Multivariate graph learning for detecting aberrant connectivity of dynamic brain networks in autism

Alterations in static functional brain networks have previously been reported in Autistic Spectrum Disorder (ASD). Although functional brain networks are known to be time-varying, alterations in time-varying or dynamic brain networks in ASD is largely unknown. Hence, in this study, we analyze resting-state fMRI data of ASD group versus Typically Developing Control (TDC) group to understand alterations in dynamic functional brain networks in ASD vis-à-vis healthy controls. We introduce a new framework for extracting overlapping dynamic functional brain networks to study these alterations. We utilize sliding window approach along with the recent Multivariate Vector Regression-based Connectivity (MVRC) method to construct functional connectivity (FC) matrices in each time-window. Further, we build three-mode subject-summarized spatio-temporal tensor in both ASD and TDC groups. This tensor is utilized to determine a set of overlapping dynamic functional brain networks and their temporal profiles. This helps us in studying alterations in dynamic brain networks in ASD subjects at the group-level. The proposed framework is tested on two publicly available resting-state fMRI dataset of ASD and normal controls. Our analyses on resting-state fMRI data indicate that dynamic functional brain networks of ASD subjects are altered compared to the TDC group. Two-sample t-test is used to establish the statistical significance of the differences observed in network strengths of the two groups. Compared to the TDC subjects, autistic subjects showed alterations in multiple functional brain networks including cognitive control, subcortical, auditory, visual, bilateral limbic, and default mode network. The proposed methodology is able to provide information on alterations in dynamic functional brain networks in ASD and may provide potential biomarkers for studying human brain disorders.