Utility and validity of group atlas versus personalized functional network approaches for depressive constructs

BackgroundTo understand the neurobiology underlying psychopathology, we need valid measurements of brain function. Group atlases for brain functional connectivity (FC) allow for efficient comparisons, but they fail to account for inter-individual variability in network topography, a problem that personalized methods address. We assess the validity and predictive utility of group and personalized approaches of quantifying FC by 1) comparing effect sizes of associations with clinical metrics; and 2) accounting for spatial features of brain networks when examining the association between FC and clinical metrics. Methods324 teens ages 13-16 participated. Personalized networks were estimated using a hierarchical Bayesian model. Effect size comparisons were done by comparing the correlations between FC and clinical metrics (depression, ruminative coping style, and sensitivity to punishment/reward) with Steiglers Z-test. We also conducted regressions, with clinical metrics as the dependent variables. Those models included FC and spatial features, together and alone. ResultsThe effect size comparisons did not survive FDR correction. However, exploratory permutation tests show that 1) the magnitude of the correlations with depression are larger on average for the intersection estimates of FC than the group estimates; and 2) the magnitude of the correlations with a ruminative coping style are larger on average for the intersection estimates of FC than the personalized estimate. The other comparisons conducted using permutation tests are not significant. Multiple regression analyses demonstrated that only spatial features of networks, not FC, are associated with sensitivity to reward. DiscussionThese results imply that the intersection estimates are more valid than the group estimates, and that the intersection estimates have greater predictive utility than personalized estimates. Further, spatial features of functions networks may be useful in and of themselves in certain contexts. Therefore, researchers in psychiatry should take into consideration functional network topography in order to gain a better understanding of the neurobiology underlying psychopathology.