Toward gaze-independent brain-computer interfaces.

The ability to communicate by speech, text or gestures is essential to human interaction. This ability is impaired in many people who are affected by debilitating neuromuscular disorders such as amyotrophic lateral sclerosis (ALS), brainstem stroke, or spinal cord injury. Conventional assistive devices (e.g., letter boards, cheek or tongue switches, or eye trackers) that aim to restore communication functions all require muscular control, which is often lost in the progress of neuromuscular disorders. A brain-computer interface (BCI) uses brain signals, rather than muscular control, to establish communication with the outside world. Thus, BCI systems may be useful for restoring communication functions to people with or without disabilities. Many BCI systems described in the literature are based on event-related potentials (ERPs). In an ERP-based BCI system, the user communicates his intention by selectively attending to a desired external stimulus. ERPs are different for desired and undesired stimuli. The BCI system uses this difference to determine the desired stimulus from the brain signals. One well-known implementation of an ERP-based BCI is the so-called ‘‘P300 matrix speller’’ that was first described by (Farwell and Donchin, 1988). In this system, the user pays attention to a character in a matrix while each row and column is intensified rapidly and randomly. The brain produces ERPs to the row or column containing the intended character; ERPs are smaller for the other rows or columns. The BCI typically averages several ERPs, detects the row and column with the strongest ERP, and thereby identifies the character the user wants to select. The ERPs in question are composed of an endogenous component (modulated by covert attention), as well as an exogenous visual-evoked potential (VEP) component. The communication performance of the matrix-speller depends on the extent to which these two ERP components are modulated by attention to the target stimulus. As a result, most of the work on ERP-based BCI systems has focused on optimizing the stimulation parameters to maximize the ERP response to the target stimulus. These parameters included matrix size (Allison and Pineda, 2003), stimulation frequency (Sellers et al., 2006), and stimulation intensity (Takano et al., 2009). By combining optimized stimulation parameters and improved classification algorithms (Krusienski et al., 2006), a recent study (Guger et al., 2009) showed that 80% of the healthy population can make effective use of the matrix speller BCI. The wide applicability of this approach in people without disabilities has been further demonstrated in several application contexts, such as web browser navigation (Mugler et al., 2008), environmental control (Edlinger et al., 2009), wheelchair navigation (Rebsamen et al.,