Time-on-task effect on workload level discriminability through electroencephalography
Résumé
Passive brain-computer interfaces (BCIs) allow one to monitor cognitive state, such as levels of workload or alertness, using as input an operator's brain waves recorded through electroencephalography (EEG). Previous studies showed that the power of theta and delta bands at frontal sites increases with load, while the power of the alpha band at parietal sites decreases (Schober et al., 1995). When time-on-task increases, the power of low frequency (LF) bands increases as well, while the power of high frequency bands decreases (Klimesch, 1999). However, to this day, no study has evaluated the interaction of those factors at the electrophysiological level. Yet, this interaction might prevent from clearly distinguishing load levels and thus reduce classification accuracy. This study was designed to assess and characterize load and alertness levels, as well as their interaction, by using EEG power spectral density (PSD) measures. Load was manipulated using a modified Sternberg paradigm (Sternberg, 1966) in which 20 participants encoded 2 or 6 digits and responded to a probe. Alertness varied with time-on-task. Participants' reaction times (RTs) and PSD of 7 frequency bands (delta, theta, alpha, low alpha, high alpha, beta and gamma) were evaluated using variance analyses. Preliminary results are in keeping with the literature for each factor's effect. As regards their interaction, in the alert condition, PSD of the delta, alpha, and beta frequency bands at parietal and frontal sites significantly decreased with load, which made the load levels easy to distinguish. While in the low alertness condition, no significant change was observed between load levels, but for a decrement with load in the alpha band at parietal sites. Those results suggest that a low alertness reduces the discriminability of load levels through EEG. Passive BCIs that would take this interaction into account could adjust their parameters accordingly and increase their classification performances.