps, and level-ground ambulation and transitions. Journal of Biomechanics, 119, Article 110320. https://doi.org/10.1016/j.jbiomech.2021.110320 Gurchiek, R. D., Cheney, N., & McGinnis, R. S. (2019). Estimating biomechanical time-series with wearable sensors: A systematic review of machine learning techniques. Sensors, 19(23), Article 5227. https://doi.org/10.3390/s19235227 Lee, C. J., & Lee, J. K. (2022). Inertial motion capture-based wearable systems for estimation of joint kinetics: A systematic review. Sensors, 22(7), Article 2507. https://doi.org/10.3390/s22072507 Moghadam, S. M., Yeung, T., & Choisne, J. (2023). A comparison of machine learning models’ accuracy in predicting lower-limb joints’ kinematics, kinetics, and muscle forces from wearable sensors. Scientific Reports, 13(1), Article 5046. https://doi.org/10.1038/s41598-023-31906-z