Automatic Recognition of Error Potentials in a P300-Based Brain-Computer Interface

An error potential (ErrP) is an innate event-related potential generated when a subject makes a mistake, and, more relevant to brain-computer interface (BCI) applications, when the BCI itself behaves differently from the user intent. For this reason, error potentials are nowadays attracting attention in the BCI field and the presence of ErrPs has been studied already in a few BCI paradigms. In this paper we investigate the presence and the detection of error potentials in a P300-based BCI speller similar to the one described in [1], where 36 symbols are disposed on a 6× 6 grid, and entire rows and columns of symbols are flashed one after the other in random order. The aim of our research is twofold; first of all, we are interested in developing a method for the automatic detection of ErrPs in a P300 speller, and, secondly, we want to evaluate the real improvement of the performance obtained when ErrP detection is used. Experiments are conducted on five subjects in a controlled scenario, where the outcome of the BCI is actually programmed to generate errors with a 20% probability; users are unaware of this, and they believe to be interacting with a real BCI. Results show that it is indeed possible to recognize an ErrP in an automated fashion when a user interacts with a P300-based BCI, and we also provide a measure of how an automatic error correction based on ErrPs impacts on the overall BCI performance.