Comparison of different models of orientation selectivity based on distinct intracortical inhibition rules

A Model of the Early Visual System Based on Parallel Spike-Sequence Detection, Showing Orientation Selectivity

Simple Summary

A computational model of primates’ early visual processing, showing orientation selectivity, is presented. The system importantly integrates two key elements: (1) a neuromorphic spike-decoding structure that considerably resembles the circuitry between layers IV and II/III of the primary visual cortex, both in topology and operation; (2) the plasticity of intrinsic excitability, to embed recent findings about the operation of the same area. The model is proposed as a tool for the analysis and reproduction of the orientation selectivity phenomenon, whose underlying neuronal-level computational mechanisms are today the subject of intense scrutiny. In response to rotated Gabor patches the model is able to exhibit realistic orientation tuning curves and to reproduce responses similar to those found in neurophysiological recordings from the primary visual cortex obtained under the same task, considering different stages of the network. This demonstrates its aptness to capture the mechanisms underlying the evoked response in the primary visual cortex. Our tool is available online, and can be expanded to other experiments using a dedicated software library developed by the authors, to elucidate the computational mechanisms underlying orientation selectivity.

Abstract

Since the first half of the twentieth century, numerous studies have been conducted on how the visual cortex encodes basic image features. One of the hallmarks of basic feature extraction is the phenomenon of orientation selectivity, of which the underlying neuronal-level computational mechanisms remain partially unclear despite being intensively investigated. In this work we present a reduced visual system model (RVSM) of the first level of scene analysis, involving the retina, the lateral geniculate nucleus and the primary visual cortex (V1), showing orientation selectivity. The detection core of the RVSM is the neuromorphic spike-decoding structure MNSD, which is able to learn and recognize parallel spike sequences and considerably resembles the neuronal microcircuits of V1 in both topology and operation. This structure is equipped with plasticity of intrinsic excitability to embed recent findings about V1 operation. The RVSM, which embeds 81 groups of MNSD arranged in 4 oriented columns, is tested using sets of rotated Gabor patches as input. Finally, synthetic visual evoked activity generated by the RVSM is compared with real neurophysiological signal from V1 area: (1) postsynaptic activity of human subjects obtained by magnetoencephalography and (2) spiking activity of macaques obtained by multi-tetrode arrays. The system is implemented using the NEST simulator. The results attest to a good level of resemblance between the model response and real neurophysiological recordings. As the RVSM is available online, and the model parameters can be customized by the user, we propose it as a tool to elucidate the computational mechanisms underlying orientation selectivity.

Power-law input-output transfer functions explain the contrast-response and tuning properties of neurons in visual cortex

We develop a unified model accounting simultaneously for the contrast invariance of the width of the orientation tuning curves (OT) and for the sigmoidal shape of the contrast response function (CRF) of neurons in the primary visual cortex (V1). We determine analytically the conditions for the structure of the afferent LGN and recurrent V1 inputs that lead to these properties for a hypercolumn composed of rate based neurons with a power-law transfer function. We investigate what are the relative contributions of single neuron and network properties in shaping the OT and the CRF. We test these results with numerical simulations of a network of conductance-based model (CBM) neurons and we demonstrate that they are valid and more robust here than in the rate model. The results indicate that because of the acceleration in the transfer function, described here by a power-law, the orientation tuning curves of V1 neurons are more tuned, and their CRF is steeper than those of their inputs. Last, we show that it is possible to account for the diversity in the measured CRFs by introducing heterogeneities either in single neuron properties or in the input to the neurons. We show how correlations among the parameters that characterize the CRF depend on these sources of heterogeneities. Comparison with experimental data suggests that both sources contribute nearly equally to the diversity of CRF shapes observed in V1 neurons.

Both the response and membrane potential of neurons in the primary visual cortex (V1) are selective to the orientation of elongated stimuli. The widths of the tuning curves, which characterize this selectivity, hardly depend on stimulus contrast whereas their amplitude does. The contrast dependence of this amplitude, the contrast response function (CRF), has a sigmoidal shape. Saturation of the spike response is substantially lower than the neurons' maximal firing rate. These well established facts constrain the possible mechanisms for orientation selectivity in V1. Furthermore, the single neuron CRFs in V1 display a broad diversity in their shape. This adds other constraints. Many theoretical works have tried to elaborate mechanisms of orientation selectivity that are compatible with the contrast invariant tuning widths. However, these mechanisms are usually incompatible with sigmoidal CRFs. We propose a mechanism which accounts simultaneously for contrast invariant tuning width for both rate and voltage response and for the shape and diversity of the CRFs. This mechanism relies on the interplay between power-law frequency-current transfer functions of single neurons, as measured in vivo in cortex, and on the recurrent interactions in the cortical circuit.

Contrast adaptation contributes to contrast-invariance of orientation tuning of primate V1 cells

Background

Studies in rodents and carnivores have shown that orientation tuning width of single neurons does not change when stimulus contrast is modified. However, in these studies, stimuli were presented for a relatively long duration (e. g., 4 seconds), making it possible that contrast adaptation contributed to contrast-invariance of orientation tuning. Our first purpose was to determine, in marmoset area V1, whether orientation tuning is still contrast-invariant with the stimulation duration is comparable to that of a visual fixation.

Methodology/Principal Findings

We performed extracellular recordings and examined orientation tuning of single-units using static sine-wave gratings that were flashed for 200 msec. Sixteen orientations and three contrast levels, representing low, medium and high values in the range of effective contrasts for each neuron, were randomly intermixed. Contrast adaptation being a slow phenomenon, cells did not have enough time to adapt to each contrast individually. With this stimulation protocol, we found that the tuning width obtained at intermediate contrast was reduced to 89% (median), and that at low contrast to 76%, of that obtained at high contrast. Therefore, when probed with briefly flashed stimuli, orientation tuning is not contrast-invariant in marmoset V1. Our second purpose was to determine whether contrast adaptation contributes to contrast-invariance of orientation tuning. Stationary gratings were presented, as previously, for 200 msec with randomly varying orientations, but the contrast was kept constant within stimulation blocks lasting >20 sec, allowing for adaptation to the single contrast in use. In these conditions, tuning widths obtained at low contrast were still significantly less than at high contrast (median 85%). However, tuning widths obtained with medium and high contrast stimuli no longer differed significantly.

Conclusions/Significance

Orientation tuning does not appear to be contrast-invariant when briefly flashed stimuli vary in both contrast and orientation, but contrast adaptation partially restores contrast-invariance of orientation tuning.

Lack of orientation and direction selectivity in a subgroup of fast-spiking inhibitory interneurons: cellular and synaptic mechanisms and comparison with other electrophysiological cell types

Neurons in cat area 17 can be grouped in 4 different electrophysiological cell classes (regular spiking, intrinsically bursting, chattering, and fast spiking [FS]). However, little is known of the functional properties of these different cell classes. Here we compared orientation and direction selectivity between these cell classes in cat area 17 and found that a subset of FS inhibitory neurons, usually with complex receptive fields, exhibited little selectivity in comparison with other cell types. Differences in occurrence and amplitude of gamma-range membrane fluctuations, as well as in numbers of action potentials in response to optimal visual stimuli, did not parallel differences observed for orientation and direction selectivity. Instead, differences in selectivity resulted mostly from differences in tuning of the membrane potential responses, although variations in spike threshold also contributed: weakly selective FS neurons exhibited both a lower spike threshold and more broadly tuned membrane potential responses in comparison with the other cell classes. Our results are consistent with the hypothesis that a subgroup of FS neurons receives connections and possesses intrinsic properties allowing the generation of weakly selective responses. The existence of weakly selective inhibitory neurons is consistent with orientation selectivity models that rely on broadly tuned inhibition.

Direction selectivity of simple cells in the primary visual cortex: comparison of two alternative mathematical models. I: response to drifting gratings

Two models of a single hypercolumn in the primary visual cortex are presented, and used for the analysis of direction selectivity in simple cells. The two models differ as to the arrangement of inhibitory connections: in the first ("antiphase model") inhibition is in phase opposition with excitation, but with a similar orientation tuning; in the second ("in-phase model"), inhibition is in phase with excitation, but with broader orientation tuning. Simulation results, performed by using drifting gratings with different orientations, and different spatial and temporal frequencies, show that both models are able to explain the origin of direction preference of simple cells.

Dependence of Visual Cell Properties on Intracortical Synapses Among Hypercolumns: Analysis by a Computer Model

The role of intracortical synapses in affecting the property of visual cells is investigated by means of an original mathematical model of cortical circuitry in V1. The model represents a compromise between computational simplicity and physiological reliability. The model incorporates four different inputs into a cortical cell: thalamic input from the lateral geniculate nucleus, according to an even Gabor function; short-range inhibition confined within the hypercolumn; a long-range excitation, which emphasizes the properties of the input; and a long-range inhibition. In the model we assume that all cells receive a similar thalamic input, which differs simply according to its position in the retina and orientation preference. Simulations were performed, with different parameter values, to assess the main characteristics of cell response (i.e., the width and locations of subregions in the receptive field (RF), orientation tuning curve, and response to drifting and counterphase gratings) as a function of the strength and extension of intracortical excitatory synapses. Results suggest that, if intracortical excitation is confined within the hypercolumn, the cells exhibit the same properties as simple cells, both with regards to the width and shape of the RF, orientation tuning curve, and response to drifting and counterphase gratings. By contrast, if excitatory synapses extend beyond the hypercolumn with sufficient strength, the cells exhibit the typical characteristics of complex cells. A progressive shift from complex to simple cells can be realized with a monotonic variation in parameters. Simulations are also performed with a hierarchical model, to suggest possible experiments able to discriminate the present recurrent mechanism from the classical hierarchical one. Results support the assumptions of previous simpler models (Chance et al., 1999) and may help to understand and assess the role of intracortical synapses in rigorous quantitative terms.

A model of contextual interactions and contour detection in primary visual cortex

A new model of contour extraction and perceptual grouping in the primary visual cortex is presented and discussed. It differs from previous models since it incorporates four main mechanisms, according to recent physiological data: a feed-forward input from the lateral geniculate nucleus, characterized by Gabor elongated receptive fields; an inhibitory feed-forward input, maximally oriented in the orthogonal direction of the target cell, which suppresses non-optimal stimuli and warrants contrast invariance; an excitatory cortical feedback, which respects co-axial and co-modularity criteria; and a long-range isotropic feedback inhibition. Model behavior has been tested on artificial images with contours of different curvatures, in the presence of considerable noise or in the presence of broken contours, and on a few real images. A sensitivity analysis has also been performed on the role of intracortical synapses. Results show that the model can extract correct contours within acceptable time from image presentation (30-40 ms). The feed-forward input plays a major role to set an initial correct bias for the subsequent feedback and to ensure contrast-invariance. Long-range inhibition is essential to suppress noise, but it may suppress small contours due to excessive competition with greater contours. Cortical excitation sharpens the initial bias and improves saliency of the contours. Model results support the idea that contour extraction is one the primary steps in the visual processing stream, and that local processing in V1 is able to solve this task even in difficult conditions, without the participation of higher visual centers.