一种嵌入群体先验的功能磁共振时空双流脑网络分析方法OA

Brain network analysis methods based on functional magnetic resonance imaging(fMRI)serve as a critical tool for diagnosing brain disorders.However,existing deep learning approaches predominantly focus on spatial structures while neglec-ting temporal dynamics and group-level prior information among subjects,resulting in insufficient exploration of temporal fea-tures and utilization of inter-subject similarity relationships.To address these limitations,this paper proposed an STDSM incor-porating group-level prior information.Firstly,it established a parallel spatio-temporal feature learning framework based on Graph Isomorphism Networks(GIN)and Mamba,which jointly captured the spatial and temporal characteristics of brain activity.Simultaneously,it incorporated group graph structures to embed prior similarity relationships among subjects,achie-ving representation optimization at the group level through neighborhood feature aggregation.Experiments on the public autism(ABIDE)and depression(REST-Meta-MDD)datasets demonstrate that STDSM exhibits outstanding disease prediction capa-bilities.By synergistically modeling spatio-temporal information and explicitly leveraging group priors,it effectively enhances the discriminative and generalization performance of brain disease prediction.