Correlates of near-infrared spectroscopy brain–computer interface accuracy in a multi-class personalization framework

Brain–computer interfaces (bcis) provide individuals with a means of interacting with a computer using only neural activity. to date,the majority of near-infrared spectroscopy (nirs) bcis have used prescribed tasks to achieve binary control. the goals of this study were to evaluate the possibility of using a personalized approach to establish control of a two-,three-,four-,and five-class nirs–bci,and to explore how various user characteristics correlate to accuracy. ten able-bodied participants were recruited for five data collection sessions. participants performed six mental tasks and a personalized approach was used to select each individual’s best discriminating subset of tasks. the average offline cross-validation accuracies achieved were 78,61,47,and 37% for the two-,three-,four-,and five-class problems,respectively. most notably,all participants exceeded an accuracy of 70% for the two-class problem,and two participants exceeded an accuracy of 70% for the three-class problem. additionally,accuracy was found to be strongly positively correlated (pearson’s) with perceived ease of session (ρ = 0.653),ease of concentration (ρ = 0.634),and enjoyment (ρ = 0.550),but strongly negatively correlated with verbal iq (ρ = −0.749). © 2015 weyand and chau.