BROADband brain Network Estimation via Source Separation (BROAD-NESS)

Abstract

Auditory memory enables the recognition of sound sequences by integrating sensory input with memory traces and predictive mechanisms. While predictive coding has been proposed as a key framework for this process, the large-scale brain networks supporting it remain poorly understood. Using BROADband brain Network Estimation via Source Separation (BROAD-NESS), we provide three fundamental insights into the neural organisation of auditory memory and predictive coding. First, auditory cortices participate in two distinct, orthogonal whole-brain networks, expanding on the conventional view focused on forward and backward information flow between single brain regions. One network involves the medial cingulate gyrus, while the other integrates prefrontal and hippocampal regions, inferior temporal cortex, and insula. Second, we present evidence for a dual-stream mechanism in auditory memory recognition, which both parallels and diverges from the well-established dual-stream hypothesis in vision. Third, predictive coding in conscious auditory memory is supported by large-scale networks generating confirmed predictions and prediction errors. While previous studies examined predictive coding in isolated regions or pairwise connections, our findings reveal how whole-brain networks coordinate these processes, highlighting fine-grained spatial gradients and distinct temporal dynamics. These findings enhance our understanding of auditory perception, memory, and prediction, as well as their underlying basis in whole-brain dynamical networks.