Energy landscape analysis of depressive brain state dynamics

Purpose: Depression biomarkers can provide objective diagnosis for major depressive disorder (MDD).Studies suggest that dynamic transitions are involved in depressive brains.Recently, a data-driven method called energy landscape analysis (ELA) has been applied to neuromaging data to understand brain state dynamics.In this study, we applied ELA to neuroimaging data to compare the brain state dynamics of healthy and depressed subjects.Methodology: We analyzed the resting state fMRI of hundreds of healthy subjects and patients with MDD.For ELA, we binarized the BOLD fMRI signals on several functional brain networks such as cingulo-opercular, default mode, fronto-parietal networks.With the samples from healthy and depression subject groups, we computed the probability distribution of all possible states, where a state corresponds to an array of active and inactive brain networks.We then fitted a Pairwise Maximum Entropy Model which is an indicator calculated from region-specific activity and their pairwise interactions.From the model we analyzed the macro-states, which are the states with highest probability with respect to neighboring states.Results: Our model generated macro-states which exhibited desynchronization pairings, where one state is the complete inverse of another.We analyzed the state-transition probabilities of healthy and depressed subjects, and observed small between-group differences.However, results are highly varied across individuals within group.Conclusion: Our study explores ELA for discovering neuroimaging depression biomarkers.Here we have shown how ELA can capture brain state dynamics of depressive brain networks.Our future endeavors are directed towards selecting significant brain networks to differentiate depressive brains.