Abstract
Objective Building efficient movement decoding models from brain signals is crucial for many biomedical applications. Moreover, decoding specific movement features, such as speed and force, may provide useful information at the expense of increasing the complexity of the decoding problem. Recent attempts to predict movement kinetics from the electroencephalogram (EEG) achieved classification accuracy levels not better than chance, stressing the demand for more accurate prediction strategies. Thus, the aim of this study was to determine the prediction accuracy of movement kinetics that can be achieved from single-trial EEG signals recorded from healthy volunteers and stroke patients.
Approach A strategy based on convolutional neural networks (ConvNet) was tested, since it has recently shown good performance in the classification of EEG signals. EEG data were minimally pre-processed, in order to mimic online classification scenarios.
Main results Overall accuracy for the 4-class classification problem using ConvNets was close to 80% for healthy volunteers and around 60% for stroke patients.
Significance These results represent a substantial improvement over previously reported results, suggesting that movement kinetics can be accurately predicted from single-trial EEG using ConvNets.
Footnotes
E-mail: rgatti{at}ingenieria.uner.edu.ar