| Brain-computer interface (BCI) for visuomotor tasks requires identification of task-specific patterns of brain activity. During the visuomotor task, the subject encodes the visual information for a decision and then makes a motor action based on that decision. Here, a gradual change of the visual target can unmask the decision-making process leading to a motor preparatory activity such as contralateral movement--selective activity. The consequent neuronal activity in the brain gives rise to small transmembrane currents that can be measured in the extracellular medium, using electroencephalogram (EEG), when recorded from the scalp. EEG activity is quite small, that is very sensitive to noise. Therefore, BCI technologies have relied on finding noise and then removing it before feature extraction. Then, common spatial patterns or manually selecting channels are two common tactics before feature extraction in EEG--based BCI.;In this thesis, we present a different approach wherein the task-relevant signal is extracted from the EEG data leaving behind the task-irrelevant signal and other noise. This approach uses an ICA coupled with its scalp topography to identify relevant task-relevant source signals. This method used an automated approach to find task-relevant sources, with scalp topography related to the task performed, in the same way an expert screens a plausibility of a source. This method is automated and is invariant of the placement of the electrodes. The mapping matrix for the sensor data to the task-relevant sources can be used to generate spatial filters to extract the task-relevant signal from noisy EEG data in a real-time BCI. Moreover, the spatial filter can be used for beam forming of non-invasive electrical brain stimulation (NIBS) under Helmholtz reciprocity principle and the task-relevant sources during motor task performance can be targeted in near real-time with NIBS to facilitate the performance in health and disease, which is future work. |
