Predicting Vertical Ground Reaction Force during Treadmill Running Using Principal Component Analysis and Wavelet Neural Network / 医用生物力学

ABSTRACT

Objective To establish the method of predicting the vertical ground reaction force (vGRF) during treadmill running based on principal component analysis and wavelet neural network (PCA-WNN). Methods Nine rearfoot strikers were selected and participated in running experiment on an instrumented treadmill at the speed of 12, 14 and 16 km/h. The kinematics data and vGRF were collected using infrared motion capture system and dynamometer treadmill. A three-layer neural network framework was constructed, in which the activation function of the hidden layers was the Morlet function. Velocities of mass center of the thigh, shank and foot as well as joint angles of the hip, knee and ankle were input into the WNN model. The prediction accuracy of the model was evaluated by the coefficient of multiple correlation (CMC) and error. The consistencies between predicted and measured peak GRF were analyzed by Bland-Altman method. Results The CMC between the predicted and measured GRF at different speeds were all greater than 0.99; the root mean square error (RMSE) between the predicted and measured vGRF was 0.18-0.28 BW; and the normalized root mean square error (NRMSE) was 6.20%-8.42%; the NRMSE between the predicted and measured impact forces and propulsive forces were all smaller than 15%. Bland-Altman results showed that the predicted peak errors of propulsive force at 12 km/h and that of impact force and propulsive force at 14 km/h were within the 95% agreement interval. Conclusions The PCA-WNN model constructed in this study can accurately predict the vGRF during treadmill running. The results provide a new method to obtain kinetic data and perform real-time monitoring on a treadmill, which is of great significance for studying running injuries and rehabilitation treatment.