Lateral interactions in visual cortex.

Abstract

The findings presented in these studies have brought out different ideas concerning the mechanisms of processing in primary visual cortex than were held at the outset. Rather than thinking of receptive fields as being restricted in their extent, with the process of integration of the components of an image occurring at a much later stage along the visual pathway, we have shown that the integrative process is a progressive one, beginning in the primary visual cortex (or perhaps even earlier) and building up in a cascading series of converging and diverging connections. Rather than thinking of the filter characteristics of a cell as being fixed, it is apparent that they are dynamic and can be modified by the context in which features are presented. Finally, rather than a cortex with a functional architecture that is fixed after a critical period ending in infancy, we find that perturbing the system can lead to long-term topographical reorganization. Other examples of contextual interactions have been demonstrated in the submodalities of motion, where a cell's directional selectivity is modulated by the presence of movement in the surround (Allman et al. 1985; Tanaka et al. 1986; Gulyas et al. 1987; Orban et al. 1987). In the domain of color, the phenomenon of color constancy, reported for cells in visual area V4 (Zeki 1983), also requires lateral interactions in visual space, comparing the wavelength distribution of light coming from surfaces in different parts of the visual field. The influences presented in these studies, as in our own work in the domain of orientation, are modulatory. The long-term changes in cortical topography following removal of somatosensory input (Merzenich et al. 1984, 1988) or by retinal lesions suggest that with the appropriate manipulations the lateral interactions can be enhanced to the point of activating the postsynaptic cells. Although retinal lesions clearly represent an abnormal disruption of sensory input, they may nevertheless be representative of long-term reorganizations of neural networks occurring under normal circumstances, such as those required for memory.