Dynamic stopping in P300 speller with convolutional neural network

Abstract

In P300 speller brain-computer interface (BCI), the stimulus sequence is presented to subject for several rounds to achieve reliable P300 detection. Traditionally, the number of rounds is fixed and relatively large (e.g., 15 in the Wadsworth Dataset of BCI Competition 2005), which results in low information transfer rate. In order to improve the speed of character recognition without affecting the spelling accuracy, we propose to use convolutional neural network (CNN) into the dynamic stopping. Compared with the traditional static stopping criterion (SSC), our method can effectively improve the information transfer rate of the system.