DDRT: Decoding the Dynamic Interaction of Hierarchical Brain Hubs Using Reinforcement Learning in Task fMRI

Identifying pivotal hubs within brain networks and elucidating their role in the dynamic expression of functional networks is fundamental to understanding brain functions. Conventional methods are constrained by time-invariant models that overlook the dynamic reconfiguration of brain functional networks, making it difficult to disentangle transient emotional states from enduring affective traits. To address these limitations, we propose a universal brain analysis framework called DDRT and implement it on the emotion task. This framework employs a reinforcement learning methodology that integrates dynamic influence with static network topology to identify key brain regions within task-specific temporal windows, and characterize the dynamics of their functional expression. Based on the frequency and persistence of these identified regions, we uncover a three-tiered hierarchical architecture of network hubs. This hierarchy reveals structure-function consistency and features a tightly integrated core (Core/Flexible Hubs) competes with the Periphery Hubs for global influence. Our work offers a new perspective on the mechanisms of the brain.