Motor imagery (MI) is a multidimensional cognitive ability which recruited multiple brain networks. However, how connections and interactions are adjusted among distributed networks during MI remains unknown. To investigate these issues, we analyze the reconfiguration patterns of large-scale networks for different MI states. In our work, we explored the specific patterns of large-scale functional network organization from rest to different MI tasks using group independent component analysis (ICA), and evaluated the potential relationships between MI and the patterns of large-scale networks. The results indicate that task-related large-scale networks show the balanced relation between the within- and between-network connectivities during MI, and reveal the somatomotor network and dorsal attention network play critical roles in switching context-specific MI, and also demonstrate the change of large-scale networks organization toward effective topology could facilitate MI performance. Moreover, based on the large-scale network connectivities, we could differentiate an individual's three states (i.e., left-hand MI, right-hand MI and rest) with an 72.73% accuracy using a multi-variant pattern analysis, suggesting that the specific patterns of large-scale network can also provide potential biomarkers to predict an individual's behavior. Our findings contribute to the further understanding of the neural mechanisms underlying MI from large-scale network patterns and provide new biomarkers to predict the individual's behaviors.
Keywords: ICA; Large-scale network; Machine learning; Motor imagery.