Spatial Frequency Maps in Human Visual Cortex: A Replication and Extension

1360 

Abstract Submission 

Jiyeong Ha¹, William Broderick², Kendrick Kay³, Jonathan Winawer⁴

¹Department of Psychology, New York University, New York, NY, ²Center for Computational Neuroscience, Flatiron Institute, New York, NY, ³Center for Magnetic Resonance Research, Department of Radiology, University of Minnesota, Minneapolis, MN, ⁴Department of Psychology, Center for Neural Science, New York University, New York, NY

Jiyeong Ha  
Department of Psychology, New York University
New York, NY

William Broderick  
Center for Computational Neuroscience, Flatiron Institute
New York, NY

Kendrick Kay  
Center for Magnetic Resonance Research, Department of Radiology, University of Minnesota
Minneapolis, MN

Jonathan Winawer  
Department of Psychology, Center for Neural Science, New York University
New York, NY

Neurons in primary visual cortex (V1) of non-human primates are tuned to different spatial scales, often quantified in terms of spatial frequency. This tuning is known to vary systematically, such that preferred spatial frequency decreases with eccentricity. fMRI studies in human have shown that spatial frequency tuning can be measured at the millimeter scale, showing the expected decline in preferred frequency with eccentricity. However, there is still wide variability in estimates of peak spatial frequency tuning across studies, with measures at a given eccentricity differing by up to four-fold, except for the two recent studies by Aghajari, Vinke, & Ling (2020) and Broderick, Simoncelli, & Winawer, (2022), which show good agreement. To better understand the discrepancies, it is crucial to investigate the reproducibility of spatial frequency maps obtained in previous studies using identical stimulus classes and analysis techniques. In this study, we aimed to replicate Broderick et al.'s results using an independent dataset. Furthermore, we extended the analyses of Broderick et al, which was limited to V1, to extrastriate maps (V2 and V3).

We applied the parameterized approach of Broderick et al. (2022) to model spatial frequency preferences, which fits all the voxels in an entire visual area simultaneously with a relatively small number of parameters. This enables compact characterization of spatial frequency tuning across all of V1 as a function of 4 attributes: stimulus spatial frequency, voxel eccentricity, stimulus orientation (absolute orientation), and stimulus orientation relative to voxel polar angle preference (relative orientation). For this replication and extension, we used an unreleased extension of the Natural Scenes Dataset (NSD) (Allen et al., 2020) in which fMRI responses to a set of log-polar grating stimuli similar to those used by Broderick et al. were measured in the NSD subjects. To assess the reproducibility of spatial frequency maps, we compared the final parameters estimated from NSD and the original study.

Despite many experimental differences between Broderick et al and NSD, including field strength (3T vs 7T), number of stimulus presentations per observer (96 vs 32), and stimulus field of view (12° vs 4.2° maximal eccentricity), NSD dataset also showed good responses to scaled gratings (Fig. 1). Specifically, the data are well captured by log Gaussian tuning function and preferred spatial frequency exhibited a noticeable decrease with eccentricity. Notably, there was good agreement in most model parameters, capturing the dependency of preferred spatial frequency on voxel eccentricity. Moreover, the effect of absolute stimulus orientation on spatial frequency maps was similar: a higher spatial period for vertical and oblique orientations compared to horizontal and cardinal orientations in both studies. Lastly, we also found systematic changes with visual hierarchy. From V1 to V3, there was an increasingly large bandwidth in the voxel spatial frequency tuning functions.

Using an independent dataset and the same parametric modeling approach, our results showed good agreement on spatial frequency preference maps in V1 with the two most recent studies (Fig. 2), and some systematic changes in spatial frequency representation between V1 and extrastriate areas. Together, our findings show robust reproducibility of visual fMRI experiments, and bring us closer to a systematic characterization of spatial encoding in the human visual system.

Activation (eg. BOLD task-fMRI) ¹

Multivariate Approaches

BOLD fMRI

Perception: Visual ²

FUNCTIONAL MRI

Modeling

Vision

Other - Spatial encoding