Deep Learning Architecture Based on the Combination of Convolutional and Recurrent Layers for ERP-Based Brain-Computer Interfaces

Abstract

In this study, we present a novel deep learning architecture for brain-computer interfaces based on event related potentials (ERP). The topology of the neural network combines convolutional and recurrent layers in order to learn high-level spatial and temporal features. Specifically, our model uses a convolutional layer, intended to detect spatial patterns over the scalp in short periods of time, followed by two bidirectional long-short term memory (BLSTM) layers to extract long-term temporal dependencies within the data. To the best of our knowledge, this is the first time that BLSTM layers are explored for ERP classification. This study takes part in the MEDICON 2019 IFMBE scientific challenge. The model has been evaluated using the provided dataset for the competition (15 subjects with autism spectrum disorder, 7 BCI sessions), achieving an average accuracy of 84% in command selection. In the course of our experiments, this approach outperformed traditional methods, such as step-wise linear discriminant analysis (SWLDA), and other deep learning architectures.