World's Fastest Brain-Computer Interface: Combining EEG2Code with Deep Learning

Abstract

In this paper, we present a Brain-Computer Interface (BCI) that is able to reach an information transfer rate (ITR) of more than 1200 bit/min using non-invasively recorded EEG signals. By combining the EEG2Code method with deep learning, we present an extremely powerful approach for decoding visual information from EEG. This approach can either be used in a passive BCI setting to predict properties of a visual stimulus the person is viewing, or it can be used to actively control a BCI spelling application. The presented approach was tested in both scenarios and achieved an average ITR of 701 bit/min in the passive BCI approach with the best subject achieving an online ITR of 1237 bit/min. The presented BCI is more than three times faster than the previously fastest BCI and allows to discriminate 500,000 different visual stimuli based on 2 seconds of EEG data with an accuracy of up to 100 %. When using the approach in an asynchronous BCI for spelling, we achieved an average utility rate of 175 bit/min, which corresponds to an average of 35 error-free letters per minute. As we observe a ceiling effect where more powerful approaches for brain signal decoding do not translate into better BCI control anymore, we discuss if BCI research has reached a point where the performance of non-invasive BCI control cannot be substantially improved anymore.