User-centered brain-computer interface design by optimizing auditory and visual stimuli
A common strategy in Brain-Computer Interface (BCI) research is to first investigate paradigms and data analysis methods in the lab with healthy subjects, before applying them to target users. While research prototypes might neglect user centered design principles, they become important in end user tests. Thus we investigated design parameters of visual and auditory stimuli for fast event-related potential (ERP) BCIs. In the visual domain, six highlighting schemes were compared in a row-column paradigm: brightness, scaling, rotation, color inversion, grid overlay and a combination thereof. Calibrating a media application, the effects were used to enhance rows and columns of photos. EEG data of healthy users (n=6) was collected together with subjective ratings of the effects. ERP analysis revealed that different effects lead to changed distributions of discriminative information. In the light of individual preferences, the careful selection of an effect per subject is advisable, as it improves the estimated classification accuracy. In the auditory domain, samples of spoken phonemes were compared to artificial stimulus tones. Brisk phonemes ('ta', 'to' or 'it') of 3 human speakers resulted in 9 different stimuli. Despite of a more diffuse temporal structure, phonemes represent over-trained stimuli and should thus be easy to perceive. They were used to drive a 9-class BCI with spatial cues. In contrast to a setup with artificial tones, all participants (n = 5) judged the phonemes as pleasant and easy to concentrate on. Observed discriminative N200 and P300 ERPs of the EEG were very similar to those of tones, and as the classification performance was almost equal for both types of stimuli, phonemes are considered a good choice for future patient tests. Funding: This work is supported by the European ICT Programme Project FP7-224631.