Abstract
Introduction: Visual object categories are represented in a reliable topology within ventral temporal cortex (VTC): neural representations in lateral VTC differ from those in medial VTC. Despite this regularity, it is unknown if the two subdivisions (1) represent visual information at the same category-abstraction level and (2) if they differ in the spatial scale of their representations. We hypothesized that visual information at different category-abstraction levels is represented at different spatial scales within VTC. Methods: Seven participants completed a visual category fMRI experiment. Stimuli were drawn from ten categories (e.g., “Corridors”, “Houses”) grouped into five domains (e.g., “Places”). Data were collected at the ultra-high-resolution of 0.8mm using 7T fMRI, affording dense sampling both parallel to the cortical surface and through cortical depth (Fig S1). In each subject, we computed correlations between distributed responses to different stimuli in each VTC subdivision and each of three cortical depths (Fig S2). We evaluated the capacity of each subdivision to discriminate responses to objects from different categories and domains by computing (1) the difference between within- and between-domain correlations (domain-level discriminability), and (2) the difference between within- and between-category correlations (category-level discriminability). Results: Domain-level discriminability was stronger in lateral than medial VTC, whereas category discriminability was higher in medial than lateral VTC, (F(1, 24) = 31.0, p < .0001). Analysis by cortical depth revealed that category-level discriminability was stable across cortical depth, while domain-level discriminability decreased from the superficial to the deepest depth, (F(1, 24) = 78.5, p < 10–8). Simulated downsampling of our data to 2.4mm eliminated these effects. Conclusions: These results suggest that (1) discovering differences in category representations across VTC requires ultra-high-resolution fMRI, and (2) visual category representations in VTC occur at a variety of abstraction levels, affording downstream regions different perspectives of the visual world.
Acknowledgement: NSF GRFP, NEI 1R01EY02391501A1, NIH Grant P41 EB015894, NIH Grant P30 NS076408, NIH Grant S10 RR026783, NIH Grant S10 OD017974-01, W. M. Keck Foundation