Transformation of Stimulus Correlations by the Retina

Redundancies and correlations in the responses of sensory neurons may seem to waste neural resources, but they can also carry cues about structured stimuli and may help the brain to correct for response errors. To investigate the effect of stimulus structure on redundancy in retina, we measured simultaneous responses from populations of retinal ganglion cells presented with natural and artificial stimuli that varied greatly in correlation structure; these stimuli and recordings are publicly available online. Responding to spatio-temporally structured stimuli such as natural movies, pairs of ganglion cells were modestly more correlated than in response to white noise checkerboards, but they were much less correlated than predicted by a non-adapting functional model of retinal response. Meanwhile, responding to stimuli with purely spatial correlations, pairs of ganglion cells showed increased correlations consistent with a static, non-adapting receptive field and nonlinearity. We found that in response to spatio-temporally correlated stimuli, ganglion cells had faster temporal kernels and tended to have stronger surrounds. These properties of individual cells, along with gain changes that opposed changes in effective contrast at the ganglion cell input, largely explained the pattern of pairwise correlations across stimuli where receptive field measurements were possible.     

Electrical coupling of photoreceptors in retinal network models

The spatial width of photoreceptor receptive fields affects the processing of signals in neural networks of the retina. This effect has been examined using the simple recurrent and non-recurrent network models, where lateral interaction strength was adjusted to approximate a prescribed receptive field profile. The results indicate that the optimal performance of the networks is obtained with photoreceptor receptive fields wider than the ganglion cell separation. It is thus concluded that while electrical coupling of photoreceptors in the retina reduces the intrinsic noise in the system, it also improves the sampling efficiency of the laterally coupled neural network of the retina.     

Predictive coding: a fresh view of inhibition in the retina

Interneurons exhibiting centre--surround antagonism within their receptive fields are commonly found in peripheral visual pathways. We propose that this organization enables the visual system to encode spatial detail in a manner that minimizes the deleterious effects of intrinsic noise, by exploiting the spatial correlation that exists within natural scenes. The antagonistic surround takes a weighted mean of the signals in neighbouring receptors to generate a statistical prediction of the signal at the centre. The predicted value is subtracted from the actual centre signal, thus minimizing the range of outputs transmitted by the centre. In this way the entire dynamic range of the interneuron can be devoted to encoding a small range of intensities, thus rendering fine detail detectable against intrinsic noise injected at later stages in processing. This predictive encoding scheme also reduces spatial redundancy, thereby enabling the array of interneurons to transmit a larger number of distinguishable images, taking into account the expected structure of the visual world. The profile of the required inhibitory field is derived from statistical estimation theory. This profile depends strongly upon the signal: noise ratio and weakly upon the extent of lateral spatial correlation. The receptive fields that are quantitatively predicted by the theory resemble those of X-type retinal ganglion cells and show that the inhibitory surround should become weaker and more diffuse at low intensities. The latter property is unequivocally demonstrated in the first-order interneurons of the fly's compound eye. The theory is extended to the time domain to account for the phasic responses of fly interneurons. These comparisons suggest that, in the early stages of processing, the visual system is concerned primarily with coding the visual image to protect against subsequent intrinsic noise, rather than with reconstructing the scene or extracting specific features from it. The treatment emphasizes that a neuron's dynamic range should be matched to both its receptive field and the statistical properties of the visual pattern expected within this field. Finally, the analysis is synthetic because it is an extension of the background suppression hypothesis (Barlow & Levick 1976), satisfies the redundancy reduction hypothesis (Barlow 1961 a, b) and is equivalent to deblurring under certain conditions (Ratliff 1965).     

Temporal filtering in retinal bipolar cells. Elements of an optimal computation?

Recent experiments indicate that the dark-adapted vertebrate visual system can count photons with a reliability limited by dark noise in the rod photoreceptors themselves. This suggests that subsequent layers of the retina, responsible for signal processing, add little if any excess noise and extract all the available information. Given the signal and noise characteristics of the photoreceptors, what is the structure of such an optimal processor? We show that optimal estimates of time-varying light intensity can be accomplished by a two-stage filter, and we suggest that the first stage should be identified with the filtering which occurs at the first anatomical stage in retinal signal processing, signal transfer from the rod photoreceptor to the bipolar cell. This leads to parameter-free predictions of the bipolar cell response, which are in excellent agreement with experiments comparing rod and bipolar cell dynamics in the same retina. As far as we know this is the first case in which the computationally significant dynamics of a neuron could be predicted rather than modeled.     

Fast, scalable, Bayesian spike identification for multi-electrode arrays

We present an algorithm to identify individual neural spikes observed on high-density multi-electrode arrays (MEAs). Our method can distinguish large numbers of distinct neural units, even when spikes overlap, and accounts for intrinsic variability of spikes from each unit. As MEAs grow larger, it is important to find spike-identification methods that are scalable, that is, the computational cost of spike fitting should scale well with the number of units observed. Our algorithm accomplishes this goal, and is fast, because it exploits the spatial locality of each unit and the basic biophysics of extracellular signal propagation. Human interaction plays a key role in our method; but effort is minimized and streamlined via a graphical interface. We illustrate our method on data from guinea pig retinal ganglion cells and document its performance on simulated data consisting of spikes added to experimentally measured background noise. We present several tests demonstrating that the algorithm is highly accurate: it exhibits low error rates on fits to synthetic data, low refractory violation rates, good receptive field coverage, and consistency across users.     

Understanding auditory spectro-temporal receptive fields and their changes with input statistics by efficient coding principles

Spectro-temporal receptive fields (STRFs) have been widely used as linear approximations to the signal transform from sound spectrograms to neural responses along the auditory pathway. Their dependence on statistical attributes of the stimuli, such as sound intensity, is usually explained by nonlinear mechanisms and models. Here, we apply an efficient coding principle which has been successfully used to understand receptive fields in early stages of visual processing, in order to provide a computational understanding of the STRFs. According to this principle, STRFs result from an optimal tradeoff between maximizing the sensory information the brain receives, and minimizing the cost of the neural activities required to represent and transmit this information. Both terms depend on the statistical properties of the sensory inputs and the noise that corrupts them. The STRFs should therefore depend on the input power spectrum and the signal-to-noise ratio, which is assumed to increase with input intensity. We analytically derive the optimal STRFs when signal and noise are approximated as Gaussians. Under the constraint that they should be spectro-temporally local, the STRFs are predicted to adapt from being band-pass to low-pass filters as the input intensity reduces, or the input correlation becomes longer range in sound frequency or time. These predictions qualitatively match physiological observations. Our prediction as to how the STRFs should be determined by the input power spectrum could readily be tested, since this spectrum depends on the stimulus ensemble. The potentials and limitations of the efficient coding principle are discussed.     

Retinal remodeling in inherited photoreceptor degenerations

Photoreceptor degenerations initiated in rods or the retinal pigmented epithelium usually evoke secondary cone death and sensory deafferentation of the surviving neural retina. In the mature central nervous system, deafferentation evokes atrophy and connective re-patterning. It has been assumed that the neural retina does not remodel, and that it is a passive survivor. Screening of advanced stages of human and rodent retinal degenerations with computational molecular phenotyping has exposed a prolonged period of aggressive negative remodeling in which neurons migrate along aberrant glial columns and seals, restructuring the adult neural retina (1). Many neurons die, but survivors rewire the remnant inner plexiform layer (IPL), forming thousands of novel ectopic microneuromas in the remnant inner nuclear layer (INL). Bipolar and amacrine cells engage in new circuits that are most likely corruptive. Remodeling in human and rodent retinas emerges regardless of the molecular defects that initially trigger retinal degenerations. Although remodeling may constrain therapeutic intervals for molecular, cellular, or bionic rescue, the exposure of intrinsic retinal remodeling by the removal of sensory control in retinal degenerations suggests that neuronal organization in the normal retina may be more plastic than previously believed.     

End-to-end neural system identification with neural information flow

Neural information flow (NIF) provides a novel approach for system identification in neuroscience. It models the neural computations in multiple brain regions and can be trained end-to-end via stochastic gradient descent from noninvasive data. NIF models represent neural information processing via a network of coupled tensors, each encoding the representation of the sensory input contained in a brain region. The elements of these tensors can be interpreted as cortical columns whose activity encodes the presence of a specific feature in a spatiotemporal location. Each tensor is coupled to the measured data specific to a brain region via low-rank observation models that can be decomposed into the spatial, temporal and feature receptive fields of a localized neuronal population. Both these observation models and the convolutional weights defining the information processing within regions are learned end-to-end by predicting the neural signal during sensory stimulation. We trained a NIF model on the activity of early visual areas using a large-scale fMRI dataset recorded in a single participant. We show that we can recover plausible visual representations and population receptive fields that are consistent with empirical findings.     

Adaptation to Changes in Higher-Order Stimulus Statistics in the Salamander Retina

Adaptation in the retina is thought to optimize the encoding of natural light signals into sequences of spikes sent to the brain. While adaptive changes in retinal processing to the variations of the mean luminance level and second-order stimulus statistics have been documented before, no such measurements have been performed when higher-order moments of the light distribution change. We therefore measured the ganglion cell responses in the tiger salamander retina to controlled changes in the second (contrast), third (skew) and fourth (kurtosis) moments of the light intensity distribution of spatially uniform temporally independent stimuli. The skew and kurtosis of the stimuli were chosen to cover the range observed in natural scenes. We quantified adaptation in ganglion cells by studying linear-nonlinear models that capture well the retinal encoding properties across all stimuli. We found that the encoding properties of retinal ganglion cells change only marginally when higher-order statistics change, compared to the changes observed in response to the variation in contrast. By analyzing optimal coding in LN-type models, we showed that neurons can maintain a high information rate without large dynamic adaptation to changes in skew or kurtosis. This is because, for uncorrelated stimuli, spatio-temporal summation within the receptive field averages away non-gaussian aspects of the light intensity distribution.     

Chicken retinal ganglion cells response characteristics: multi-channel electrode recording study

The first stage of visual processing occurs in the retina, the function of which is to process the raw information obtained from the outside world. In the present study, the electrical activities of a group of retinal ganglion cells were recorded from a small functioning piece of retina, using multi-electrode array (MEA), and the action potentials were detected by applying nonlinear algorithm. By analyzing the ensemble retinal ganglion output characteristics, it is revealed that both firing rates and correlated activity between adjacent neurons in the retina contribute to visual information encoding.     

Visual threshold is set by linear and nonlinear mechanisms in the retina that mitigate noise: how neural circuits in the retina improve the signal-to-noise ratio of the single-photon response

In sensory biology, a major outstanding question is how sensory receptor cells minimize noise while maximizing signal to set the detection threshold. This optimization could be problematic because the origin of both the signals and the limiting noise in most sensory systems is believed to lie in stimulus transduction. Signal processing in receptor cells can improve the signal-to-noise ratio. However, neural circuits can further optimize the detection threshold by pooling signals from sensory receptor cells and processing them using a combination of linear and nonlinear filtering mechanisms. In the visual system, noise limiting light detection has been assumed to arise from stimulus transduction in rod photoreceptors. In this context, the evolutionary optimization of the signal-to-noise ratio in the retina has proven critical in allowing visual sensitivity to approach the limits set by the quantal nature of light. Here, we discuss how noise in the mammalian retina is mitigated to allow for highly sensitive night vision.     

Chicken retinal ganglion cells response characteristics: multi-channel electrode recording study

A fundamental question in neuroscience is how the information relevant to behavior is presented in the activity of neurons[1]. The visual system, especially the retina, offers some advantage to explore the neural code owing to its explicitly layered structure and relatively simple neuron types[2]. However, most of what we know about retinal signaling is derived from single neuron recordings[2,3]. The assumptions underlying this approach are that individual neuron acts as a unique element dedi…     

Temporal encoding of spatial information during active visual fixation

Humans and other species continually perform microscopic eye movements, even when attending to a single point. These movements, which include drifts and microsaccades, are under oculomotor control, elicit strong neural responses, and have been thought to serve important functions. The influence of these fixational eye movements on the acquisition and neural processing of visual information remains unclear. Here, we show that during viewing of natural scenes, microscopic eye movements carry out a crucial information-processing step: they remove predictable correlations in natural scenes by equalizing the spatial power of the retinal image within the frequency range of ganglion cells' peak sensitivity. This transformation, which had been attributed to center-surround receptive field organization, occurs prior to any neural processing and reveals a form of matching between the statistics of natural images and those of normal eye movements. We further show that the combined effect of microscopic eye movements and retinal receptive field organization is to convert spatial luminance discontinuities into synchronous firing events, beginning the process of edge detection. Thus, microscopic eye movements are fundamental to two goals of early visual processing: redundancy reduction and feature extraction.     

Simple model for encoding natural images by retinal ganglion cells with nonlinear spatial integration

A central goal in sensory neuroscience is to understand the neuronal signal processing involved in the encoding of natural stimuli. A critical step towards this goal is the development of successful computational encoding models. For ganglion cells in the vertebrate retina, the development of satisfactory models for responses to natural visual scenes is an ongoing challenge. Standard models typically apply linear integration of visual stimuli over space, yet many ganglion cells are known to show nonlinear spatial integration, in particular when stimulated with contrast-reversing gratings. We here study the influence of spatial nonlinearities in the encoding of natural images by ganglion cells, using multielectrode-array recordings from isolated salamander and mouse retinas. We assess how responses to natural images depend on first- and second-order statistics of spatial patterns inside the receptive field. This leads us to a simple extension of current standard ganglion cell models. We show that taking not only the weighted average of light intensity inside the receptive field into account but also its variance over space can partly account for nonlinear integration and substantially improve response predictions of responses to novel images. For salamander ganglion cells, we find that response predictions for cell classes with large receptive fields profit most from including spatial contrast information. Finally, we demonstrate how this model framework can be used to assess the spatial scale of nonlinear integration. Our results underscore that nonlinear spatial stimulus integration translates to stimulation with natural images. Furthermore, the introduced model framework provides a simple, yet powerful extension of standard models and may serve as a benchmark for the development of more detailed models of the nonlinear structure of receptive fields.     

A possible mechanism of stochastic resonance in the light of an extra-classical receptive field model of retinal ganglion cells

Traditionally the intensity discontinuities in an image are detected as zero-crossings of the second derivative with the help of a Laplacian of Gaussian (LOG) operator that models the receptive field of retinal Ganglion cells. Such zero-crossings supposedly form a raw primal sketch edge map of the external world in the primary visual cortex of the brain. Based on a new operator which is a linear combination of the LOG and a Dirac-delta function that models the extra-classical receptive field of the ganglion cells, we find that zero-crossing points thus generated, store in presence of noise, apart from the edge information, the shading information of the image in the form of density variation of these points. We have also shown that an optimal image contrast produces best mapping of the shading information to such zero-crossing density variation for a given amount of noise contamination. Furthermore, we have observed that an optimal amount of noise contamination reproduces the minimum optimal contrast and hence gives rise to the best representation of the original image. We show that this phenomenon is similar in nature to that of stochastic resonance phenomenon observed in psychophysical experiments.     

The influence of the retinal inhomogeneity on the perception of spatial patterns

From the fact that the retina is rather inhomogeneous, it can be inferred that the perception of spatial patterns of appreciable extent will be dependent on the retinal location. Anatomical, electrophysiological and psychophysical findings substantiate the claim that the retina is very inhomogeneous of composition. In order to investigate the influence of this inhomogeneity on the perception of patterns, a model of spatiotemporal signal processing in the retina was developed on the basis of a paradigm for the Weber type adaptation. Such “scaling-ensembles” proved successful in the prediction of spatiotemporal modulation transfer in the human fovea (Koenderink et al., 1971). One prediction of the present model is that certain spatial patterns are optimally detected at well defined retinal locations, dependent on the spatial frequency content of the stimulus. A confrontation of the model's predictions with measurements published by Bryngdahl (1966) enabled us to estimate some of the relevant parameters of the retinal receptive fields as a function of the eccentricity. We obtained estimates that compare reasonably well with previously known values; for instance with values of acuity and anatomical measurements. The present discussion bears relevance on the question of whether the retina is composed of independently tuned spatial frequency filters at any retinal location, or whether the tuning is with respect to the eccentricity.     

Visual advantage in deaf adults linked to retinal changes

The altered sensory experience of profound early onset deafness provokes sometimes large scale neural reorganisations. In particular, auditory-visual cross-modal plasticity occurs, wherein redundant auditory cortex becomes recruited to vision. However, the effect of human deafness on neural structures involved in visual processing prior to the visual cortex has never been investigated, either in humans or animals. We investigated neural changes at the retina and optic nerve head in profoundly deaf (N = 14) and hearing (N = 15) adults using Optical Coherence Tomography (OCT), an in-vivo light interference method of quantifying retinal micro-structure. We compared retinal changes with behavioural results from the same deaf and hearing adults, measuring sensitivity in the peripheral visual field using Goldmann perimetry. Deaf adults had significantly larger neural rim areas, within the optic nerve head in comparison to hearing controls suggesting greater retinal ganglion cell number. Deaf adults also demonstrated significantly larger visual field areas (indicating greater peripheral sensitivity) than controls. Furthermore, neural rim area was significantly correlated with visual field area in both deaf and hearing adults. Deaf adults also showed a significantly different pattern of retinal nerve fibre layer (RNFL) distribution compared to controls. Significant correlations between the depth of the RNFL at the inferior-nasal peripapillary retina and the corresponding far temporal and superior temporal visual field areas (sensitivity) were found. Our results show that cross-modal plasticity after early onset deafness may not be limited to the sensory cortices, noting specific retinal adaptations in early onset deaf adults which are significantly correlated with peripheral vision sensitivity.