Abstract/Summary

Brain Computer Interface (BCI) empowers individuals with severe movement impairing conditions to interact with the computers directly by their thoughts, without the involvement of any motor pathways. Motor-based BCIs can offer intuitive control by merely intending to move. Hence, to develop effective motor-based non-invasive BCIs, it is essential to understand the mechanisms of neural processes involved in motor command generation in electroencephalography (EEG). The EEG consists of complex narrowband oscillatory and broadband arrhythmic processes. However, there is more focus on the oscillations in different frequency bands for studying motor command generation in the literature. The narrowband processes such as event-related (de)synchronisation (ERD/S) and movement-related cortical potential (MRCP) are commonly used for movement detection. Analysis of these narrowband EEG components disregards the information existing in the rest of the frequencies and their dynamics. Hence, this thesis investigates various facets of previously unexplored temporal dynamics of neuronal processes in the broadband arrhythmic EEG to fill the gap in the knowledge of motor command generation on a single trial basis in the BCI framework. The temporal dynamics of the broadband EEG were characterised by the decay of its autocorrelation. The autocorrelation decayed according to the power-law resulting in the longrange temporal correlations (LRTC). The instantaneous ongoing changes in the broadband LRTC were uniquely quantified by the Hurst exponent on very short EEG sliding windows. There was an increase in the temporal dependencies in the EEG leading to slower decay of autocorrelation during the movement and significant increase in the LRTC (p<0.05). Different types of temporal dependencies in the broadband EEG were comprehensively examined further by modelling the long and short-range correlations together using autoregressive fractionally integrated moving average model (ARFIMA). The short-range correlations also changed significantly (p<0.05) during the movement. These ongoing changes in the dynamics of the broadband EEG were able to predict the movement 1 s before its onset with accuracy higher than ERD and MRCP. The LRTCs were robust across participants and did not require determination of participant specific parameters such as most responsive spectral or spatial components.