EEG Analysis of Selective Attention during Error Potential BCI experiments

Brain Computer Interfaces (BCI) permit to control external devices through the detection and classification of brain activity. Electroencephalographic (EEG) signal is recorded to interpret this activity and the cerebral responses to specific stimuli can be used as drivers for the BCI system. During BCI tasks the attention of the subject plays a key role on the good performance of the system and engaging protocols are crucial for obtaining reliable results. Attention is particularly related to the Error Potential (ErrP), a specific EEG Evoked Potential (EP) that is elicited whenever an error is detected by the subject, either if it is performed by the subject itself or by another person or by a machine. In this paper an analysis of EEG features related to attention during ErrP–based BCI tasks is presented in order to assess how attention varies during a BCI experiment and how this affects the performance of the final system. The Power Spectral Density (PSD) in $\alpha$ band and the ratio of the PSD in $\beta$ and $\theta$ bands have been chosen as attention descriptors. The obtained results suggest that two subsets of subjects can be distinguished one more focused than the other in terms of attention related EEG features. The more attentive subset also resulted in better performance when in terms of balanced accuracy, using a Convolutional Neural Network for classifying between ErrP and Non-ErrP epochs. These results confirm how crucial subject's attention is during BCI experiments to obtain good performances. Moreover, the differences in ErrP and Non–ErrP epochs in terms of attention related EEG features, suggest that they can be useful descriptors for machine learning algorithms for classifying this EP in BCI application.