Concurrent large-scale brain dynamics during the emotional face matching task and their relation to behavior and mental health

Prior investigations of emotion processing’s neural underpinnings rely on a priori models of brain response, obscuring detection of task-relevant neurobiological processes with complex temporal dynamics. To overcome this limitation, we applied unsupervised machine learning to functional magnetic resonance imaging data acquired during the emotional face matching task (EFMT) in healthy young adults from the Human Connectome Project (n=413; n=416 replication). Tensorial independent component analysis showed that the EFMT engages 10 large-scale brain networks – each recruiting visual association cortex in distinct temporal fashions and in tandem with diverse non-visual regions – that collectively recruit 74% of cortex, posterior cerebellum, and amygdala. Despite prominent use of the EFMT to probe negative affect and related psychopathology, EFMT-recruited networks strongly reflected individual differences in cognition but not internalizing/negative affect. Overall, we characterize a richer-than-expected tapestry of concurrent EFMT-recruited brain processes, their diverse activation dynamics, and their relations to task performance and latent mental health phenotypes.