NETWORK MODELS FOR IDENTIFYING MENTAL HEALTH DISORDERS

Abstract

Mental health disorders, much like the neural system responsible for them, are multidimensional concepts of compounding complexity. Characterising psychopathology may require examining large-scale network organisation rather than isolated regions or small-scale networks. Despite the growth of network neuroscience as a field, relatively little work has examined whether global topological features of whole-brain functional networks differentiate affective disorders, particularly in the context of comorbidity. This study investigates how the topological features of complex functional brain networks differ across anxiety, depression and control groups in an adolescent sample. A novel methodological feature of the study was the combination of participant-specific permutation thresholding, separate positive and negative network analyses, and graph theoretic metrics. Resting-state fMRI data from the Boston Adolescent Neuroimaging of Depression and Anxiety (BANDA) cohort of the Human Connectome Project (HCP) were used to construct individual-level correlational networks. To control the family-wise error rate, correlation matrices were corrected using a two-tailed threshold calculated from a participant-specific permuted null distribution. Null values were generated by phase-randomizing each participant’s fMRI time series, computing correlations, and extracting the minimum and maximum values across 10,000 permutations. Each corrected network was split into a positive network and negative network. The negative networks were converted to absolute values and analysed with the same metrics as positive networks. Graph-theoretic metrics included clustering coefficient as a measure of network segregation, average shortest path length as a measure of network integration, and assortativity as an index of network resistance. Group differences in these metrics were assessed using an ANOVA. Significant differences were further analysed using a post-hoc Tukey’s Honestly Significant Difference (HSD) to determine whether comorbid conditions exhibit significantly distinct topological signatures. ANOVA results showed a significant difference between groups only for the assortativity of the positive network. Tukey’s HSD test showed a significant difference between the anxiety and control groups. The difference between anxiety and depression came close but did not reach statistical significance, whereas the difference between depression and control was not significant.