OHBM 2022 | 101 | Talk | Yann Harel | Decoding the neural correlates of sustained attention using …

Title: Decoding the neural correlates of sustained attention using MEG and supervised learning.
Session: Talk
Speaker: Yann Harel
The ability to maintain attentional focus over time is known as sustained attention, and is thought to be subtended by cross-frequency interactions between frontomedial theta and sensorimotor alpha and gamma oscillations, responsible for cortical inhibition and excitation respectively. Previous research, primarily using MRI, have shown fluctuations between two modes of performance during a rhythmic sustained attention task : optimal performance (IN-the-zone) in which errors are due to mind-wandering and sub-optimal performance (OUT-of-the-zone) in which errors are caused by cognitive overload. Interestingly, optimal performance has been associated with moderate levels of default-mode-network (DMN) activity. However, little is known about the electrophysiological correlates of optimal and sub-optimal attentional performance. This project aims to uncover the oscillatory correlates of fluctuations in sustained attention performance using magnetoencephalography (MEG) and machine learning classification. Thirty-five participants were recruited to complete 6 blocs of the GradCPT (Esterman et al., 2013), a rhythmic go/no-go task with continuous transitions between stimuli, while their brain activity was recorded using MEG. To differentiate IN and OUT performance, we computed the normalized absolute deviation of reaction times across time, or Variance Time Course (VTC). Periods of stable (optimal) and variable (sub-optimal) reaction times were obtained by a median split of the VTC. We found higher lapses rate and omission errors rate when OUT of the zone. Comparing the oscillatory activity for baseline trials between IN and OUT periods, we found increased posterior alpha activity when IN the zone and increased widespread gamma activity when OUT of the zone. These results were reproduced using a LDA classifier with decoding accuracies reaching 52%. Overall, the differences found in alpha and gamma oscillation are consistent with models of sustained attention that integrate both mind-wandering and executive-control failure accounts of attentional lapses (e.g., Thomson, Besner & Smilek, 2015).