Large-scale signal propagation modes in the human brain

The brain's temporal dynamics across large-scale networks are essential to understanding cognitive processes. However, a coherent framework remains elusive due to the combinatorial complexity of sequential states and methodological inconsistencies across studies. Here, we address these challenges by identifying a limited set of brain-wide signal propagation modes that capture diverse spatiotemporal features reported across studies. Applying dynamic mode decomposition to resting-state fMRI data, we identified five distinct propagation modes, referred to as dynamic modes (DMs), that effectively predict future BOLD signals. These modes correspond to key dimensions of neural resource allocation: (1) multisensory integration and top-down cognitive control along the cortical hierarchy; (2) salience network's modulation of interactions between the default mode and central executive networks; (3) visual-sensorimotor mapping with frontal regulation; and (4) interhemispheric coordination. Four of these modes involve the default mode network, highlighting its role as a dynamic hub. Crucially, we demonstrate that the superposition of these modes generates canonical spatiotemporal phenomena—including wave propagation, time-varying functional connectivity, and quasi-periodic patterns. These modes enable the modeling of individual differences in brain temporal dynamics within a low-dimensional manifold, with each dimension aligning with a critical operational aspect. The individual differences correlate with general cognitive abilities, exhibit heritability, and show cross-task stability. Together, our framework positions DMs as parsimonious units for modeling coherent dynamics in the brain, offering a scalable and system-level tool to study cognition and individual differences.