2.9 Auditory vs. Visual Correlation Difference Analysis

The probabilistic IPS/TOS and STG/S ROIs were used as seeds in a correlation difference analysis on individual subjects of the HCP469 dataset in order to assess whether additional lateral frontal regions possess differential functional connectivity with the sensory-biased attention networks beyond those identified by the VASA t-fMRI contrast (Michalka et al., 2015). For each subject, the average rs-fMRI signal from ipsilateral STG/S and IPS/TOS ROIs was extracted and used to conduct separate vertex-wise rsFC analyses. This analysis was performed separately for each hemisphere using ipsilateral ROIs. Correlation maps were thresholded at zero to exclude negative correlations before subtracting the IPS/TOS correlation map from the STG/S correlation map. Thresholding was performed for the dual purpose of minimizing the influence of ambiguous negative correlations and to remove anti-correlation with the default mode network often reported for rsFC analyses using an IPS seed. Individual results were aggregated and a group level comparison was performed using PALM, correcting for multiple comparisons (p<0.0001, FWE-corrected, 10000 sign-flips, cluster extend threshold z = 3.1) (Figure 5, Supplemental Figure 5). The resulting brain maps identify regions of significantly greater functional connectivity to ipsilateral STG/S or IPS/TOS ROIs.

Group level results of the correlation difference regression rsFC analysis. By directly contrasting connectivity to IPS/TOS and STG/S probabilistic ROIs, several additional bilateral frontal regions with divergent connectivity to posterior sensory-biased attention regions are revealed. Each prefrontal probabilistic frontal region is extended to include adjacent cortex with matching sensory-bias.

Following correction for multiple comparisons, the 2D spatial gradient for each vertex of the Cohen’s d effect size map was calculated across the cortex by computing the local first spatial derivative using the -cifti-gradient function of the wb_command software package. Local extreme values (min and max) of the effect size gradient map were calculated with wb_command’s -cifti-extrema function. The probabilistic labels were overlayed on these maps and new ROIs, which were refer to as ‘extended network’ ROIs, were manually delineated on the surface using combined information from all three metrics (Cohen’s d, effect size gradient map and the local extrema of the gradient map) and the probabilistic labels (Supplemental Figure 5). Test statistics, significance and average effect size for each ROI are reported in Table 2.

Correlation difference analysis results. Surface area, t statistic and Cohen’s d effect size are reported for equivalent probabilistic LFC ROIs and extended network sensory-biased ROIs identified in the correlation difference analysis. T statistic and Cohen’s d were extracted following group level comparison.