Improving the Performance of Brain-Computer Interface using Deep Learning Algorithms and Event-Related Spectral Perturbation

Abstract

Brain Computer Interface allows disabled people to communicate with the external world by using their brain signals. The main goal of a BCI is to provide patients who suffer form any neuromuscular disorders whith a communication channel based on their brain signals. In this paper, the aim is to explore the effects of applying deep learning algorithms and Event Related Spectral Perturbation analyses on the performance of different EEG-based BCI paradigms. Two paradigms were investigated: one is based on the Matrix paradigm (known as oddball); and the other one utilizes the Rapid serial visual Presentation (RSVP) for presenting the stimuli. Deep learning algorithms of convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) were utilized to evaluate the two paradigms. Our findings showed that Matrix paradigm is more effective in detecting P300 signal. In terms of classification methods, deep learning of CNN algorithm has shown superiority performance in comparison with the other machine learning algorithms.