Value-driven attentional capture enhances distractor representations in early visual cortex

Abstract

When a behaviorally relevant stimulus has been previously associated with reward, behavioral responses are faster and more accurate compared to equally relevant but less valuable stimuli. Conversely, task-irrelevant stimuli that were previously associated with a high reward can capture attention and distract processing away from relevant stimuli (e.g., seeing a chocolate bar in the pantry when you are looking for a nice, healthy apple). Although increasing the value of task-relevant stimuli systematically up-regulates neural responses in early visual cortex to facilitate information processing, it is not clear whether the value of task-irrelevant distractors influences behavior via competition in early visual cortex or via competition at later stages of decision-making and response selection. Here, we measured functional magnetic resonance imaging (fMRI) in human visual cortex while subjects performed a value-based learning task, and we applied a multivariate inverted encoding model (IEM) to assess the fidelity of distractor representations in early visual cortex. We found that the fidelity of neural representations related to task-irrelevant distractors increased when the distractors were previously associated with a high reward. This finding suggests that value-driven attentional capture begins with sensory modulations of distractor representations in early areas of visual cortex.

Conflict of interest statement

The authors have declared that no competing interests exist.

Figures

Fig 1. Value-based decision-making task.

Participants selected 1 of the 2 target stimuli to learn values associated with their colors while ignoring a task-irrelevant distractor that could never be selected and was thus unactionable. Across trials, the colors of the targets and the distractor changed randomly so that the distractor color on a given trial could match the color of a previously selected target that yielded either a low or a high monetary reward (i.e., low- or high-valued distractor).

Fig 2. High-valued distractors increased RTs.

(A) Choice preference for high-valued targets for different distractor types. CW and CCW targets are targets located clockwise and counterclockwise to the distractor location, respectively. (B) The same choice preference data, overlaid with the best-fit cumulative Gaussian function (see Table 1). (C) Distractor-value modulation (high − low distractor value) of the standard deviation (or sigma) and the mean (or mu) of the cumulative Gaussian function that explains choice preference in (B) (also see Table 1). Overall, we observed no distractor-value modulation on choice preference functions: sigma and mu did not change with distractor value in trials in which the current distractor was previously selected or unselected. (D) Unlike choice preference data, we observed a robust distractor-value modulation on RTs. The RT effect was significant only for trials in which the distractor was previously selected. Black *** shows a significant distractor-value modulation compared to 0 with p < 0.001 (2-tailed; resampling test). Red * shows a significant difference between trials in which the current distractors were previously selected and unselected with p < 0.05 (1-tailed). All error bars show ±1 SEM. CW, clockwise; CCW, counterclockwise; n.s., no significant difference; RT, response time.

Fig 3. Quantifying stimulus representations with an IEM.

(A) The IEM was trained using fMRI data from the visuospatial mapping task, in which flickering-checkerboard mapping stimuli were randomly presented at each of 111 locations (center locations shown in blue, red, and yellow dots in the first panels; these dots were not physically presented to participants). We filtered individual stimulus locations using 64 Gaussian-like spatial filters to predict channel responses for each trial. We then use the predicted channel responses and fMRI data of all trials to predict channel weights for each voxel within each visual area. (B) The IEM was tested using fMRI data from the value-based learning task (an independent data set). We inverted the estimated channel weights to compute channel responses within each visual area, resulting in a spatial reconstruction centered at 3 stimulus locations in the value-based learning task. fMRI, functional magnetic resonance imaging; IEM, inverted encoding model.

Fig 4. Distractor value boosted the activation of distractor representations in early visual cortex.

(A) Averaged spatial reconstructions of the selected target, unselected target, and distractor based on fMRI activation patterns in early visual areas (collapsed across V1–V3). The data were sorted based on the distractor value (high and low distractor value) and the selection of previous choices (whether the current distractor was previously selected at least once in 3 prior trials: selected and unselected; also see Materials and methods). Before averaging, reconstructions were rotated so that the positions of each respective stimulus type were in register across subjects. In each color plot, a black dot marks the location of the central fixation, and 3 surrounding dots at 30°, 150°, and 270° polar angle indicate the centers of the selected target, unselected target, and distractor locations, respectively. The bottom panels show difference plots between high- and low-distractor-value conditions. (B) The distractor-value modulation (high − low distractor value) from the reconstruction activation (averaged across black dashed circles in [A]). Overall, we found significant distractor-value modulations in extrastriate visual areas V2 and V3, only in trials in which the current distractor was previously selected. Black ** and *** show significant distractor-value modulations compared to 0 with p < 0.01 and p < 0.001 (2-tailed). Red * and ** show a significant difference between trials in which the current distractors were previously selected and unselected with p < 0.05 and p < 0.01 (1-tailed). The statistics computed for different visual areas were corrected using the Holm-Bonferroni method. All error bars show ± 1 SEM. Blue, red, and black dashed circles in (A) represent the spatial extents of unselected targets, selected targets, and distractors, respectively. fMRI, functional magnetic resonance imaging; n.s., no significant difference.

Fig 5. Target-selection modulations in early visual areas.

(A) The difference between the selected and unselected target reconstruction activation for different target types. The activation values were obtained from averaging the reconstruction activation over circular spaces spanning the spatial extents of target stimuli (red and blue dashed circles in Fig 4A). The data in (A) were collapsed across visual areas. (B) The same data as (A) but plotted separately for different target-value conditions and for different visual areas. ** and *** indicate significant target-selection modulations compared to 0 with p < 0.01 and < 0.001, respectively (2-tailed). ++Significant difference across visual areas V1 and V3. Statistics in (B) were corrected for multiple comparisons with the Holm-Bonferroni method. All subfigures are plotted with ±1 SEM. val, valued.