FIGURE 9.

Association and prediction results. (A) Partial least squares regression maps independent (all included network measures combined) and dependent (all neurobehaviors combined) variables into a latent space to find an aggregate relationship between them. The regression displayed in the figure is performed in the latent space, which contains categorical variables representing all included network measures of the four integration paths (MFG → Insula, Insula → Amygdala, Amygdala → Hippocampus, Hippocampus → Precuneus) and two functional segregation nodes (MFG, Insula) (both of which fit our hypothesis), and the neurobehaviors in latent space. Their association in the latent space could be considered as the net association of all these network measures with all the neurobehaviors. (A) Shows the linear fit in latent space (R² = 0.56, R = 0.75, p = 3.5 × 10^–32). (B) Machine learning classification result, which was obtained through recursive cluster elimination based support vector machine (RCE-SVM) classifier, to classify between control, PTSD and PCS + PTSD groups. (B) Shows worst-case classification accuracies obtained using recursively reducing number of discriminative features (poorer features are successively eliminated). Classification was independently performed with both complex-network measures obtained from the entire brain and non-imaging measures (NIMs). We observed that network measures outperformed NIMs, with approximately 10% superior performance in the final RCE step using top-predictive features of network measures.