Radial Basis Function (RBF) networks for improved gait analysis

The gait analysis is an important component for assessment of rehabilitation of individuals with paralyzed legs. One measure of recovery is the resemblance of muscular activities to patterns characteristic for healthy individuals. We present a method for estimation of muscular activities based on accelerometers, force sensing resistors and Radial Basis Function (RBF) neural networks. This method eliminates the need to record the EMG during walking, being very difficult, especially if electrical stimulation is applied as the orthotic assistive system. The input for the training of RBF network were data from accelerometers and force sensing resistors (FSR), and output data were electromyographic (EMG) activities recorded from major joint flexor and extensor muscles. In order to assess the quality of mapping we used correlation of recorded and RBF predicted EMG signals from data used for training of RBF network, and data not used for training (validation data). The correlations were between 0.9 and 0.94 when network was tested with validation data.