Activation network improves spatiotemporal modelling of human brain communication processes

Dynamic functional networks (DFN) have considerably advanced modelling of the brain communication processes. The prevailing implementation capitalizes on the system and network-level correlations between time-series. However, this approach does not account for the continuous impact of non-dynamic dependencies within the statistical correlation, resulting in relatively stable connectivity patterns of DFN over time with limited sensitivity for communication dynamic between brain regions. Here, we propose an activation network framework based on the activity of functional connectivity (AFC) to extract new types of connectivity patterns during brain communication process. The AFC captures potential time-specific fluctuations associated with the brain communication processes by eliminating the non-dynamic dependency of the statistical correlation. In a simulation study, the positive correlation (r=0.966,p<0.001) between the extracted dynamic dependencies and the simulated "ground truth" validates the method's dynamic detection capability. In an application to autism spectrum disorders (ASD) and COVID-19 datasets, the proposed activation network extracts richer topological reorganization information, which are largely invisible to the DFN. Detailed, the activation network exhibits significant inter-regional connections between function-specific subnetworks and reconfigures more efficiently in the temporal dimension. Furthermore, the DFN fails to distinguish between patients and healthy controls. However, the proposed method reveals a significant decrease (p<0.05) in brain information processing abilities in patients. Finally, combining two types of networks successfully classify ASD (83.636% ± 11.969%,mean±std) and COVID-19 (67.333% ± 5.398%). These findings suggest the proposed method could be a potential analytic framework for elucidating the neural mechanism of brain dynamics.