A Brain-Computer Interface for Decoding Decisions During Learning in a Multidimensional Environment

Abstract

The aim of this work was to develop a brain-computer interface (BCI) for decoding decision from neuronal recordings. This was accomplished by collecting a novel dataset of neural recordings during a feature learning task, identifying the information encoded in the data, and lastly using this knowledge to create a cognitive BCI for decoding decision A variational autoencdoer based decoder is developed. The encoder and decoder portions were constructed using gated recurrent units and a classifer off of the latent space used long short-term memory units. This network architecture was specifically designed for decoding decision features, achieving peak decoding accuracy approaching 15%. The findings reveal significant correlations between neural activity and cognitive variables, underscoring the roles of the anterior cingulate cortex, prefrontal cortex, and caudate in the decision-making process. Additionally, an importance analysis identifies key channels contributing to decision decoding. The conclusions drawn from this work highlight the potential for real-time monitoring of latent behavioral variables.