A spatiotemporal pRF model of sensory suppression in human visual cortex

SFN 2021 virtual poster presentation of "A spatiotemporal pRF model of sensory suppression in human visual cortex" by Eline R. Kupers, Insub Kim & Kalanit Grill-Spector.

Abstract

When multiple visual stimuli are presented simultaneously in the receptive field the neurophysiological response is surprisingly lower compared to when the identical stimuli are presented sequentially. The prevailing hypothesis suggests this suppression is due to competition effects within the receptive field (Kastner et al. 1998, 2001; Kim et al.; 2021; Reynolds, Chelazzi & Desimone, 1999). Alternatively, this suppression may be due to nonlinear, compressive summation within the receptive field—nonlinearities may be spatial (compressive spatial summation, CSS, Kay et al. 2013), temporal (compressive temporal summation, CTS, Zhou et al. 2017; Stigliani et al. 2019), or both. Here, we built a new spatiotemporal population receptive field (pRF) model and tested CSS and CTS models using fMRI. We collected data from two experiments: (i) retinotopic mapping with cartoon stimuli (Finzi et al. 2021) to define visual areas in ventral, dorsal, and lateral streams, as well as to estimate spatial pRF parameters for each voxel, and (ii) simultaneous vs sequential presented stimuli. Each trial contained 4 colorful squares (2x2°) presented in the upper right quadrant (3-9° eccentricity). In sequential trials, each square appeared for 250 ms and the 4 squares appeared in random order. In simultaneous trials, the identical 4 squares appeared at once for 250 ms followed by 750 ms blank. Each sequential and simultaneous block had 8 trials with random stimuli followed by 8 s blank. Results replicate prior findings of lower BOLD responses for simultaneous vs sequential blocks in individual voxels as early as V2 in all 3 observers. We then tested which model best predicts these suppressive effects: (i) linear temporal + spatial summation, (ii) compressive spatial + linear temporal summation (CSS), (iii) linear spatial + compressive temporal summation (CTS, 2-channel temporal model by Stigliani et al. 2019). Both nonlinear models predict lower responses for simultaneous than sequential blocks in single voxels, but the linear model predicts the same level of response across sequential and simultaneous blocks. Across the visual hierarchy, the CTS model explains most variance in the data (27±5.2%, up to ~60% in ventral and lateral areas), and significantly more than both the CSS (23±4.3%, up to ~50%) or linear model (24±4.6%, up to ~54%). These results indicate that a linear pRF model cannot explain the observed sensory suppression and the underlying mechanism involves nonlinear summation. Our new spatiotemporal pRF model offers a quantitative approach to study dynamic visual stimuli and provides insight into neural computations of sensory suppression in the human visual pathways.