Novel Spatiospectral Features of ERPs Enhances Brain-Computer Interfaces

Abstract

Constructing accurate predictive models for the detection of event-related potentials (ERPs) is a crucial step to obtain robust Brain-Computer Interface (BCI) systems. In this regard, the majority of previous studies have used spatiotemporal features of ERPs for classification. Recently, we showed that the spatiospectral features of ERP signals also contain significant discriminatory effects in predicting users’ mental intent. In this study, we compare the discriminatory effect of spatiospectral features and spatiotemporal features of electroencephalographic signals. Spectral features are extracted by modeling ERP signals as a sum of sinusoids with unknown amplitudes, frequencies, and phases. Temporal features are the magnitude of ERP waveforms across time. As the classification rule Logistic Regression with L2-Ridge penalty (LRR) is used. We chose this classifier as we recently showed it could achieve high performance using spatiospectral features. We observe that generally by directly using temporal features rather than extracted spectral features even a higher classification performance is achieved.