Figure 5.

Discovering the Functional Exceptions Driving the Turbulent Core in Cognitive Tasks

(A) The functional connectivity in resting state between the 1,000 regions in the Schaefer1000 parcellation averaged across all HCP 1,003 participants.

(B) Similarly, from (A), we computed the global brain connectivity (GBC) as the mean correlation between each region with the rest of the brain (Demirtas et al., 2019) that characterizes the node-level connectivity. The left panel shows the GBC vector (averaged over all participants), and the right panel shows a rendering on the brain of position of the top 20% quantile GBC regions. The top part shows them rendered on the left hemisphere, the middle part shows a rendering on the midline, and the bottom part shows a rendering on a flat map of the left hemisphere.

(C) The left panel shows the top regions with myelination (T1w/T2w) rendered on the brain. As can be seen in the middle panel, there is a strong spatial overlap between the top GBC regions and the top regions with myelination. This can also be seen in the right panel of the overlapping histograms of the top GBC and top myelin regions, as indexed by the spatial location (Schaefer parcellation number), which shows a 46.5% overlap.

(D) Example of the pipeline finding the functional exceptions applied to HCP relational task (see STAR Methods). Here is shown the contrast between the relational task (brown) and the resting state (gray), with the shaded error showing the dispersion across nodes, i.e., all pairs across all participants. The inertial subrange (r = [8.13 33.82] mm) is highlighted with a light-yellow background, and the long-distance correlation subrange (r > 33.82 mm) is shown on a light-gray background. As can been clearly seen, the long-range correlations are mainly increased in task, whereas the inertial subrange correlations remain unchanged (p < 0.001, Wilcoxon rank-sum test). The shadows indicate the standard deviation.

(E) We show a histogram of the difference of the average correlation for each spatial location across the long-distance subrange. The histogram for task (in brown) is clearly showing higher correlations than the histogram for resting (in gray, p < 0.001, Wilcoxon rank-sum test).

(F) We found the most changed long-distance regions in the relational task by thresholding the pair correlation by the maximum pair correlation of the resting condition. The left panel shows a rendering of the top changing regions in the relational task overlaid on the top GBC regions. The overlap is very low (18.2%) as can be seen in the right panel, which shows the overlapping histograms of the top relational (red) and top GBC (blue) regions, as indexed by the spatial location. This is strong evidence that the most changed regions in task are complementary to the unchanged resting GBC regions.