Modeling default mode network patterns via a universal spatio-temporal brain attention skip network

Hang Yuan, Xiang Li, Benzheng Wei

Abstract: Highlights•To better preserve the integrity of the spatio-temporal features of 4D Rs-fMRI, we corrected the deviation information learned by the attention mechanism. The proposed shallow attention feature enrichment module in this study enhances the capacity to concentrate on and characterize the detailed features of 4D fMRI and can provide guidance for subsequent temporal feature extraction.•We dexterously use spatiotemporal fusion information, original temporal information, and feature extracted temporal information as the input of the temporal multi-head attention block. This operation not only enriches the temporal features and reduces the risk of overfitting but also achieves feature learning constraints for spatial pattern modeling.•To our knowledge, this is one of the earliest studies to personally model the spatio-temporal feature patterns of abnormal brain DMN in EMCI patients and correlate this methodology with clinical research. This study is expected to provide a framework and approach for the characterization and construction of abnormal brain DMNs in the future.