Resolving the Spatial Profile of Figure Enhancement in Human V1 through Population Receptive Field Modeling

Abstract

The detection and segmentation of meaningful figures from their background is one of the primary functions of vision. While work in nonhuman primates has implicated early visual mechanisms in this figure-ground modulation, neuroimaging in humans has instead largely ascribed the processing of figures and objects to higher stages of the visual hierarchy. Here, we used high-field fMRI at 7 Tesla to measure BOLD responses to task-irrelevant orientation-defined figures in human early visual cortex (N = 6, four females). We used a novel population receptive field mapping-based approach to resolve the spatial profiles of two constituent mechanisms of figure-ground modulation: a local boundary response, and a further enhancement spanning the full extent of the figure region that is driven by global differences in features. Reconstructing the distinct spatial profiles of these effects reveals that figure enhancement modulates responses in human early visual cortex in a manner consistent with a mechanism of automatic, contextually driven feedback from higher visual areas.SIGNIFICANCE STATEMENT A core function of the visual system is to parse complex 2D input into meaningful figures. We do so constantly and seamlessly, both by processing information about visible edges and by analyzing large-scale differences between figure and background. While influential neurophysiology work has characterized an intriguing mechanism that enhances V1 responses to perceptual figures, we have a poor understanding of how the early visual system contributes to figure-ground processing in humans. Here, we use advanced computational analysis methods and high-field human fMRI data to resolve the distinct spatial profiles of local edge and global figure enhancement in the early visual system (V1 and LGN); the latter is distinct and consistent with a mechanism of automatic, stimulus-driven feedback from higher-level visual areas.

Keywords: LGN; early visual system; fMRI; figure–ground; pRF modeling; perception.

Copyright © 2020 the authors.

Figures

Figure 1.

Figure–ground modulation in the wild and in the literature. A, An example of animal camouflage, which is pervasive in the natural visual world. Detection of figures is much more difficult when they resemble the color, luminance, and spatial frequency content of the surrounding background. B, A sample orientation-defined figure display, typically shown to an animal such that the receptive field of the recorded neuron falls over the figure region. C, Proposed mechanisms of figure–ground modulation in the primate V1: after an initial visual response (40 ms), local boundary detection mechanisms enhance responses near the figure–surround boundary (60 ms). Subsequently (90 ms), the entire figure region is enhanced. D, The contribution of intrareal horizontal inhibition and excitatory feedback from higher-level visual areas to boundary detection (left) and figure enhancement (right). Adapted from Self et al., (2013) with permission from Elsevier.

Figure 2.

Boundary responses and figure enhancement as a function of eccentricity. A, Examples of spatially filtered oriented-noise stimulus displays used in the experiment 2. The following three main conditions are depicted: left, incongruent figure, in which the figure and surround were orthogonally oriented, creating both a perceptual border and an orientation difference; middle, congruent figure, in which the figure and surround are iso-oriented but sampled from distinct patches of noise, creating a visible phase-defined border but a weaker figure percept; right, ground-only condition, in which a single oriented texture fills the visual field. The spatial frequency depicted here is much lower than the experimental stimuli (0.5–8 cycles/°) to improve visibility. See also Movie 1. B, V1 responses plotted using mean beta weights for each experimental condition as a function of the eccentricity of the pRF center of each voxel. Dotted line at 2° eccentricity indicates the figure–surround border; error bars depict ±1 SEM across voxels.. C, V1 BOLD beta weight difference between congruent figure and ground-only conditions, associated with a predicted boundary response. D, V1 BOLD beta weight difference between incongruent and congruent figure conditions, associated with a predicted figure enhancement due to differences in orientation. The spatial profile of this effect as a function of eccentricity is clearly distinct from that of the top panel and appears to extend through the full eccentricity range of the figure but to decline beyond the 2° figure–surround border.

Figure 3.

Behavioral performance in the main experiment. A, Behavioral hit rate relative to the onset and offset of 16 s stimulus blocks, plotted in 200 ms increments. Performance is averaged across blocks; shaded error bars indicate ±SEM across subjects. There is no apparent difference between task performance when the stimulus is on the screen versus off, save for a slight dip immediately after the stimulus block that is likely attributable to increased blinking. B, Average hit rate for targets coinciding with the three main experimental condition stimulus blocks, and stimulus-off blocks. Error bars indicate ±SEM across subjects, and grayscale lines correspond to individual participants' performance (some overlap). There is no significant effect of stimulus condition on performance of the fixation task (F(3,20) = 0.46, p = 0.72), nor is there an overall difference between stimulus-on and stimulus-off performance (t(5) = 0.59, p = 0.58). While the fixation task in these experiments is not particularly challenging, we would expect some variation in performance if the task-irrelevant figures were actively or preferentially attracting subjects' attention.

Figure 4.

Details of the pRF mapping and binning procedure used in V1 analyses of Figure 2. A, Estimated pRF size (as measured by Gaussian FWHM) plotted as a function of eccentricity for all voxels used in the analyses for the main experiment. pRF size increases both as a function of eccentricity and as one ascends the visual hierarchy from V1 to V3. Inset, Total pRF coverage of V1 across subjects; blue dots mark pRF centers, and gray outlines mark FWHM. Blue circles indicate the figure location (2° eccentricity) and the full display extent (4.5° eccentricity). B, Scatterplot of voxel pRF size (Gaussian FWHM) as a function of eccentricity, with gray lines indicating bin edges. Bins were 0.25° wide with the exception of the most foveal, which included all voxels with a center pRF eccentricity <0.75°. C, Goodness-of-fit of the pRF model across voxels in each bin. Error bars indicate ±SEM across subjects. D, Number of voxels in each bin, across subjects. E, Proportion of voxels in each bin that overlap with the boundary (as defined in Fig. 5A), indicating clear differentiation of voxels in the figure, boundary, and surround regions.

Figure 5.

Boundary response and figure enhancement in voxels centered within the figure, plotted as a function of pRF extension beyond boundary. A, As illustrated, two pRFs with centers at the same eccentricity within the figure might, by virtue of differences in their size, lead to differential sensitivity to the location of the figure–surround boundary. We calculated whether the central FWHM region of each pRF fell short of the boundary (negative overlap values) or extended past the boundary (positive overlap values). B, V1 boundary responses and figure enhancement as a function of distance; while boundary responses (congruent minus ground-only condition, gray) is evident primarily in voxels whose pRFs overlap with the figure/surround boundary (overlap >0), figure enhancement (incongruent minus congruent, blue) occurs along the full extent of measured distances. Bins are 0.25° wide, and those containing <10 data points were trimmed. C, Results from a control experiment using 6° diameter figures. While boundary responses again primarily occur in voxels whose pRFs overlap with the boundary (overlap <0), figure enhancement persists even in voxels with up to 2° spatial separation with the figure–surround boundary.

Figure 6.

Average BOLD responses across figure, boundary, and surround-selective voxels. A, V1 pRF locations of one representative subject (circles, FWHM), illustrating how pRFs were sorted to create ROIs with voxels primarily responsive to the figure (orange, 205 voxels across participants), surround (purple, 293 voxels), or on the boundary (magenta, 476 voxels). Dotted lines denote the spatial extent of our mapping stimulus (9° diameter) and the figure region (4° diameter). B, V1 results for these ROIs; left panels depict average BOLD time courses in each condition, with dotted lines marking the onset and offset of 16 s stimulus blocks. Right panels depict averaged estimated GLM beta weights in each condition. Error bars indicate ±1 SEM between subjects, and asterisks indicate significance at p < .05, n.s. = no significance. C, Results from voxels in the LGN that overlapped with the center, surround, and boundary.

Figure 7.

pRF-based visualization of voxelwise BOLD responses in stimulus space. A, Schematic of the multivariate regression-based method of estimation, which allows us to estimate the spatial profile of the observed BOLD effect (yellow) from the measured pRFs (red) and BOLD response (green) of each voxel. The resulting estimate of the spatial profile (yellow) is in the form of a vector of length n2, which is then reshaped to yield images of size n × n like those in B. The precise formula used appears in the Materials and Methods. B, The projected spatial profiles of the differential responses associated with boundary response (congruent minus ground-only condition, left) and figure enhancement (incongruent minus congruent) for V1 data in the main experiment (middle) and in a control experiment which introduced a 0.5° gap between figure and surround (right). Each pixel corresponds to 0.25° of visual angle of the original stimulus space; color depicts predictor values, normalized and averaged across subject. See also Figure 9.

Figure 8.

A–C, Stimuli (A) and results (B, C) of control experiment A, in which a 0.5° grayscale gap separated the figure and surround. Plotting follows conventions of Figure 2; here, since both conditions have a visible boundary, we can subtract responses of the congruent figure condition (green) from the incongruent figure condition (red) to yield orientation-dependent figure enhancement. Results closely follow the pattern of the main experiment, with the observed enhancement extending throughout the figure region.

Figure 9.

A comparison of two methods for visualization of the boundary response (congruent minus baseline) in areas V1, V2, and V3. The top panel shows our ridge regression-based approach, while the bottom panel is a simpler average of each pRF weighted by the BOLD amplitude of boundary response-associated BOLD beta weight difference in that voxel. The regression-based approach generally led to better resolved reconstructions, particularly in higher visual areas, which typically have larger pRFs.