EEG Functional Connectivity Predicts Individual Behavioural Impairment During Mental Fatigue.

Abstract

Mental fatigue deteriorates ability to perform daily activities - known as time-on-task (TOT) effect and becomes a common complaint in contemporary society. However, an applicable technique for fatigue detection/prediction is hindered due to substantial inter-subject differences in behavioural impairment and brain activity. Here, we developed a fully cross-validated, data-driven analysis framework incorporating multivariate regression model to explore the feasibility of utilizing functional connectivity (FC) to predict the fatigue-related behavioural impairment at individual level. EEG was recorded from 40 healthy adults as they performed a 30-min high-demanding sustained attention task. FC were constructed in different frequency bands using three widely-adopted methods (including coherence, phase log index (PLI), and partial directed coherence (PDC)) and contrasted between the most vigilant and fatigued states. The differences of individual FC (diff (FC)) were considered as features; whereas the TOT slop across the course of task and the differences of reaction time ( ∆ RT) between the most vigilant and fatigued states were chosen to represent behavioural impairments. Behaviourally, we found substantial inter-subject differences of impairments. Furthermore, we achieved significantly high accuracies for individualized prediction of behavioural impairments using diff(PDC). The identified top diff(PDC) features contributing to the individualized predictions were found mainly in theta and alpha bands. Further interrogation of diff(PDC) features revealed distinct patterns between the TOT slop and ∆ RT prediction models, highlighting the complex neural mechanisms of mental fatigue. Overall, the current findings extended conventional brain-behavioural correlation analysis to individualized prediction of fatigue-related behavioural impairments, thereby moving a step forward towards development of applicable techniques for quantitative fatigue monitoring in real-world scenarios.