Figure 1
Pre-processing and analysis strategy of the four data sets. The HCP, NKI, RUB, and UMN samples were pre-processed in the same manner (light gray boxes). First, T1-weighted anatomical images were delineated into 180 cortical and 8 subcortical areas per hemisphere. Second, these areas were used as landmarks to extract mean time courses from resting-state images. Third, functional connectivity matrices were built by computing edge weights in the form of BOLD signal correlations. Fourth, pruning was applied to every data set to remove spurious connections from the network. Fifth, networks of all participants were pruned using OMST. Sixth, all functional connectivity matrices and graph theoretical metrics were computed. Seventh, partial correlation coefficients were calculated, and elastic-net regression was applied to investigate the association of graph theoretical metrics and g in all data sets. Eighth, the overlap between the results of the four different samples were compared and prediction models resulting from the elastic-net analysis in each sample were tested in the other independent samples. Finally, reliability of graph metrics was defined using the HCP and NKI samples.
