Models of visual processing derived from cortical microelectrode recordings

Abstract

A single object generally activates neurons in many visual cortical areas corresponding to a distributed representation of its features. While single neurons in the “lower” cortical areas, including V1–V5, represent a variety of local features it is still under debate how the distributed representation of an object is bound into a coherent whole and how unrelated features are separated. Synchronization of neural signals has been proposed to code spatial feature binding, supported by the discovery of synchronized neural assemblies in cat and monkey visual cortex which occurred stimulus dependent — either as oscillatory (30–100 Hz) events due to internal processes of self-organization, or as non-rhythmical stimulus-locked responses. Spatial coherence of synchronized oscillations covered generally larger areas in visual cortical representations than the classical receptive field of single neurons. However, coherence was laterally confined to a few millimeters of cortical surface which means that synchronized cortical regions do only span the representational range of small visual objects or parts of larger ones. To relate such restricted segments to perceptual processes we introduced the concept of the “linking or association field” of local neural assemblies in accordance with receptive fields of single neurons. The linking field is defined by the aggregate receptive fields of a neural assembly engaged in a common synchronized state. We further argue that spatial continuity of an object might either be coded 1.) by a continuum of overlapping linking fields, i.e. by overlapping synchronized regions spanning the entire object representation or (and) 2.) by a hierarchy of linking fields in which the assemblies engage in phase locked oscillations at different frequencies — where large objects or larger parts are represented by larger linking fields defined by synchronization at lower frequencies and smaller subparts are represented by smaller linking fields at higher frequencies. Our cortical recordings showed the presence of both types of synchronized states, including phase-locking among different frequencies.

However, besides feature binding spatial and temporal feature separation play an equally important role in scene segmentation. To generalize the “binding-by-synchronization” hypothesis accordingly it is argued that the spatio-temporal distribution of short synchronized and desynchronized excitation-inhibition cycles can provide the visual system with precise internal representations of a scene's spatio-temporal feature relations. Spatial binding may be coded by a short simultaneous activation cycle (of a few milliseconds) in neurons belonging together, while spatial separation may be characterized by activation cycles occurring temporally separate. Finally, temporal separation in a single assembly may be coded by short excitations interrupted by short inhibition. Several recent recordings from the visual cortex of awake monkeys are supportive for this proposal.

The present paper also aims at relating signals of stimulus dependent synchronization and desynchronization with basic neural mechanisms and circuits, including 1.) neurons with converging feed forward sensory connections establishing classical receptive fields representing local visual features, 2.) feedback inhibition in local assemblies via a common interneuron subserving local feature linking and separation as well as suppression of incoherent stimuli, 3.) lateral and feedback associative connections for establishing fast context dependent feature linking by synchronization among assemblies driven by the same stimulus, 4.) symmetric linking circuits providing local and distant common-input at equal activation delay for supporting zero-delay relative timing. Other recently discovered effects of signal correlations potentially supporting feature grouping are shortly reviewed. Finally, our linking field hypothesis is critically discussed with respect to contradictory psychophysical work and supportive new recording results including first evidence for perception-related synchronizations in monkey visual cortex.