Experimental paradigm.

Each trial began with the presentation of one of three visual cues for 200 ms. Two of the cues—the instructional cues— instructed the participant to covertly attend to either the left or right visual field, the third cue—the choice cue—prompted participants to choose whether to attend to the left or right field. Shown in the figure is an example of the cue symbol-to-attention mappings which were randomized across participants. After a variable cue-to-target interval (2,000–8,000 ms), a target stimulus (a grating with high or low spatial frequency) appeared briefly (100 ms) with equal probability in either visual field. Participants were asked to discriminate the spatial frequency of the grating presented at the cued or chosen location (attended location), while ignoring stimuli on the uncued or unchosen side (unattended location). Following a variable inter-stimulus interval (ISI; 2,000–8,000 ms), a “?SIDE?” prompt appeared, asking participants to report the visual field they attended to on that trial. Finally, a variable inter-trial interval (ITI; 2,000–8,000 ms) elapsed before the onset of the next trial.

Behavioral results for UF and UCD datasets.

Univariate BOLD activation analysis.

(A) The DAN was activated by both the instructional cues and the choice cue in both datasets. (B) Choice cue, however, additionally activated a frontoparietal network consisting of AI, APFC, DLPFC, dACC, and IPL in both datasets. (AI: anterior insula; APFC: anterior prefrontal cortex; DLPFC: dorsal lateral prefrontal cortex; dACC: dorsal anterior cingulate cortex; IPL: inferior parietal lobule; SFG: superior frontal gyrus).

The MNI coordinates of the frontoparietal regions preferentially activated consistently by the choice cue across both datasets.

These regions are chosen as regions of interest (ROIs) in the sequel.

Decoding accuracy between attend-left and attend-right for instructed (cued) and choice (willed) trials in frontoparietal ROIs during the cue-to-target period.

(A) UF dataset. (B) UCD dataset. (C) Combining UF and UCD via meta-analysis, it was observed that in all 7 ROIs, decisions about where to attend could be decoded only in choice trials but not in instructed trials. *: p≤0.05.

Precue alpha power patterns and postcue direction of attention.

(A) Alpha power decoding accuracy (-500 to 0 ms) for UF and UCD datasets. For both datasets, the postcue direction of postcue attention can be decoded doe choice trials, but not for instructional trials. (B) Precue decoding accuracy time course suggests that only alpha activity immediately preceding the cue onset predicted the post cue direction of attention. *: p<0.05, ***: p<0.001.

EEG informed fMRI analysis.

Subjects were divided into two groups (median split) based on their precue alpha decoding accuracy being high or low. (A) The BOLD activation in the frontoparietal decision network for the two datasets (top) and meta-analysis combining the two datasets (bottom). (B) The decoding accuracy in the frontoparietal decision network for the two datasets (top) and meta-analysis combining the two datasets (bottom). (C) The bootstrap distributions of the neural efficiency for the two groups for the two datasets. * p<0.05, ** p<0.01, ** p<0.001.

The model of willed attention control.

The frontoparietal decision network, under the biasing influence of the ongoing precue brain state, makes the decision about where to attend upon reception of the choice cue and sends the decision to the dorsal attention network. The dorsal attention network executes attention control and issues top-down signals to bias visual areas in anticipation of the upcoming visual processing. Attention selection of sensory input occurs as the result of top-down biasing. Sensory biasing and stimulus section mechanisms are shared between willed and instructed attention.