Fig. 2: Functional heterogeneity within ROI sets.
a–c Top panel shows the region of interest parcellation sets colour coded to reflect task positive (orange) or negative (blue) associations quantified using mean activation for each ROI. a The intersection conjunction was segmented using our in-house watershed algorithm. b The task negative to positive20 statistical volume was segmented using the same method. c Unbiased ROI set capturing the whole cortex as defined by24 during resting state fMRI. d Using the above ROI sets we use custom box-violin plots to show the grand mean voxel-wise activation distribution contained within each brain mask (boxes height is defined as the inter quantile range (IQR); i.e. the range from the 3rd to the 1st quantile, the median is drawn as the central line. Lower inner fence is defined as the 1st quantile minus 1.5 times IQR; upper inner fence is defined as the 3rd quantile plus 1.5 times IQR. Finally, the violin plot is calculated from the winsorized kernel density distribution. Full code is avaliable in the GitHub repository under fws.plot.group_violin.m. As expected, the INTR contains only positive values, the MDDM has a bimodal distribution reflecting the task positive and negative networks it captures; and the CRTX mask has a skewed unimodal distribution spanning positive and negative values with a long right tail. e Task-wise distribution plots across ROI sets emphasise the heterogeneity of the different tasks. f Top panel shows the pairwise correlation coefficient matrices for each ROI set across tasks. Note, the inner clustering of tasks is consistent across ROI sets. f Bottom panel shows scree plots for principle component analysis (PCA) of task × ROI, separately for each ROI set across tasks. Note, the optimal number of principle components needed to explain the variance across tasks is at least three for each ROI set, thereby demonstrating the heterogeneity of their activation responses to different tasks. Source data are provided as a Source Data file.
