Elastic net ensemble classifier for event-related potential based automatic spelling

Abstract

Abstract Background Brain–computer interface (BCI) refers to a direct communication between the brain and an object to be controlled by information embedded in various signals originated from the brain. Electroencephalogram (EEG) is the most common brain signal modality due to its noninvasiveness, ease-of-use, and high time-resolution. P300-based automatic spelling is the most celebrated EEG-based BCI system, which allows its user to relay a message by just staying focused on letters and numbers displayed on a screen. The objectives of this work are (1) to introduce a novel P300 detection algorithm and (2) to compare its performance against the best current practice algorithm for P300 detection. Results Four volunteers tried to spell letters, “PIRATES”, using our automatic spelling system after proper training. Our results indicate that the proposed P300 detection algorithm can speed up the automatic spelling process since it requires a smaller number of flashing sequences than the best current practice algorithm does to recognize target event-related potentials. While the proposed P300 detection algorithm performs better than the best current practice algorithm, it does not necessarily require a heavy computational burden. Conclusion We designed a novel P300 detection algorithm assuming that the sparsity of EEG signals could be effectively utilized to detect target event-related potentials such as P300. Our pilot study results indicate that utilizing the sparsity of EEG signals can improve the automatic spelling experience.