Brain network classification based on dynamic graph attention information bottleneck

Changxu Dong, Dengdi Sun

Abstract: Highlights•The proposed DGAIB framework dynamically purified and updated features of the brain graphs.•GIB is applied to optimize graph connections and GAT is used for further enhancing brain node features.•We validated the efficacy of our approach through patient-specific and cross-patient experiments.•The DGAIB demonstrated superior information representation capabilities while optimizing the upper bounds.