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 been of intense interest to better understand connectome-scale brain networks. Though simultaneously modeling spatial-temporal patterns of functional brain networks has long been a research topic, the development of a unified spatial-temporal model to realize such a purpose is challenging. For instance, some deep learning methods have been proposed recently in order to model functional brain networks, most of them can only represent either spatial or temporal perspective of fMRI data and rarely model both domains simultaneously. Inspired by the success in applying sequential auto-encoders for brain encoding/decoding, we propose a novel deep sparse recurrent auto-encoder (DSRAE) in an unsupervised way to learn spatial patterns and temporal fluctuations of brain networks jointly. We evaluate and validate the proposed DSRAE on three tasks of the publicly available human connectome project (HCP) fMRI dataset with promising results. To our best knowledge, the proposed DSRAE is among the early efforts in developing unified models that can extract connectome-scale spatial-temporal networks from 4D fMRI data simultaneously.