Forecasting Perception Before It Happens: Context-Specific Connectivity Patterns Predict Perceptual Outcomes

The behavioral relevance of the static functional connectome is well-established, but the real-time relationship between ongoing connectome dynamics and behavior remains unclear. Because behavior shifts from moment to moment, an important question arises: Are changes in behavioral outcomes linked to fluctuations in task-specific or task-general processes like arousal? Task-specific versus task-general processes should manifest as distinct versus shared connectome patterns that predict moment-to-moment behavioral outcomes across tasks. We analyzed fMRI data from three groups of healthy participants (total N=35), each performing one of three ambiguous perception tasks maximizing behavioral fluctuations: (1) Reporting faces or vase percepts on Rubin's figure, (2) detecting near-random motion, or (3) identifying a near-threshold tone. Leveraging long inter-stimulus intervals (>20s), we examined how pre-stimulus connectome states influenced perception on a trial-by-trial basis. Using SVM models, pre-stimulus connectome reliably predicted upcoming post-stimulus perception. Distinct, non-overlapping sets of task-specific connections supported prediction in each task, spanning task-relevant sensory networks as well as higher-order cognitive control networks. Only when predictive connections were aggregated over the well-known intrinsic connectivity networks (ICNs), did we observe partial overlap across tasks. Our findings underscore the functional impact of ongoing connectome dynamics in shaping moment-to-moment behavior, and show that this impact is predominantly context-dependent.