Eccentricity strongly modulates visual processing delays

Abstract

The temporal dynamics of visual information processing varies with the stimulus being processed and with the retinal location that initiates the processing. Here, we present psychophysical data with sub-millisecond precision showing that increasing eccentricity decreases the delay with which stimuli are processed. We show that, even within the central +/-6° of the visual field, processing delays change by a factor of up to three times. A simple model, grounded in retinal physiology, provides a good account of the data. The relative delays are on the order of only milliseconds. But if later processing leaves the delays unresolved, they can cause dramatic misperceptions of motion and 3D layout. We discuss the implications for how the human visual system solves the temporal binding problem across eccentricity. The results highlight the severe computational challenge of obtaining accurate, temporally-unified percepts of the environment with spatiotemporally-staggered processing across the visual field.

Keywords: 3D orientation; binocular disparity; perceptual latency; periphery; retina; temporal sensitivity.

Figure 1.. Processing speed, stimulus, and task.

A Stimulus properties vary with location in natural scenes (top). Dark stimulus patches (#1) are processed more slowly than bright stimulus patches (#2) when fixated (middle). It is largely unknown how visual field location impacts processing speed (bottom). B Left- and right-eye onscreen stimuli. Free-fuse to see the stimuli in depth. Divergent-fusing will produce a ‘top-back’ percept. Cross-fusing will produce a ‘bottom-back’ percept. C If the left eye receives less light than the right eye, the left-eye image will be processed with more delay. With moving stimuli, a neural disparity will occur, and the beads will be perceived as rotated ‘top back’ with respect to the screen. The effective neural disparity at a given bead location Δx is given by v_xΔt the product of the horizontal velocity component of the bead motion, and the interocular neural delay assuming that onscreen delays are zero. The critical onscreen delay was that which nulled the neural delay, and made the beads appear to rotate in the plane of the screen. D Critical onscreen delays were measured for a five different eccentricities (0.5°, 1.0°, 2.0°, 4.0°, 6.0°). These critical onscreen delays should be equal in size but have the opposite sign of the neural delay. Circular bead sizes (10, 20, 40, 80, 120arcmin) and rotational speeds (1, 2, 4, 8, 12deg/sec) scaled exactly with eccentricity.

Figure 2.. Main experiment.

A Psychometric functions for the first human subject. Proportion ‘top back’ chosen as a function of onscreen delay for each luminance difference between the eyes (colors) and each eccentricity (subplots) for subject S1. Luminance differences are expressed as interocular differences in optical density (ΔO=[−0.6,−0.3,0.3,0.6], colors; see Methods). Points of subjective equalities (PSEs; arrows) indicate the critical onscreen delays required to make the stimulus appear to rotate in the plane of the screen. B Critical onscreen delay as a function of eccentricity for the four interocular luminance differences. Delays decrease systematically with eccentricity. Shaded regions indicate bootstrapped 68% confidence intervals (approximately ±1 standard error). In cases where shaded regions are not visible, it is because the confidence interval is smaller than the datapoint. C As in B, but for averaged data across subjects. Shaded regions indicate standard deviations across subjects. D Interocular delays as a function of interocular difference in optical density, for five different eccentricities (replotted data in A). Best-fit regression lines, computed via weighted linear regression, are also shown. The slopes of the best-fit lines decrease systematically with eccentricity, again indicating that the same luminance difference causes smaller interocular delays as eccentricity increases. E As in D, for group averaged data. F Best-fit slopes (see D and E) at each eccentricity for all individual subjects and the average data (symbols and solid lines). The best-fit power function—computed from the average data (stars)—is also shown (dashed black line; see Methods). For the group average data, the delays caused by a given luminance difference decrease with eccentricity raised to a power of −0.33. (Individual subject best-fit powers, in order, are $m$ = [−0.34, 0.34, −0.21, −0.34, −0.40].) Hence, 8-fold increases in eccentricity are associated with 2-fold decreases in visual processing delay.

Figure 3.. Size-control experiment.

A In the original experiment, bead size increased in proportion to eccentricity. In the control experiment, bead size was held constant at all eccentricities. B Critical onscreen delays as a function of eccentricity at each luminance difference (colors) in both the control experiment (left) and the original experiment (right) for the first human subject. C Delays from the size-control experiment plotted against delays from the original experiment, after having regressed out average delays due to luminance differences (dashed lines in B). The eccentricity-dependent changes in critical onscreen delay were significantly correlated between the two experiments ($\rho =0.97,p=5.6\times {10}^{-10}$). The slope a of best-fit regression line, obtained using Deming regression, was not significantly different from 1.0 $(a=0.91,C{l}^{95}=[0.781.07]$). Shaded regions show 95% confidence intervals on the best-fit slopes obtained from 1000 bootstrapped datasets. D As in B but with group averaged data. Shaded regions indicate the standard deviation across subjects in each condition. E As in C, but for group averaged data ($\rho =0.64,p=0.008;a=0.80,{Cl}^{95}=\left[0.311.30\right]$). The individual subject data, in order, are best-fit by $a=[0.91,1.10,1.49,0.96,3.25]$ (see Supplement Fig. S2D). The evidence indicates that bead-size has no significant effect on delay. F Predicted 3D percepts of a binocularly viewed spinning object (true), assuming interocular delays that either are constant with (dashed) or vary with (solid) eccentricity. All predictions assume that no computations subsequent to the delays function to eliminate them.

Figure 4.. Retinal physiology-based model predictions.

A Changes in retinal physiology with eccentricity, expressed as proportional change relative to the fovea. Cone inner-segment (IS) diameters increase (blue), cone outer-segment (OS) lengths decrease (red), and macular pigment transmittance (yellow) increases with eccentricity. Insets (adapted from Curcio et al., 1990) show peripheral and foveal cone IS diameters (top) and cone OS lengths (bottom), with individual examples outlined in white. B Photon catch rate, expressed as proportional change relative to the fovea, based on differences shown in A, for the L-, M-, and S-cones. The cone-specific catch rates differ because of how the cone spectral sensitivities interact with the spectral transmittance of the macular pigment, which passes less light in the short- than in the long-wavelength portion of the spectrum. Also shown are catch rates for the inner-segments only and for the midget ganglion cells (mRGCs). C Visualization showing how interocular delays $(\mathrm{\Delta }t)$ are predicted from catch rates (slopes) for a 2-fold difference in light intensity $(I)$) between the eyes. Higher catch rates in the periphery (see B) predict smaller interocular delays than in the fovea. D Model fit (solid curves) to group averaged data (stars) for all eccentricities and luminance differences. The model uses the mean catch rate of L- and M-cone catch rate (i.e. [L+M]/2). A single free-parameter—the threshold number of photons—was used to fit the data across all conditions. Changing the threshold value multiplicatively scales the model predictions symmetrically around zero delay. E Model fits (solid curves) and group averaged data replotted as in Fig. 2F. The [L+M]/2 model fit is the solid curve. Best-fit model predictions from S-cones (blue curve), inner-segment diameters alone (dashed curve), and midget ganglion cells (dotted curve) are also shown.