Simultaneous Spatial-Temporal Decomposition of Connectome-Scale Brain Networks by Deep Sparse Recurrent Auto-Encoders.

Exploring the spatial patterns and temporal dynamics of human brain activities has long been a great topic, yet development of a unified spatial-temporal model for such purpose is still challenging. To better understand brain networks based on fMRI data and inspired by the success in applying deep learning for brain encoding/decoding, we propose a novel deep sparse recurrent auto-encoder (DSRAE) in an unsupervised spatial-temporal way to learn spatial and temporal patterns of brain networks jointly. The proposed DSRAE has been validated on the publicly available human connectome project (HCP) fMRI datasets with promising results. To our best knowledge, the proposed DSRAE is among the early unified models that can extract connectome-scale spatial-temporal networks from 4D fMRI data simultaneously.