Abstract
In most primates, neurons from nasal retina project to the brain’s contralateral hemisphere, and those from temporal retina to the ipsilateral hemisphere. If contralateral and ipsilateral projections were precisely split along the vertical meridians, objects near the midsagittal plane would send signals to opposite hemispheres, making disparity estimation difficult. But anatomical studies of macaque reveal overlapping projections near the vertical meridians (Fukuda et al., 1989). The overlap expands with eccentricity and is biased toward crossed disparities in the lower visual field (not known in the upper field). We asked whether the overlap is well suited to ensure that common disparities project to the same hemisphere thereby aiding precise stereopsis. To answer this, we must know the distribution of disparities near the vertical meridians. We measured natural retinal disparities in humans using a custom device. Participants performed everyday tasks at close, medium, and far range. The resulting database has ~880,000 video frames. The disparity distributions are peaked with long tails. The most likely disparity depends on position near the vertical meridian: uncrossed in the upper field and crossed in the lower. Variance increases with eccentricity. From those data, we calculated the percentage of disparities that would projection to opposite hemispheres if human contralateral and ipsilateral projections were the same as those in macaque. The macaque overlap would encompass 75–85% of natural disparities if it was symmetric in the two eyes. It would encompass 85–90% if it were biased toward uncrossed disparity in the upper field and crossed in the lower. Because of the long tails of the disparity distribution, wider overlap would not provide significantly better coverage. The pattern of nasal-temporal overlap in retinal-cortical projections is well suited for ensuring that common disparities produce direct projections to the same hemisphere thereby aiding precise stereopsis.
Acknowledgement: NSF Research Grant BCS-1734677, Corporate University Research, Intel Labs, and the Center for Innovation in Vision and Optics