Predicting moment-to-moment attentional state.

Although fluctuations in sustained attention are ubiquitous, most psychological experiments treat them as noise, averaging performance over many trials. The current study uses multi-voxel pattern analysis (MVPA) to decode whether, on each trial of a cognitive task, participants are in an optimal or suboptimal attentional state. During fMRI, participants performed n-back tasks, composed of central face images overlaid on distractor scenes, with low, perceptual, and working memory load. Instructions were to respond to novel faces and withhold response to rare repeats. To index attentional state, reaction time variability was calculated at each correct response. Participants' 50% least variable trials were labeled optimal, or "in the zone," and their 50% most erratic trials were labeled suboptimal, or "out of the zone." Support vector machine classifiers trained on activity in the default mode network (DMN), dorsal attention network (DAN), and task-relevant fusiform face area (FFA) distinguished in-the-zone and out-of-the-zone trials in all tasks. Consistent with evidence that distractors are processed when central task load is low, parahippocampal place area (PPA) classifiers were only successful in the low load task. Classification in anatomical regions across the brain revealed widespread coding of attentional state. In contrast to these robust pattern analyses, univariate signal in DMN, DAN, FFA, and PPA did not distinguish states, suggesting a nuanced relationship to sustained attention. In sum, MVPA can be used to decode trial-by-trial attentional state throughout much of cortex, helping to characterize how attention network fluctuations correlate with performance variability.