Functional fractionation of large-scale brain networks in the human subcortex.

Abstract

The human brain operates through large-scale networks whose subcortical components are critical for consciousness, emotion, and cognition. While cortical network connectivity has been mapped with increasing precision, subcortical network mapping has lagged far behind due to two fundamental barriers: the overlapping and correlated nature of brain networks, which conventional analytic methods cannot disentangle, and the inherently low signal-to-noise ratio of functional imaging data within the subcortex. As a result, no consensus or normative atlas for subcortical functional brain connectivity exists-a critical gap that has impeded both basic neuroscience and the development of targeted neuromodulatory therapies. In this work, we address these barriers using NASCAR, a tensor decomposition method explicitly designed to separate overlapping and correlated networks, applied to resting-state functional MRI data from 1,000 healthy individuals in the Human Connectome Project. This approach enabled us to fractionate four large-scale brain networks into 15 highly reproducible subnetworks spanning both cortical and subcortical structures, revealing their sites of neuroanatomic overlap and defining a normative whole-brain functional atlas grounded in subcortical connectivity. As proof of principle for the translational potential of this framework, we show that individual patterns of subnetworks predict levels of consciousness in patients with severe traumatic brain injury. By establishing a gold-standard, openly accessible reference for subcortical functional brain organization, this work expands the landscape of human brain network mapping and opens avenues for precision targeting in the treatment of a broad spectrum of neurological and psychiatric disorders.