BCI Illiteracy: It’s Us, Not Them. Optimizing BCIs for Individual Brains

Abstract

Brain-computer-interfaces allow a person to control a device directly via their brain signals. Around 15-30% of people are unable to control a BCI accurately; this is known as the BCI illiteracy problem. However, the problem may lie more with the BCI designer than with the user. Here we explore three potential sources of variability that may contribute to so-called BCI illiteracy, and solutions that may help overcome each one. These approaches bring us closer to having individualized, open-ended BCIs. First, we examine the impact of individual expertise. We show that people who have specialized expertise can often master a mental-imagery-driven BCI task when that task is well matched to their specific training, such as auditory imagery tasks for those with music training. Second, we take into account the high individual variability in brain structure, by using beamforming to transform the EEG signal from electrode space to the spatial coordinates of the brain; we show that individually trained classifiers, when fed features derived from source space, are much more accurate when users are performing non-conventional mental imagery tasks. Third, in our ongoing and future research, we are addressing the issue of cognitive variability – the fact that individuals may perform mental imagery tasks using very different cognitive strategies and engage correspondingly different functional networks. We will have BCI users perform a large battery of imagery tasks based on "approach" versus "avoidance" behaviours. Given that approach versus avoidance processes differentially recruit left versus right hemispheric networks, respectively, we predict that each user will be able to generate highly distinctive and discriminable brain signatures for at least two of these tasks, leading to accurate classification by a BCI. We conclude with some recommendations for mental-imagery-based BCI design based upon our findings to date.