Detection of high frequency steady state visual evoked potentials for Brain-computer interfaces

Brain-computer interfaces (BCI) based on steady-state-visual-evoked-potentials (SSVEP) offer higher information throughput and require shorter calibration periods than other BCI modalities. SSVEPs are oscillatory responses elicited by oscillatory visual stimuli (e.g. using flickering LEDs) that can be detected in the electroencephalogram (EEG). The SSVEP is more prominent in occipital sites and consists of oscillatory components matching that of the stimulus and/or its harmonics. The electrode sites for optimal SSVEP detection change with the frequency of the stimulus. The emphasis here is on SSVEPs elicited by high-frequency stimuli (>30 Hz) because they are minimally perceptible and prevent safety hazards linked to photo-induced epileptic seizures. Linear combinations of EEG signals (spatial filters) are used to construct signals exhibiting large SSVEP components. As in most applications relying on biosignals, individual specificity needs to be taken into account. Thus, the spatial filters need to be customized for each BCI user through a (preferably short) calibration procedure. In this study, we present an approach to automatically obtain the optimum spatial filters to detect the SSVEP at a given stimulation frequency. Our experiments on six subjects resulted on detection rates characterized by values of the area-under-the-ROC ranging from 0.8 to 1 for stimulation frequencies in the 30-45 Hz range.