Receptive fields of disparity-tuned simple cells in macaque V1

Abstract

Binocular simple cells in primary visual cortex (V1) are the first cells along the mammalian visual pathway to receive input from both eyes. Two models of how binocular simple cells could extract disparity information have been put forward. The phase-shift model proposes that the receptive fields in the two eyes have different subunit organizations, while the position-shift model proposes that they have different overall locations. In five fixating macaque monkeys, we recorded from 30 disparity-tuned simple cells that showed selectivity to the disparity in a random dot stereogram. High-resolution maps of the left and right eye receptive fields indicated that both phase and position shifts were common. Single cells usually showed a combination of the two, and the optimum disparity was best correlated with the sum of receptive field phase and position shift.

Figure 1.. Position-Shift versus Phase-Shift Models for Generating Disparity-Tuned Simple Cells

(A) Adapted from Figure 3 in Qian (1997). In the position-shift model, the left and right eye receptive fields have the same phase, but different horizontal positions. In the phase-shift model, the left and right eye receptive fields are in the same position, but differ in phase. (B) The disparity-tuning curves predicted by the position-shift model (left) and phase-shift model (right) assuming that inputs from the two eyes interact multiplicatively. The position-shift model predicts symmetric disparity-tuning curves, while the phase-shift model predicts asymmetric disparity-tuning curves.

Figure 2.. Classifying Cells as Simple

(A) Top row: left and right eye spacetime maps of a disparity-tuned simple cell mapped with light bars. Bottom row: left and right eye spacetime maps of the same cell mapped with dark bars. The light- and dark-bar responses were spatially and temporally complementary. (B) Light- and dark-bar monocular spacetime maps for a cell sensitive to only a single contrast in each eye. (C) Scatter plot of temporal versus spatial phase shift for the 30 binocular simple cells (open circles, asterisks), compared to 154 disparity-tuned complex cells (open squares). For the two simple cells marked with an asterisk, phase differences were computed across eyes.

Figure 3.. A Position-Shifted Near Simple Cell

This is the same cell whose light- and dark-bar response maps are shown in Figures 2B and 2C. (A) A raster plot showing that the cell responded preferentially to a near random dot stereogram. (B) The disparity-tuning curve to a disparity bar moving in opposite directions (top), as well as to a flashed bar (bottom). For both moving and flashed bars, the left and right eye monocular responses are shown. The spontaneous firing rate is also shown for flashed bars. The peak of the flashed bar disparity-tuning curve was −0.10°. (C) First column: the receptive-field profile in the left and right eyes fitted to a Gabor function (red, left eye Gabor fit; green, right eye Gabor fit; magenta, left eye raw data; blue, right eye raw data). Second and third columns: left and right eye spacetime maps. The x axis represents the stimulus location, and the y axis represents the time following the stimulus presentation. The colorscale is calibrated symmetrically about zero; red regions represent excitation to light bars, and blue regions represent excitation to dark bars. The two horizontal lines in each map mark the 10 ms time interval over which the average receptive field profile was computed. The monocular receptive field profiles are also shown superimposed on both spacetime maps. This cell had a position shift of −0.12° and a phase shift of 0.00°. Both spacetime maps have been filtered with a pseudo-Gaussian 3 × 3 matrix with sigma = 0.14° in space and 3 ms in time. (D) The same set of maps as for (C), without filtering.

Figure 4.. The Cross-Correlogram Method Is a Simpler Way to Compute Interocular Phase and Position Shift

(A) The cross-correlation of the left and right eye spacetime maps for the cell in Figure 3, scaled to an amplitude of 2. Superimposed is a cross-correlation profile (magenta curve), averaged over the temporal range indicated by the blue horizontal lines. The Gaussian envelope of the cross-correlation profile is shown in black, and the center of the Gaussian is indicated by a black vertical line. The interocular cross-correlogram of this cell had a position of −0.15° and a phase of 0.05°. (B) The actual disparity-tuning curve (blue, raw data; cyan, Gabor fit). The predicted disparity-tuning curve (magenta) is redrawn from (A), with rescaling.

Figure 5.. Spacetime Maps and Interocular Cross-Correlation Functions for Nine Disparity-Tuned Simple Cells Reveal that Various Combinations of Phase and Position Shifts Exist between the Two Eyes’ Receptive Fields

The first column shows the monocular receptive field profiles, the second and third columns show the left and right eye spacetime maps with receptive field profiles superimposed, the fourth column shows the cross-correlation of the left and right eye spacetime maps together with the predicted disparity-tuning curve, and the fifth column shows the actual disparity-tuning curve together with the predicted disparity-tuning curve. (A) A near cell whose left and right eye receptive fields were related mainly through a position shift. The predicted disparity-tuning curve (magenta) for this cell was symmetric. (B–D) Three more near cells whose left and right eye receptive fields were related mainly through a phase shift. The predicted disparity-tuning curve (magenta) for these cells was asymmetric, and the envelope position was close to zero. (E and F) Two hybrid phase- and position-shift cells in which phase and position disparity were in the same direction (both near for [E], both far for [F]). (G) Another hybrid phase- and position-shift cell (mapped at 11.6° eccentricity), in which phase and position disparities were in opposite directions (near phase shift, far position shift). The actual disparity-tuning curve of this cell was symmetric. (H and I). Two tuned inhibitory cells. The monocular receptive fields of these cells were monophasic. The actual disparity-tuning curves showed a trough at zero disparity.

Figure 6.. Disparity Selectivity Is Generated by Both Position and Phase

(A) A scatter plot of the RF phase shift (computed from the cross-correlogram of the two eyes’ receptive field profiles) versus the phase of the actual disparity-tuning curve, both in degrees phase angle. Angular correlation coefficient r = 0.67. (B) RF position shift versus the position of the actual disparity-tuning curve. (C) RF phase shift versus the peak of the actual disparity-tuning curve. (D) RF position shift versus the peak of the actual disparity-tuning curve. (E) Sum of RF phase and position shifts versus the peak of the actual disparity-tuning curve. (F) Peak of the predicted disparity-tuning curve versus peak of the actual disparity-tuning curve. n = 29 for all plots. The cell in Figure 5G is not shown because its phase and position shifts were too large (phase shift = −0.42°, position shift = 0.73°).

Figure 7.. Distribution of RF Phase and Position Shifts

(A) Scatter plot of phase disparity versus position disparity (both in degrees visual angle), and histograms for each. The two types of disparity spanned similar ranges. There was an insignificant negative correlation between the two (r = −0.22, p < 0.1). (B) Histogram of phase in degrees phase angle. All cells have a phase angle disparity of <90°, except for the four tuned inhibitory cells.

Figure 8.. Repeatability of Predicted Phase and Position Parameters

(A) Repeated monocular spacetime maps and interocular cross-correlation maps for the cell in Figure 5A. The two sets of maps (top and bottom) were generated from two independent sets of spikes. The cross-correlation of the left and right eye maps is shown in the far right column. Note the stability of the phase and position of the predicted disparity-tuning curve. (B) Repeated spacetime maps and cross-correlation functions for the cell in Figure 5G. (C) Scatter plots of the phase (left) and envelope position (right) of the interocular cross-correlogram, computed off spikes in block A versus spikes in block B (see text for details).

Figure 9.. Da Vinci Stereopsis Generates Interocular Image Correlation Profiles with Both Phase and Position Shifts

(A) Two da Vinci stereograms. Divergent fusion of the top stereogram reveals a zero disparity square with two black flanks at a far depth. Divergent fusion of the bottom stereogram reveals a zero disparity square behind a near illusory window. Crossed fusion switches the percepts generated by the two stereograms. The diagrams on the right depict the physical situations that could give rise each stereogram. For example, stereogram 1 could be generated by viewing a white square in front of a black wall. (B) Magnified views of the left edge of stereograms 1 and 2, together with the interocular image correlation profiles computed from the magnified views. The left edge of stereogram 1 contains a crossed phase shift and an uncrossed position shift. The left edge of stereogram 2 contains an uncrossed phase shift and a crossed position shift.