| Summary: | The present study investigates a Brain-Computer Interface (BCI) spelling procedure based on the P300 evoked potential. It uses a small matrix of words arranged in a tree-shaped organization (‘multimenu’), and allows the user to build phrases one word at a time, instead of letter by letter. Experiments were performed in two sessions on a group of seven healthy volunteers. In the former, the ‘multimenu’ was tested with a total of 60 choices: 30 ‘externally-imposed’ selections and 30 ‘free-choice’ selections. In the latter, 3x3 matrices were compared with 6x6 matrices. Each matrix was composed of letters or words, for a total of four matrices. Differences in classifier accuracy, bit rate and amplitude of the evoked P300 were evaluated.Average accuracy in all subjects was 87% with no differences between the selection methods. The 3x3 ‘multimenu’ obtained the same level of classifier accuracy as the 6x6 matrices, even with a significantly lower amplitude of the P300. Bit rate was increased when using the 3x3 matrices compared to the 6x6 ones. The ‘multimenu’ system was equally effective, but faster than conventional, letter-based matrices. By improving the speed of communication, this method can be of help to patients with severe difficulties in communication.
|
|---|
