Spontaneously emerging patterns show category-biased similarity to task-evoked visual patterns beyond human visual cortex

In the absence of any task, spontaneous brain activity exhibits patterns that resemble those evoked by natural stimuli and behaviors. This raises the possibility that resting-state activity patterns may relate to maintenance and consolidation of behaviorally relevant information states. Using functional MRI (fMRI) data from 100 healthy adults in the Human Connectome Project, we investigated whether neural signals evoked by specific visual stimuli (body parts, faces, places, tools) show similarity to resting-state activity across non-canonical visual areas, i.e., motor and association cortices. Participants underwent a working memory task and resting-state scanning. We analyzed multivoxel pattern similarity between task-evoked activations and spontaneous activity in bilateral hand motor regions and attention-related networks. Spatial correlations between task-evoked multivoxel patterns for each visual category and frame-by-frame resting-state signals were computed, with the upper 90% distribution percentile (U90) as a similarity metric. We found that resting-state patterns in the left hemisphere hand motor region showed significantly higher similarity to tool-evoked patterns than place-evoked patterns, consistent with the interpretation that tools recruit motor affordances and hand‑object associations in motor cortex. In the bilateral dorsal attention network, resting-task similarity for non-face evoked patterns significantly exceeded that for faces. Moreover, effects localized to posterior dorsal attention network, aligning with the dorsal visual pathway's role in spatial-action coding of behaviorally relevant objects. These results suggest that resting-state activity in non-visual cortices shows category-biased similarity to task-evoked visual patterns, consistent with the view that spontaneous neural dynamics preferentially reactivate ecologically salient stimuli.