The application of a neural network to improve plantar pressure mapping accuracy

Pressure mapping sensors provide portability but to draw valid conclusions their accuracy compared to a gold standard force platform must be established. This study compared original Tekscan readings and those scaled to a neural network (NN) to force plate values obtained during walking and jogging. Correlations between force plate and both original and scaled Tekscan readings were very strong (r > 0.95). Root mean square errors (RMSE) were improved from 3.31 N/kg (jogging) and 1.37 N/kg (walking) in the original Tekscan readings to 0.45 N/kg and 0.21 N/kg, respectively after being scaled by the neural network. This study demonstrates that with minor post-processing, the Tekscan F-Scan system is an accurate and portable tool for measuring ground reaction forces (GRF) compared to gold standard force plate measurements.