Character encoding based on occurrence probability enhances the performance of SSVEP-based BCI spellers

Abstract

ObjectiveSteady-state visual evoked potential (SSVEP) is a control signal which is widely used in brain-computer interface (BCI) systems. The SSVEP-based spellers with hierarchical structure have a limitation of low ITR. To improve the ITR in these spellers, we effectively applied the character encoding based on the character frequency rate. MethodsWe proposed the 1–2 level hierarchical structure that allows the user to spell the most used characters just in one stage, while other characters will be selected through two stages. We also considered the latency at the start of each trial, to enhance the SSVEP classification accuracy. To estimate the ITR more accurately, we used a novel ITR definition for the first time, which considers the symbol occurrence probability. ResultsThe proposed speller achieved the mean classification accuracy of 90.5%, the ITR of 48.3 bit/min, and the speed of 13.2 char/min. The latency varies for different subjects, and the mean value of 0.2 was determined across all individuals. ConclusionConsidering the character encoding enhances the performance of SSVEP-based BCI spellers. SignificanceThe proposed speller provides a reliable and easy-to-use assistive communication system for locked-in patients.