Maximum entropy model reveals frequent brain state switching in psychotic disorders in a multiversal brain function analysis

Abstract

To design network-based treatments, it is critical to understand dynamic changes in brain networks over time. We comprehensively examined network structure using regional activation-only, pairwise-coactivation-only, and systems-level maximum entropy models (MEM) applied to the Human Connectome Project-Early Psychosis (HCP-EP) resting fMRI data (patients=109, controls=56). The MEM integrates regional activation levels and pairwise correlation-based functional connectivity (FC), providing the potential for better characterization of clinically meaningful brain states and how they change over time. Using the HCP/Glasser atlas to define brain regions within the default mode (DMN) and dorsal attention (DAN) networks, group differences in regional activation, graph measures of FC, and MEM features such as transition rates between minima in an energy landscape along with correlations with cognitive and psychopathological measures were examined. Psychosis was associated with reduced activation in several regions and multiple FC graph metric alterations. MEM demonstrated a wider variety of DMN and DAN activation/deactivation configurations, with more frequent switching between them, more pronounced network configuration differences and higher basin transition rates with reduced basin dwell times suggesting temporally unstable network configurations, meaning the brain cannot rely on any particular configuration for cognitive processing. Activation of only three regions correlated with working memory, but none of the FC metrics did. MEM features, in particular basin transitions and total energy, correlated negatively with working memory. Our findings suggest that MEM provides unique information, specifically attenuated inter-network connectivity with reduced stability of DMN and DAN brain states, to characterize networks as targets and their features as markers for novel treatment development.