Neural Decoding: Exploring Inputs for Generalisable Brain Computer Interfaces

: (Alternative Format Thesis)

Abstract

Brain-Computer Interfaces (BCIs) are a rapidly advancing field with the potential to enhance human cognition, communication and mobility. The goal is to develop an ideal BCI system that can: 1) process input with abundant, rich information, 2) use accessible neuroimaging tools to decode the neural information associated with the input, and 3) generalise its performance across tasks and individuals. This doctoral thesis explores various information inputs, neuroimaging tools and the impact of individual differences on neural activity related to visual information. Through this doctoral research, we find that visual input offers considerable information and there is promise for decoding simple visual stimuli using EEG. However, the effectiveness of EEG-based BCIs for decoding complex visual stimuli is hindered by the inherent spatial resolution constraints of EEG. Furthermore, we explore semantic inputs for BCIs, driven by the rationale that incorporating semantic-level information may enhance generalisabilty across tasks and subjects. To facilitate this exploration of semantic-level inputs in EEG-based BCIs, a novel dataset was created, encompassing visual, auditory, perception and imagination across three semantic categories.

Additionally, we discover that a hybrid EEG-fMRI BCI can improve decoding performance for inner-speech, although the degree of effectiveness varies among individuals. We further explore inter-individual variability in two secondary fMRI datasets, and find notable differences in neural activity related to visual information and imagery capacity. Such differences can influence generalisability of BCI systems.

Finally, based on these findings, this thesis offers valuable recommendations for future research directions and proposes scaling up the current body of work. By addressing these recommendations, the field of BCIs can be further advanced, harnessing their potential for augmenting human capabilities.

Date of Award	17 Jan 2024
Original language	English
Awarding Institution	University of Bath
Supervisor	Eamonn O'Neill (Supervisor), Mohammad Golbabaee (Supervisor) & Michael Proulx (Supervisor)