Signal processing algorithms for brain computer interfacing
mesheuropmc: genetic structures
A brain computer interface (BCI) allows the user to communicate with a computer using only brain signals. In this way, the conventional neural pathways of peripheral nerves and muscles are bypassed, thereby enabling control of a computer by a person with no motor control. The brain signals, known as electroencephalographs (EEGs), are recorded by electrodes placed on the surface of the scalp. A requirement for a successful BCI is that interfering artifacts are removed from the EEGs, so that thereby the important cognitive information is revealed. Two systems based on second order blind source separation (BSS) are therefore proposed. The first system, is based on developing a gradient based BSS algorithm, within which a constraint is incorporated such that the effect of eye blinking artifacts are mitigated from the constituent independent components (ICs). The second method is based on reconstructing the EEGs such that the effect of eye blinking artifacts are removed. The EEGs are separated using an unconstrained BSS algorithm, based on the principles of second order blind identification. Certain characteristics describing eye blinking artifacts are used to identify the related ICs. Then the remaining ICs are used to reconstruct the artifact free EEGs. Both methods yield significantly better results than standard techniques. The degree to which the artifacts are removed is shown and compared with standard methods, both subjectively and objectively. The proposed BCI systems are based on extracting the sources related to finger movement and tracking the movement of the corresponding signal sources. The first proposed system explicitly localises the sources over successive temporal windows of ICs using the least squares (LS) method and characterises the trajectories of the sources. A constrained BSS algorithm is then developed to separate the EEGs while mitigating the eye blinking artifacts. Another approach is based on inferring causal relationships between electrode signals. Directed transfer functions (DTFs) are also applied to short temporal windows of EEGs, from which a time-frequency map of causality is constructed. Additionally, the distribution of beta band power for the IC related to finger movement is combined with the DTF approach to form part of a robust classification system. Finally, a new modality for BCI is introduced based on space-time-frequency masking. Here the sources are assumed to be disjoint in space, time and frequency. The method is based on multi-way analysis of the EEGs and extraction of components related to finger movements. The components are localised in space-time-frequency and compared with the original EEGs in order to quantify the motion of the extracted component.