Deep recurrent spatio-temporal neural network for motor imagery based BCI

In this paper, we propose a novel architecture of a deep neural network for EEG-based motor imagery classification. Unlike the existing deep neural networks in the literature, the proposed network allows us to analyze the learned network weights from a neurophysiological perspective, thus providing an insight into the underlying patterns inherent in motor imagery induced EEG signals. In order to validate the effectiveness of the proposed method, we conducted experiments on the BCI Competition IV-IIa dataset by comparing with the competing methods in terms of the Cohen's k value. For qualitative analysis, we also performed visual inspection of the activation patterns estimated from the learned network weights.