Hemispheric Asymmetry of Feedback-Driven Synergistic Information Processing During Face Perception

The human brain excels at integrating complex sensory information to support coherent perception. However, the mechanisms through which distributed neural signals interact across time, space, and hierarchical levels remain incompletely understood. Here, we applied partial information decomposition to magnetoencephalography data to quantify synergistic information—a form of higher-order, non-additive integration—during the processing of normal and two-tone (Mooney) faces. We observed two temporally distinct synergy peaks corresponding to the initial stimulus processing and later memory-based task stages. Notably, Mooney faces elicited delayed and amplified information synergy, reflecting feedback-dominant processing under ambiguity. Spatial correlation analyses revealed hemispheric asymmetries: unlike the distributed additive patterns on the left, the right hemisphere showed nonlinear, emergent integration from fewer sources. At the network level, synergistic information flow was rerouted toward right-lateralized hubs, particularly under ambiguity. Moreover, feedback processing exhibited enhanced synergy information, highlighting the computational role of top-down signals. Finally, the synergy information during early perception can predict synergy information when the stimulus disappears in the right OFA, suggesting its important role in temporal integration. Together, our findings demonstrate how advanced information-theoretic approaches can reveal the underlying computational architecture of perceptual inference, capturing dynamic, directional, and sustained neural computations.