313 — Modelling the brain response to arbitrary visual stimulation patterns for a flexible high-speed BCI
This research proposes a new technique for interpreting the onset of visual evoked potentials (VEPs) in EEG signal — a common strategy employed in order to close the feedback look from a brain-computer interface. While techniques exist to identify and measure VEPs, the authors explain that these are highly bandwidth-constrained due to their assumptions about how VEPs are composed in the brain.
Nagel & Spüler propose EEG2Code, a model which can convert spatially filtered EEG into a prediction of the stimulation pattern that was used to arrive at that EEG signal. The data are windowed into a sliding 250ms window, and each of these windows is converted to a binary representation of stimulation pattern.
But the authors go further and also propose the reverse model — given a stimulation pattern, Code2EEG predicts a reasonable EEG recording that might have corresponded to that stimulation pattern.
Both of these methods can be used for BCI control (for either stimulation pattern prediction or brain response prediction, respectively). Furthermore, this can be used to begin to bound the SNR of the EEG recording, since the trained model couldn’t predict the noise inherent in the data and instead returns a lower-frequency output.