A multicomponent estimation method of single-trial ERPS for BCI applications

Abstract

A novel method is introduced to detect and estimate the P300 and other components in the applications of Brain Computer Interface, where the automatically detection of P300 from single-trial EEGs is the key problem. Recent research work has demonstrated that the amplitudes and latencies of the event-related potentials (ERPs) vary from trial to trial, the features of P300 is unstable and make the detection more difficult. In order to acquire better recognizing performance of P300 component, the Variable Signal Plus Ongoing Activity (VSPOA) model are employed to analyze the EEG waves. Based on this model, the amplitudes and latencies of the components are estimated trial by trial through maximizing the likelihood function. With the estimated scale and shift from this component analysis tool, further analysis is made to determine the existence of typical P300 and its stimulus style. Finally, the trials containing target components can be distinguished from the non-target ones successfully in both tests. Hence this method can be used in the BCI applications. Out method is tested on the simulated datasets and the BCI Competition III datasets, results indicate that this approach is effective and efficient in BCI applications.