CES-531: Collaborative Brain-Computer Interfaces for Target Detection and Localisation in Rapid Serial Visual Presentation

The rapid serial visual presentation protocol can be used to show images sequentially on the same spatial location at high presentation rates. We used this technique to present aerial images to participants looking for predefined targets (airplanes) at rates ranging from 5 to 12 Hz. We used linear support vector machines for the single-trial classification of event-related potentials from both individual users and pairs of users (in which case we averaged either their individual classifiers' analogue outputs before thresholding or their electroencephalographic signals associated to the same stimuli) with and without the selection of compatible pairs. We considered two tasks - the detection of targets and the identification of the visual hemifield in which targets appeared. While single users did well in both tasks, we found that pairs of participants with similar individual performance provided significant improvements. In particular, in the target-detection task we obtained median improvements in the area under the receiver operating characteristic curve (AUC) of up to 8.3% w.r.t. single-user BCIs, while in the hemifield classification task we ob- tained AUCs up to 7.7% higher than for single users. Furthermore, we found that this second system allows not just to say if a target is in on the left or the right of an image, but to also recover the target's approximate horizontal position.