A Deep Autoencoder with Sparse and Graph Laplacian Regularization for Characterizing Dynamic Functional Connectivity During Brain Development

Deep-layer autoencoder (DAE) provides a powerful way for medical image analysis, while it remains a daunting challenge due to the limited samples but high dimension. In this paper, a DAE with sparse and graph Laplacian regularization, termed as GSDAE, is presented to identify significant differences of dynamic functional connectivity (dFC) between child and young adult groups. The proposed model incorporates prior knowledge into sparse learning, i.e., the intrinsic structural information defined by manifold in the data. In this way, the reconstruction ability of unsupervised DAE can be improved, which facilitates the extraction of most discriminative features of dFC changing with age. Results on the fMRI data from the Philadelphia Neurodevelopmental Cohort project reveal essential differences lying in the reoccurrence patterns of dFC and in the connectivity of resting state networks with increasing age, e.g., there exist different trajectories of connectivity patterns in brain functions: those associated with complex cognitive functions generally decreased, while those associated with basic visual or motor control functions usually enhanced. In addition, the brain circuitry moves from segregation to integration during brain development.