Mechanisms of Stereopsis in Monkey Visual Cortex

Stereopsis, by which certain species obtain perception of depth and distance relies in mam- mals, at least, upon cerebral cortical processing. In this article Gian Poggio describes experi- ments which have been performed to establish the existence of three types of neurones which respectively are sensitive to fine depth disparities, to stimuli nearer than the fixation distance, and to stimuli farther than the fixation distance. In the primary visual afferent pathways of monkeys, and presumably of man as well, vis~ual information from the two eyes remains essentially segregated as far as the input layer of the primary visual area of the cerebral cortex, the striate cortex, or area 17 of Brodmann. Projection fibres from the thalamic lateral geniculate nucleus, functionally associated with one or the other eye, terminate in separate and alter- nating patches in the middle layer (layer IV) of the striate cortex 5, and it is only through intracortical networks that influ- ences from the two eyes are brought to bear on neurones in the more superficial (II and III) and deeper (V and VI) cortical layers. Beyond the striate cortex, in pre- striate and inferotemporal visual areas, nearly all neurones are influenced from both eyes. The evidence, therefore, strongly suggests that binocular vision evolves from cortical neuronal mechan- isms, and that it is in the primary visual cor- tex that the search for and the analysis of interaction of signals from the two eyes should begin. Depth perception The essential function of binocular vis- ion in man is that it makes possible the unique experience of stereoscopic depth perception. The horizontal separation of the two eyes establishes a geometrical rela- tionship between objects in the external world and their images on the retina, such that the images of objects closer to or more distant from the eyes than the point of fixa- tion, occupy horizontally disparate posi- tions on the two retinas. Since Wheatstone invented the stereoscope in 1838 a wealth of psychophysical findings have demon- strated conclusively that the cue of hori- zontal binocular disparity is sufficient for the stereoscopic perception of relative position of depth, and that humans and monkeys are both able to judge correctly the depth relationships between objects whose retinal images are relatively shifted by only a few seconds of arc. A vivid illusion of depth is experienced when viewing stereograms from which normally occurring monocular depth cues such as interposition, texture changes, and size changes, have been excluded. The studies of Julesz ~ using random-dot stereograms have gone further, and have demonstrated that, using binocular vision, humans are able to detect the presence of form and movement in stereoscopic depth when each monocula r image is a homogeneous field of randomly placed elements. Only differences in the horizon- tal disparity between binocularly corre- sponding elements of figure and ground make the figure visible. Little or nothing is known about the central neural mechan- isms underlying the 'global' stereopsis required to resolve the ambiguities present in random-dot stereograms, but recent ablation studies in monkeys by Cowey in Oxford suggest that a normal infero- temporal cortex is necessary for global stereopsis, but not for the 'local' stereo- scopic mechanisms required to detect hori- zontal disparity between simple bar stimuli presented binocularly. Depth discrimination Much more is known about neural mechanisms that appear to be appropriate for 'local' stereopsis. In recent years a number of studies, first in the cat and then in the monkey, have shown that there exist in the visual cortex neurones whose responses to binocular stimulation are dependent on positional disparity. In 1967, Barlow, Blakemore, and Pettigrew ~ pre- sented evidence that the elements of a neural mechanism for binocular depth dis- crimination could be identified in the cat visual cortex based on phenomenon of receptive field disparity. These investiga- tions showed that the separation between receptive fields of binocular cortical cells in 199 the two eyes varies from cell to cell and led to the inference that some cells have fields in exact binocular correspondence, others have fields with convergent or divergent disparity. The properties of disparity-sensitive neurones in area 17 of the cat cortex have been analysed in some detail by several investigators, especially by Bishop and his collaborators in Canberra='L Evidence has been presented that these neurones have a tuned sensitivity to depth, and respond optimally to objects whose retinal images have a particular disparity, that is, to objects at some specific depth in front of or behind the point of fixation. Disparity-sensitive neurones have been described in other animals: in the visual cortex (V1 and V2) of the sheep 3 and in the Wulst of the owl 9. Studies of binocular interaction in the macaque monkey were conducted by Hubel and WieseP who did not find disparity-sensitive cells in primary visual cortex, but did observe numerous binocular neurones in prestriate area 18 with receptive fields in the two eyes at vari- ous degrees of disparity and responding exquisitely to binocular stimulation. They concluded that the chief mechanisms sub- serving stereopsis were likely to lie outside the primary visual cortex. Central representation of depth All of these neurophysiological observa- tions were made in anaesthetized animals that had been paralysed to prevent eye movements. In the past several years, tech- niques have been developed that make it possible to investigate the central neural representation of three-dimensional depth under conditions of normal binocular vision in alert and behaving macaques. Rhesus monkeys, animals known to pos- sess excellent stereopsis, can be trained to maintain a steady fixation of gaze for sev- eral seconds, and to do so repeatedly quite precisely more than 1000 times each day for a period of 4--6 h. While these monkeys fixate, it is possible to record with microclectrodes the impulse activity of single neurones in the visual cortex and to assess with considerable accuracy and resolution the response properties of these neurones to stimuli in depth. With these techniques, we '° first investi- gated the neural interaction of inputs from the two eyes on neurones in the visual cor- tex of the alert macaque. Stereoscopic response properties were examined by means of moving patterns over the 'field of view' of the cell under study, while the monkey actively maintained fixation. In