Multi-task fMRI outperforms resting-state fMRI for revealing task-invariant organization of the human brain

Abstract

Resting-state functional Magnetic Resonance Imaging (fMRI) is widely used to infer the intrinsic functional organization of the brain, yet it remains unclear how well this approach can predict the structure of brain activity observed across a diverse set of mental states. Here we compare resting-state to task-based fMRI using diverse task batteries within the same individuals. We find that multi-task fMRI data consistently outperform resting-state estimates in predicting functional organization during novel tasks. This advantage persists across preprocessing strategies, brain regions, and independent datasets. While task activation estimates do show task-dependency when using only few tasks, increasing task diversity reduced task-specific bias, with convergence achieved using modest task sets. These improvements translate into superior individual parcellations and connectivity models. Together, our results dissociate reliability from validity in neuroimaging and challenge the prevailing assumption that rest provides a privileged window into intrinsic brain organization. Instead, functional architecture appears most faithfully revealed when the brain is actively driven through diverse task states.