Neural Patterns Reveal Lateral Occipital Complex Representation of Ensemble Mean Orientation

Ensemble perception refers to the visual system's ability to extract summary statistics of groups of similar elements, manifested, for example, as efficient perceptual averaging. While this phenomenon has been extensively studied and characterized behaviorally, its underlying neural mechanisms remain largely unexplored. We ask if average orientation is represented in early visual cortex, known for sensitivity to local orientation, or higher cortical levels, including object- and scene-selective regions. Kahneman et al. (1992) classically proposed that object features are represented in "object files" and Ayzenberg and Behrmann (2022a) concluded that ventral representations form basis sets of local object features. Following these suggestions, we hypothesize that ensemble statistical characteristics are represented in similar "ensemble files." This similarity leads to the possibility that ensembles may be represented in object-selective regions, specifically lateral occipital complex, LOC. Using fMRI, we studied neural representations of ensemble averages derived from bars with various orientations. Participants (8 males; 17 females) estimated their average orientation and occasionally reported this in a two-alternative forced choice task. We found modest, broadly tuned representations of average orientation in early visual area V3 and parietal regions. However, the strongest contribution emerged in the LOC, which manifested distinct voxel activity patterns reliably differentiating ensemble orientation averages. Multivoxel pattern analysis pinpointed LOC in both region of interest-based and searchlight-based decoding analyses and complementary correlation analyses. Notably, LOC decoding strength was correlated with participants' performance, suggesting functional importance of LOC representation. We conjecture that object-related LOC may represent ensemble characteristics in "ensemble files" similar to previously suggested "object files."