Coarse-scale biases for spirals and orientation in human visual cortex

Abstract

Multivariate decoding analyses are widely applied to functional magnetic resonance imaging (fMRI) data, but there is controversy over their interpretation. Orientation decoding in primary visual cortex (V1) reflects coarse-scale biases, including an over-representation of radial orientations. But fMRI responses to clockwise and counter-clockwise spirals can also be decoded. Because these stimuli are matched for radial orientation, while differing in local orientation, it has been argued that fine-scale columnar selectivity for orientation contributes to orientation decoding. We measured fMRI responses in human V1 to both oriented gratings and spirals. Responses to oriented gratings exhibited a complex topography, including a radial bias that was most pronounced in the peripheral representation, and a near-vertical bias that was most pronounced near the foveal representation. Responses to clockwise and counter-clockwise spirals also exhibited coarse-scale organization, at the scale of entire visual quadrants. The preference of each voxel for clockwise or counter-clockwise spirals was predicted from the preferences of that voxel for orientation and spatial position (i.e., within the retinotopic map). Our results demonstrate a bias for local stimulus orientation that has a coarse spatial scale, is robust across stimulus classes (spirals and gratings), and suffices to explain decoding from fMRI responses in V1.

Figure 1.

Orientation and spiral sense preference. A, Top row illustrates a voxel with radial orientation bias. The orientation tuning function (orange curve) is centered on the radial orientation, and the responses to clockwise and counter-clockwise spirals are equal because the local orientations of the two spirals relative to radial (θ₁ and θ₂) are equal. In the middle and bottom rows, the responses to the two spirals and differ because the orientation tuning functions are centered on orientations slightly off radial. B, Spiral sense preference for an individual voxel was predicted by combining measurements of: (1) the pRF of the voxel, (2) the responses to oriented gratings of the voxel, and (3) the local orientations of the spirals at the center of the pRF of the voxel.

Figure 2.

fMRI responses to spiral gratings and oriented gratings for an example subject (S1, same subject as shown in Fig. 3). A, Spirals alternated every 9 s between clockwise and counter-clockwise (inset). B, Oriented gratings cycled through 16 orientation steps ranging from 0° to 180° every 24 s. Responses from both experiments are shown on flattened representations (flat maps) of the occipital lobe. Dark gray, sulci. Light gray, gyri. Colors, Phase of best-fitting sinusoid (see color bars). For oriented gratings, the phase of the best-fitting sinusoid indicated the orientation preference of the voxel. For spiral gratings, the match of the phase to 0° or 180° reflected its spiral sense preference. White lines, V1/V2 boundaries as determined from an independent retinotopic mapping experiment.

Figure 3.

Measured and predicted spiral sense preferences for an example subject (S1, same subject shown in Fig. 2). A, Measured spiral sense preference. B, Predicted spiral sense preference, computed from measurements of orientation preference and population receptive field location (see also Fig. 1B). Each point corresponds to a single voxel, plotted in the visual field based on its estimated population receptive field location. Color (red or blue) indicates the preferred spiral sense (clockwise or counter-clockwise). Symbol size and saturation indicates the difference in response amplitudes (either measured or predicted) to the two spirals. Although the experimental stimulus subtended 18° (9° radius), the plot has been cropped slightly to emphasize the region of the visual field where spiral responses were strongest.

Figure 4.

Measured and predicted responses to spirals, for all three subjects (including subject S1, also shown in Figs. 1, 2). Ordinate, fMRI response amplitudes measured in the spiral experiment, during which stimuli alternated between clockwise and counter-clockwise; abscissa, predicted response amplitudes derived jointly from orientation preferences and population receptive fields. Each point corresponds to a voxel. Symbol size indicates the measured spiral sense preference (difference in response amplitudes to the two spirals).

Figure 5.

Spiral sense preferences arise from nonradial orientation preferences near the fovea. Both panels plot the orientation preference versus the polar angle component of the population receptive field. Each data point corresponds to a single voxel. Color (red or blue) indicates spiral sense preference (clockwise or counter-clockwise, respectively). Symbol size and saturation indicates the difference in response amplitudes to the two spirals. A, Voxels for which the radial component of the population receptive field is <5°. B, Voxels for which the radial component of the population receptive field is >5°.

Figure 6.

Orientation and retinotopic preferences were sufficient for spiral decoding. Black symbols, Decoding accuracy for pairs of supervoxels, created by averaging the time series from voxels that were selected based on the spiral sense prediction. Voxels sorted into two groups based on their predicted spiral sense preference for CCW or CW. Each of the two groups was subdivided into 10 bins based on the degree of predicted selectivity for CCW and CW spirals, and then averaged. Gray symbols, Voxels were assigned to bins randomly, not based on the spiral sense prediction. Error bars indicate SEM across three subjects (n = 3).

Figure 7.

Coarse-scale bias is sufficient for decoding both orientation and spiral sense. Slices were shifted by half a voxel (1 mm) on even-numbered runs. A, Measured slice position for each temporal frame of each scan, from one example scanning session. Slice position was measured using a robust image registration algorithm (Nestares and Heeger, 2000). B, Decoding accuracies when the classifier was trained and tested with data from either the same (green) or different (blue) slice positions (mean of two subjects).