Validity of recurrent neural networks to predict pedal forces and lower limb kinetics in cycling

Dynamic variables contribute to understand the mechanics of pedalling and can assist with injury prevention. Measuring pedal forces and joint moments and powers has a high cost, which can be mitigated by using trained artificial neural networks (ANN) to predict forces from kinematics. Thus, this study aimed at training and validating recurrent ANN to predict 3D pedal forces, lower limb joint moments and powers from lower limb kinematics. Ergometer pedalling data from seventeen cyclists recorded in a single laboratory session were used to train the ANN, where various ergometer power outputs and cadences were combined. A different dataset with ten cyclists was utilized to test the ANŃs performance. Statistical Parametric Mapping (SPM) was performed to explore significant correlations between measured and predicted kinetic variables throughout the pedal cycle. Mean correlation values ranged from 0.79 to 0.96 and all variables exhibited significant positive correlations at their peak absolute values (p < 0.05), except for the anteroposterior (p = 0.28) and mediolateral (p = 0.51) pedal forces and the knee flexion power (p = 0.33). The maximum prediction errors of the ANN in the sagittal plane were 12.1 % for the pedal forces, 17.2 % for the net joint moments and 9.4 % for the joint powers, while for non-sagittal plane were 13.0 %, 28.9 % and 24.0 %, respectively. Thus, the ANN produces kinetic data in cycling within the errors expected from the variability between assessment days.