Estimation of Knee Joint Forces in Sport Movements Using Wearable Sensors and Machine Learning

Abstract

Knee joint forces (KJF) are biomechanical measures used to infer the load on knee joint structures. The purpose of this study is to develop an artificial neural network (ANN) that estimates KJF during sport movements, based on data obtained by wearable sensors. Thirteen participants were equipped with two inertial measurement units (IMUs) located on the right leg. Participants performed a variety of movements, including linear motions, changes of direction, and jumps. Biomechanical modelling was carried out to determine KJF. An ANN was trained to model the association between the IMU signals and the KJF time series. The ANN-predicted KJF yielded correlation coefficients that ranged from 0.60 to 0.94 (vertical KJF), 0.64 to 0.90 (anterior–posterior KJF) and 0.25 to 0.60 (medial–lateral KJF). The vertical KJF for moderate running showed the highest correlation (0.94 ± 0.33). The summed vertical KJF and peak vertical KJF differed between calculated and predicted KJF across all movements by an average of 5.7% ± 5.9% and 17.0% ± 13.6%, respectively. The vertical and anterior–posterior KJF values showed good agreement between ANN-predicted outcomes and reference KJF across most movements. This study supports the use of wearable sensors in combination with ANN for estimating joint reactions in sports applications.