Statistical Learning of Distractor Suppression Downregulates Prestimulus Neural Excitability in Early Visual Cortex

Abstract

Visual attention is highly influenced by past experiences. Recent behavioral research has shown that expectations about the spatial location of distractors within a search array are implicitly learned, with expected distractors becoming less interfering. Little is known about the neural mechanism supporting this form of statistical learning. Here, we used magnetoencephalography (MEG) to measure human brain activity to test whether proactive mechanisms are involved in the statistical learning of distractor locations. Specifically, we used a new technique called rapid invisible frequency tagging (RIFT) to assess neural excitability in early visual cortex during statistical learning of distractor suppression while concurrently investigating the modulation of posterior alpha band activity (8-12 Hz). Male and female human participants performed a visual search task in which a target was occasionally presented alongside a color-singleton distractor. Unbeknown to the participants, the distracting stimuli were presented with different probabilities across the two hemifields. RIFT analysis showed that early visual cortex exhibited reduced neural excitability in the prestimulus interval at retinotopic locations associated with higher distractor probabilities. In contrast, we did not find any evidence of expectation-driven distractor suppression in alpha band activity. These findings indicate that proactive mechanisms of attention are involved in predictive distractor suppression and that these mechanisms are associated with altered neural excitability in early visual cortex. Moreover, our findings indicate that RIFT and alpha band activity might subtend different and possibly independent attentional mechanisms.SIGNIFICANCE STATEMENT What we experienced in the past affects how we perceive the external world in the future. For example, an annoying flashing light might be better ignored if we know in advance where it usually appears. This ability of extracting regularities from the environment is called statistical learning. In this study, we explore the neuronal mechanisms allowing the attentional system to overlook items that are unequivocally distracting based on their spatial distribution. By recording brain activity using MEG while probing neural excitability with a novel technique called RIFT, we show that the neuronal excitability in early visual cortex is reduced in advance of stimulus presentation for locations where distracting items are more likely to occur.

Keywords: alpha rhythm; distractor suppression; frequency tagging; magnetoencephalography; statistical learning; visual attention.

Figure 1.

Experimental stimuli and procedure. A, The visual search task. Each trial started with a fixation dot. Afterward, a placeholder screen with four Gabor patches was displayed. Then the search array was presented, and participants had to report whether the target was tilted to the left or right. In 66% of the trial, the target was presented together with a color-singleton distractor. During the whole RIFT periods, the two stimuli at the bottom were shown flickering at a broadband signal (RIFT). B, Statistical learning manipulation. The distractor was presented more often in one hemifield (75%) than in the other (25%). To counterbalance the distractor probability conditions across the two hemifields, each participant took part in two experimental sessions (indicated as day 1 and day 2), with the statistical learning manipulation swapped after the first session.

Figure 2.

RIFT protocol. A, During the frequency tagging period, the two stimuli at the bottom of the screen (indicated by the dotted orange and blue circles) were flickered at very high frequencies (>50 Hz) to generate a steady-state response in visual cortex. B, Examples of two uncorrelated (i.e., left and right stimulation signals) broadband tagging signals. C, Power spectra of the broadband tagging signals used in this study.

Figure 3.

Behavioral results. A, B, Effect of distractor probability on distractor cost expressed in terms of (A) RTs and (B) accuracy divided by session. C, D, Overall, statistical learning (SL) effect on (C) RTs and (D) accuracy. The distractor cost was significantly smaller for high compared with low distractor probability in both RTs and accuracy. Pale dots represent individual median results, white dots represent averaged medians, horizontal solid lines represent averaged means, vertical thick lines represent first (bottom edge) and third (top edge) quartiles.

Figure 4.

RIFT results. A, Results of the cross-correlation analysis for left and right stimulation across lags and MEG sensors (grand average). A correlation peak was detected at a time lag of ∼60 ms, reflecting the time taken for the visual information to reach the primary visual cortex. B, Sensor-level topological representation of normalized correlation value at peak lag for left and right stimulation and for their difference. The strongest correlation was observed over posterior-central sensors contralateral to the stimulated hemifield. C, Source localization of the tagging response. The tagging response was localized in V1/V2. D, Statistical learning (SL) effect on RIFT. The tagging response was significantly weaker for high compared with low distractor probability. Pale dots represent individual median results, white dots represent averaged medians, horizontal solid lines represent averaged means, vertical thick lines represent first (bottom edge) and third (top edge) quartiles.

Figure 5.

Alpha activity results. A, Sensor-level topological representation of baseline-corrected alpha power for the two different statistical leaning configurations and their difference. Asterisks indicate the sensors from where lateralized posterior alpha power was estimated. B, Source localization of alpha activity. C, Time course of alpha power by distractor probability. After the presentation of the placeholder display, contralateral alpha power decreased similarly for high- and low-probability condition. The left and right dotted lines represent the onset of the placeholder display and search array, respectively. D, Statistical learning (SL) effect on lateralized alpha power. There was no evidence for a modulation of distractor probability on alpha power. Pale dots represent individual median results, white dots represent averaged medians, horizontal solid lines represent averaged means, vertical thick lines represent first (bottom edge) and third (top edge) quartiles.