Modular Inter-brain Synchrony Network Associated with Social Difficulty in Autism Spectrum Disorder: a Graph Neural Network-Driven Hyperscanning Study

Understanding social difficulties in Autism Spectrum Disorder (ASD) remains challenging due to its neurobiological heterogeneity and the limited ecological validity of conventional neuroimaging methods in capturing dynamic social interactions. Hyperscanning analysis based on functional near-infrared spectroscopy (fNIRS), which measures inter-brain synchrony (IBS) during dyadic interaction, offers a novel avenue to address these challenges. However, prior studies on ASD have reported inconsistent findings, primarily focusing on intra-regional synchronization while overlooking cross-regional network dynamics. To bridge this gap, we proposed an interpretable graph neural network (GNN) model to systematically identify ASD-specific IBS modular network between child-caregiver dyads during naturalistic cooperative puzzle-solving and free-talking tasks. We identified distinctive key IBS sub-networks for the cooperative puzzle-solving task and free-talking task, with the frontal eye field (FEF) of caregivers, the dorsal lateral prefrontal cortex (DLPFC) and the motor region of children highlighted. Furthermore, the key IBS sub-networks were found to be able to predict multiple domains of the core ASD symptoms. By integrating hyperscanning with GNN-driven analysis, this work uncovers task-dependent inter-brain neural mechanisms underlying social difficulties in ASD. These findings advance the field by proposing a data-driven framework to identify IBS biomarkers tied to clinical profiles, paving the way for personalized interventions that integrate computational neuroscience with clinical practice.