Maximum entropy model reveals frequent brain state switching in a multiversal brain function analysis in early...

Regional activation and pairwise correlation are mainstay neuroimaging methods. Yet, neither approach fully captures brain networks' temporal evolution and interaction. The maximum entropy model (MEM), which integrates regional and inter-regional activity, represents a promising integrative approach to examine dynamically evolving functional networks.

Activation-only, pairwise-coactivation-only, and MEM-based analyses were applied to the Human Connectome Project-Early Psychosis dataset (patients=109, controls=56). Using the HCP/Glasser atlas to define brain regions from the default mode (DMN) and dorsal attention (DAN) networks, group differences were quantified in nodal frequency band power, functional connectivity (FC), and MEM features such as transition rates between basins in energy landscapes and their associations with cognitive and psychopathological measures.

Psychosis was associated with reduced BOLD frequency band power and FC graph metrics. The MEM demonstrated a variety of DMN and DAN nodal activation/deactivation patterns, with more frequent switching between them during the resting state in psychosis. Higher basin transitions with reduced basin dwell times were observed in DMN and DAN nodes, more prominently in DMN+DAN system. BOLD frequency band power of three regions correlated positively with working memory, but FC graph metrics did not. MEM metrics correlated negatively with working memory and positively with positive symptom severity.

The MEM provides unique information unavailable from the other two methods, namely nodal activation/deactivation patterns, basin transitions, and energy. Network-level results were better characterized in the MEM analysis as attenuated inter-network connectivity with reduced stability of DMN and DAN in psychosis that may underlie impaired working memory and psychopathology severity.