The neural basis of the stereo contrast paradox

Abstract

Disparity-selective neurons in the primary visual cortex (V1) of the primate brain initiate the computations that lead to the perception of depth. A long-standing challenge in visual neuroscience is to understand the contributions of different brain areas to perception. Phenomena that affect neuronal behaviour differently from perception can aid this endevour. A striking example is the stereo contrast paradox which arises when the contrast is increased in one eye while the other is fixed at a low level. At the perceptual level, the resulting interocular contrast differences increases stereothresholds relative to keeping contrast low in both eyes. Thus counterintuitively, an increase in signal strength impairs disparity sensitivity. This paradox is present for low spatial frequency narrowband stimuli (grating) and it is absent for high spatial frequency gratings and stimuli that are broadband in both frequency and orientation (e.g., random-dot patterns). We combined neuronal recordings in primate V1 with psychophysical measurements in primates and humans using random-dot and random-line stereograms (RDS, RLS). In V1, firing rate and disparity sensitivity for RLS were higher when contrast was low in both eyes compared with unequal interocular contrast, consistent with the stereo contrast paradox. For RDS, sensitivity was higher for the mismatch condition, consistent with the idea that increasing contrast increases disparity signal strength. However, psychophysically, the paradox was absent for both RDS and RLS, and for high spatial frequency gratings. In line with previous studies, the paradox was present for low spatial frequency gratings. We introduced a generalized binocular model that includes monocular and binocular contrast gain mechanisms to describe the measured cortical response modulations. This study is the first to examine the effects of unequal contrast ratios in broadband and narrowband stimuli both physiologically and psychophysically. The discrepancy between neuronal responses and behavioural performance shows that thresholds cannot be explained by a fixed linear decoder applied to V1.