Sonia Poltoratski, Postdoc, Vision and Perception Neuroscience lab (Grill-Spector lab) Department of Psychology, Stanford University
Title: The behavioral face inversion effect is predicted by differential spatial integration in face-selective cortical regions
Abstract: Spatial information and eccentricity biases have been characterized in face-selective regions of the ventral stream. However, little is known about the precise functional consequences of this spatial processing toward recognition behavior. Here, we tested the hypothesis that spatial integration plays a critical role in a hallmark behavioral phenomenon of face perception: holistic processing. While the neural mechanisms of holistic processing are ill-defined, we posit that, at minimum, holistic processing of faces requires spatial integration across multiple facial features. Consequently, if population receptive fields (pRFs) reflect this functional spatial integration, they should be altered in response to stimuli that disrupt it, like inverted faces. To evaluate this, we mapped pRFs using 3T fMRI in 12 adults. pRFs were mapped independently in each voxel using upright and inverted faces. In face-selective regions, pRFs mapped with inverted faces had smaller size and gain, and were shifted downward relative to those mapped with upright faces. Differences in pRF properties between upright and inverted faces increased along the ventral face processing hierarchy and were also evident in pSTS-faces. In contrast, inverted and upright faces yielded equivalent pRF estimates in retinotopic areas (V1-hV4). To explain these differences, we tested additional feature-based computational models, and found that models allowing differential weighting of facial features yield more parsimonious pRF estimates across upright and inverted conditions in face-selective regions. Finally, we tested whether differences in visual field coverage by pRFs across upright and inverted conditions predict the magnitude of the face inversion effect (FIE). Behavioral testing revealed that the magnitude of the FIE was reduced when inverted faces were presented at a position that maximally overlapped with inverted pRF coverage. Together, these data suggest that pRFs in high-level regions reflect complex stimulus-dependent neural computations that underlie variations in recognition performance.