A Review of Performance Variations in SMR-based Brain-Computer Interfaces (BCIs)

The ability to operate a brain-computer interface (BCI) varies not only across subjects but also across time within each individual subject. In this article, we review recent progress in understanding the origins of such variations for BCIs based on the sensorimotor-rhythm (SMR). We propose a classification of studies according to four categories, and argue that an investigation of the neuro-physiological correlates of within-subject variations is likely to have a large impact on the design of future BCIs. We place a special emphasis on our own work on the neuro-physiological causes of performance variations, and argue that attentional networks in the gamma-range (> 40 Hz) are likely to play a critical role in this context. We conclude the review with a discussion of outstanding problems. 1 A brief history of BCI-research From the early days of research on brain-computer interfaces (BCIs) until about a decade ago, subjects had to undergo intensive training in order to acquire the new skill of operating a BCI [1–7]. In the past ten years, machine-learning algorithms have shortened training procedures and enabled higher information transfer rates [8–12]. Even though machine-learning continues to make important contributions to the field, advances have somewhat slowed down: recent studies often report only minor enhancements in classification accuracy [13–15]. At the same time, variations in performance across subjects remain substantial. In a recent study based on a two-class sensorimotor-rhythm (SMR) BCI, 30 out of 80 healthy participants (37.5%) did not achieve a classification accuracy of or above 70%, which is considered as the lower limit for reliable communication [16]. While this constitutes an improvement of 11.2% relative to a large-scale study published in 2003 [17], in which 48.7% of subjects did not 1 In C. Guger et al. (eds.), Brain-Computer Interface Research, SpringerBriefs in Electrical and Computer Engineering, DOI: 10.1007/978-3-642-36083-1_5, 2013.