Adaptive Spontaneous Brain–Computer Interfaces Based on Software Agents

Abstract

Background: An adaptive Brain–Computer Interface (aBCI) is an extension of a traditional Brain–Computer Interface (BCI). In this work, trial rejection, median filter and software agent are included in a BCI to modify the parameters of its classifier and smoothen the signal feature during task execution. Methods: In this study, a database was used with five spontaneous mental tasks. A software agent was implemented to monitor a BCI for its adaptation. The software agent can learn from the environment and save this information. Results: The statistical significance and the effect size between a BCI and the aBCI proposed here were evaluated in this work. The Information Transfer Rate (ITR) in the aBCI was lower in comparison with BCI, however, the system has statistical significance and high effect size in the accuracy, sensitivity, specificity and kappa coefficient than the latter. Conclusions: Our aBCI improves the performance of a traditional BCI because the software agent can learn from its environment (brain signals) and adjust the BCI’s parameters. The signal quality was used as main factor to tune the feature extraction and parameters of the classifier.