An Occipitotemporal Region that Identifies Relevant Features

Poster Presentation: Tuesday, May 21, 2024, 2:45 – 6:45 pm, Pavilion
Session: Object Recognition: Basic features

Yuxuan Zeng1 (), Ren E Hentz1, David E Osher1; 1The Ohio State University, Department of Psychology

Visual stimuli are recognized first and foremost by the relevant features that differentiate them from other stimuli. How does the brain extract the features that are relevant for identifying a specific stimulus or category? While low-level features such as orientation or spatial frequency are represented as early as in V1, they still need to be integrated to form more informative features that are relevant for identifying specific object categories. Participants were trained for 1.5 hours in the lab to categorize 2 sets of novel stimuli; in the first set (featural), category membership was defined strictly by features, while in the second set (control), features were identical and category membership was defined by the locations of features. Afterwards, participants were scanned while performing the same categorization task. We also collected an fMRI localizer for high-level visual categories, including faces, bodies, scenes, and objects. We identified a bilateral region in the ventral visual stream that is specifically recruited while participants are extracting stimulus features. This region lies on the ventral surface just beyond areas V4/V8, and extends to the posterior fusiform gyrus. Interestingly, this region is not selective for any high-level visual category, and is responsive to faces, bodies, and scenes, and strongest to objects and even scrambled objects (object selectivity is typically defined by the contrast of objects > scrambled). In addition, we identified several visual fROIs, and compared featural responses in each fROI. Only the bilateral posterior fusiform sulcus (pFS, an object-selective region) showed a significant difference between conditions. Our findings reveal a region in the intermediary ventral visual stream associated with feature-based processing that does not overlap with other category-specific areas. We propose that this could be a mid-level region that extracts featural information before stimulus identity is determined.