Abstract
How does the human brain represent the behavioral significance of everyday objects? Previous research has focused on perceived similarity to compare object representations in brain and behavior. However, the similarity between objects could in principle be attributed to an unknown number of behaviorally-relevant object dimensions, and a focus on global similarity may obscure the fine-grained nature of brain-behavior correlations specific to individual dimensions. To address these challenges, here we targeted the relationship of brain activity and 49 interpretable object dimensions underlying human similarity judgments. We fit a voxel-wise encoding model on a large-scale fMRI dataset where three participants each saw 8740 unique images of 720 objects from the THINGS database. The model fit revealed highly distributed representations of object dimensions spanning across all regions of the visual system. The results capture known regional category selectivity for faces, bodies, and scenes as well as functional tuning to animacy, but also to visual features (e.g. “coarse-patterned”) and other conceptual properties (e.g. “valuable”, “disgusting”). This indicates that the behavioral significance of object dimensions is not confined to higher visual regions but instead distributed more globally. To pinpoint the tuning of individual brain regions towards behavior, we first identified regional tuning profiles across all 49 dimensions. Next, we related them to the behavioral dimension profile of thousands of individual images, including images not seen by participants. This approach identified images with strong correlations between brain and behavioral profiles, mirroring category-selective representations in higher visual regions and lower-level visual features (e.g. gratings, repeated patterns) in early visual cortex and suggesting that behaviorally-relevant object dimensions are embedded in the hierarchical representation of visual cortex. Together, these results suggest that moving beyond similarity can help identify behaviorally-relevant object representations and provide insights into the behavioral significance of different processing stages in the visual system.