Abstracts

Rationale: Many people affected by neurological or neuromuscular disorders such as amyotrophic lateral sclerosis (ALS), brainstem stroke, and spinal cord injury, are impaired in their ability or unable to communicate with their family and caregivers. A Brain-Computer Interface (BCI) uses brain signals directly - rather than muscles, to restore communication. One particular implementation of a BCI is a matrix-based speller as described by Farwell and Donchin (1988). This speller makes use of different event-related potentials (ERPs) of the EEG, including the P300 evoked response. In this system, the user pays attention to a character in a matrix while each row and column is intensified in a random sequence. Typically, several sequences of intensifications are necessary to correctly detect the row and column, and thus the character that the user intends to spell. Recent EEG-based studies (Sellers et al., 2006; Nijboer et al., 2008) report an average of 79% and 80% accuracy in selecting the correct letter (2.8% chance) using 20 and 10 intensification sequences, respectively. Thus, these BCIs supported a selection every 42 and 20 seconds, respectively. At this rate, it takes several minutes to convey a single message. Methods: A growing number of recent studies (Leuthardt et al., 2004; Schalk et al., 2008) suggest that the invasive monitoring in the course of epilepsy surgery assessments affords unique opportunities to utilize signals recorded from the brain (electrocorticography (ECoG)) for BCI research and application. Because ECoG has higher fidelity than EEG, it may allow to reduce the number of sequences, and thus to improve the communication performance of a matrix speller BCI. Results: In this study, we investigated the communication performance of a matrix speller using ECoG signals recorded from frontal, parietal, and occipital areas in one subject. The BCI was first calibrated using data from a 1 min training session. Subsequently, the subject spelled a character every 3.15 seconds during 23 minutes of data collection (444 selections total). The average spelling accuracy was 86% (2.8% chance), and the subject corrected all errors by selecting backspace. We used these data to further optimize system parameters. Using these optimized parameters, the subject was able to spell a character every 2 seconds at an accuracy of 86%. Post-hoc analysis revealed that the accuracy could have been further increased to 97%. Conclusions: The results of this study suggest that ECoG may support substantially higher performance with the matrix-based BCI than does EEG. Thus, with additional verification in more subjects, these results may further extend the communication options for people with severe neurological or neuromuscular disorders.