Neural Interactions in a Spatially-Distributed Cortical Network During Perceptual Decision-Making

Abstract

Behavioral experiments evidence that attention is not maintained at a constant level, but fluctuates with time. Recent studies associate such fluctuations with dynamics of attention-related cortical networks, however the exact mechanism remains unclear. To address this issue, we consider functional neuronal interactions during the accomplishment of a reaction time (RT) task which requires sustained attention. The participants are subjected to a binary classification of a large number of presented ambiguous visual stimuli with different degrees of ambiguity. Generally, high ambiguity causes high RT and vice versa. However, we demonstrate that RT fluctuates even when the stimulus ambiguity remains unchanged. The analysis of neuronal activity reveals that the subject's behavioral response is preceded by the formation of a distributed functional network in the β-frequency band. This network is characterized by high connectivity in the frontal cortex and supposed to subserve a decision-making process. We show that neither the network structure nor the duration of its formation depend on RT and stimulus ambiguity. In turn, RT is related to the moment of time when the β-band functional network emerges. We hypothesize that RT is affected by the processes preceding the decision-making stage, e.g., encoding visual sensory information and extracting decision-relevant features from raw sensory information.

Keywords: behavioral response fluctuations; cortical network reorganization; functional brain network; perceptual decision-making task; reaction time.

Figure 1

(A) Complete set of visual stimuli divided into two subsets according to the degree of ambiguity. Classification of cubes with high ambiguity is a high-complexity (HC) task, whereas classification of low-ambiguous cubes is a low-complexity (LC) task. (B) Schematic representation of the experimental protocol.

Figure 2

Different task conditions. Typical probability distribution functions (PDFs) of RT for HC (solid line) and LC (dashed line) tasks and corresponding box-and-whisker diagrams for a single subject (*p < 0.05 via Mann-Whitney U-test for 200 stimuli). M_LC and M_HC correspond to medians of the PDFs.

Figure 3

Time-frequency representation. (A) Typical EEG trial recorded from O1 sensor during background activity [–1,0] s and stimulus processing [0,1] s. The moment of stimulus presentation is indicated with a vertical dashed line. (B) Time-frequency representation of the EEG trial via wavelet transform with highlighted α and β frequency bands. (C) Spectral power of α and β oscillations E^{α, β} extracted from time-frequency representation of EEG trial according to Equation (2).

Figure 4

Illustration of recurrence-based approach for functional links inference. (A) β-band spectral power E^β(t) calculated for the following pair of EEG channels in a single trial: O1–Fp2 (upper panel); O2–Pz (middle panel); Oz–P4 (lower panel). The moment of stimulus presentation is indicated with a vertical dashed line. (B) Left column: RMD|^k(τ) dependence for considered pairs of E^β(t) trials in background (gray) and visual perception (green) activity. Maximal values $RM{D}_{b,t}^{*}{|}^k$ are indicated by horizontal lines. Right column: results of pairwise comparison of maximal RMD values collected over K = 20 trials for background and task-related brain activity via t-test for related samples. Here, ^* indicates significance level of p < α via t-test for related samples corrected for MCP by non-parametric permutation test. (C) β-band functional networks containing links with increasing (upper panel) and decreasing (lower panel) coupling strength related with visual stimuli processing. Bold arrows indicate links selected for recurrence-based method demonstration.

Figure 5

Reaction times. (A) Median RT for HC and LC tasks averaged over all subjects (**p < 0.01 via Wilcoxon signed-rank test, n = 20 subjects). (B) Median RT for SE2, SH2, and SH1 conditions averaged over all subjects (*p < 0.05, via repeated measures ANOVA with Bonferroni correction, n = 20 subjects). (C) Median presentation times over the course of the experiment for stimuli belonging to SE2, SH2, and SH1 conditions averaged over all subjects (^*p < 0.05 via repeated measure ANOVA with Bonferroni correction, n = 20 subjects).

Figure 6

Functional network reconfiguration. (A,B) Ratio R of the number of increasing functional links to the number of decreasing links in β- and α-frequency bands. The arrows indicate the moments of time t₁(SE2, SE2, SH1) when R exceeds the level of R = 1 shown by the dashed horizontal line. t₃(SE2) is a time moment when R reaches the maximal value for SE2 condition. The vertical dashed lines indicate median RT and the shaded area around these lines illustrates RT distributions in the 25th–75th percentile. (C) Median values of t₁ and median RT for each of the three conditions (SE2, SH2, SH1) for the group of 20 subjects. The data are shown as group mean ± SE (^*p < 0.05 via repeated measures ANOVA and post-hoc pairwise comparison with Bonferroni correction).

Figure 7

Properties of decision-making functional connectivity in β-band. (A) Schematic illustration of selected EEG sensor regions. (B–D) The degree D (mean ± SE) vs. EEG sensor region (horizontal axis) and time (different color) for different experimental conditions (SE2, SH2, and SH1) (^*p < 0.05 via repeated measures ANOVA with Bonferroni correction).