Gait phase classification from surface emg signals using neural networks

Identification and classification of different gait phases is an essential requirement to temporally characterize muscular recruitment during human walking. The present study proposes a Deep-learning methodology for the classification of the two main gait phases (stance and swing), based on the interpretation of surface electromyographic (sEMG) signals alone. Three different Multi Layer Perceptron (MLP) models are tested to this aim. The present approach does not require specific features to be extracted from the signal, differently from previous studies. 12 healthy adult subjects are analyzed during walking over-ground at comfortable speed. sEMG signals from eight leg muscles are selected. Performance of classifiers is tested vs. gold standard, represented by basographic signals measured by means of three foot-switches. A 10-fold evaluation is computed to take into account the possible variability of the results. The direct comparison among the performances of the three different MLP models shows an average high accuracy over the population (around 95%) for all the models, independent from the increasing complexity. Moreover, the accuracy in each single subject does not fall below 92.6% (range of accuracy variability = 92.6–97.2%). This present study suggests that artificial neural networks may be a suitable tool for the automatic classification of gait phases from electromyographic signals, in overall walking tasks.