🧠 neuroscience

Intrinsic Organization of Contrast Sensitivity in Human Vision

This paper proposes and experimentally validates a circuit-based model demonstrating that changes in luminance reorganize human contrast sensitivity by shifting the network's intrinsic spatial frequency through altered excitation-inhibition balance, rather than merely rescaling neural response amplitudes.

Original authors: Gepshtein, S., Savel'ev, S., Minns, A. A., Janson, N.

Published 2026-09-14

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Gepshtein, S., Savel'ev, S., Minns, A. A., Janson, N.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

The human eye is a marvel of adaptation, capable of seeing clearly whether one is stepping out of a dark movie theater into bright sunlight or navigating a dimly lit room at dusk. This ability relies on a system that must constantly recalibrate itself to handle an enormous range of light levels. For decades, scientists understood that as light increases, our vision becomes sharper, allowing us to distinguish finer details. This phenomenon is described by the contrast sensitivity function, a measure of how well we can detect patterns of light and dark. It was long known that in brighter light, the peak of our sensitivity shifts toward higher spatial frequencies, meaning we become better at seeing fine textures and small details, whereas in dim light, we rely more on coarser, larger patterns. However, the biological machinery behind this shift remained a mystery. Previous theories suggested that the brain simply turned up the volume on its neural signals as light increased, a process known as gain control, but this explanation could not account for why the brain would suddenly prefer to focus on a different scale of detail.

A team of researchers from Loughborough University, the University of California San Diego, and the Salk Institute has now proposed a new explanation for this reorganization of vision. They suggest that the shift is not merely a matter of amplifying signals, but rather a fundamental change in the state of the neural circuits themselves. By combining computer simulations of brain networks with precise human experiments, the researchers found that the brain's visual system operates like a resonant circuit. In this view, the network of excitatory and inhibitory neurons has a natural, preferred rhythm or scale at which it vibrates most strongly. This intrinsic scale is determined by how these neurons are connected to one another. The researchers discovered that changes in background light alter the balance between the excitatory and inhibitory forces within these circuits. This shift in balance effectively retunes the circuit, changing its natural preferred scale and causing the brain to suddenly favor a different level of visual detail.

To test this idea, the researchers first built a mathematical model of a network containing both excitatory and inhibitory neurons, similar to the connections found in the visual cortex. They simulated how this network would respond to a uniform background of light, which sets the baseline activity, and then to a patterned stimulus, such as a striped grating. The model showed that the network does not respond equally to all patterns. Instead, it has a specific spatial frequency, or a specific size of stripes, at which it responds most vigorously. This is similar to how a guitar string vibrates most strongly at a specific note. The model predicted that as the background light increased, the balance between the excitatory and inhibitory neurons would shift, causing this natural preferred frequency to jump from a low value to a higher one. The transition was not gradual; the model predicted a sharp, S-shaped change where the preferred frequency would remain steady in low light, then rise rapidly over a narrow range of brightness, before settling at a new, higher level.

The team then moved to the laboratory to see if human vision followed this same pattern. They recruited four adult volunteers with normal vision and placed them in a dark room equipped with a high-resolution monitor. The participants were asked to look at a central point while a striped pattern appeared briefly to either the left or the right. The researchers carefully measured the contrast sensitivity of each person across a wide range of light levels, from dim to bright, using a finely sampled set of brightness values. They determined the specific spatial frequency at which each person was most sensitive at every light level. The results were striking and matched the model's predictions almost perfectly. In low light, the preferred frequency remained constant. Then, as the light increased, the preferred frequency jumped sharply upward over a very narrow range of brightness, forming the predicted S-shaped curve. This rapid reorganization occurred consistently across all four subjects, suggesting that this is a fundamental property of human vision rather than a quirk of individual perception.

The study explicitly challenges the older view that the brain simply scales its responses up or down based on light levels. The researchers argue that if the brain were only adjusting the gain, the preferred frequency would not change in such a distinct, non-linear way. Instead, their findings support the idea that the brain's visual circuits are dynamic systems that can switch between different operating modes. The change in light alters the internal state of the network, effectively reconfiguring the connections so that the system resonates with a different spatial scale. This means that the brain's ability to see fine details is not just a matter of having more light to work with, but is an active process where the circuitry itself reorganizes to optimize performance for the current environment.

The researchers also compared their model to existing physiological data from other studies, including measurements of how neurons in the visual cortex respond to light and electrical stimulation. Their model successfully reproduced these known neural behaviors, such as the way inhibitory signals grow stronger with increased light and how excitatory and inhibitory responses scale differently. This consistency across different types of data strengthens the conclusion that the observed shift in visual preference is rooted in the fundamental dynamics of excitatory and inhibitory interactions. The work suggests that sensory tuning is not a fixed property of the brain's hardware but is a fluid characteristic that emerges from the current state of the network. By linking the perception of light and detail to the underlying physics of neural circuits, this research provides a deeper understanding of how the brain maintains its remarkable adaptability in a world of changing light.

Technical Summary: Intrinsic Organization of Contrast Sensitivity in Human Vision

Problem Statement
The visual system must maintain sensitivity to spatial structure across an enormous range of environmental luminance. While it is well-established that the peak of the contrast sensitivity function (CSF) shifts toward higher spatial frequencies as luminance increases, existing theoretical accounts fail to explain the mechanism behind this reorganization. Classical psychophysical models attribute this shift to efficiency constraints imposed by photon and neural noise, suggesting that low luminance favors integration over larger spatial scales. Physiological models typically frame luminance adaptation as a gain-control mechanism (e.g., divisive normalization) that rescales neural response amplitudes. However, these gain-control frameworks describe how luminance modulates response magnitude but do not explain why the preferred spatial frequency itself changes systematically or how distinct tuning regimes emerge from underlying circuit dynamics.

Methodology
The authors combine psychophysical measurements with a computational modeling framework based on recurrent excitatory–inhibitory (E–I) networks.

Modeling Approach: The study utilizes a reformulation of the Wilson–Cowan equations as nonlinear partial differential equations, treating neural tissue as a continuous medium. This approach captures wave-like behaviors (propagation, interference) arising from local and lateral interactions between excitatory and inhibitory populations. The model posits that the network possesses an "intrinsic spatial frequency" determined by its coupling structure. Background luminance ( $L_B$ ) is modeled as a constant input that shifts the network's operating state, thereby altering the effective balance between excitation and inhibition. This shift renormalizes the coupling parameters, changing the network's intrinsic spatial frequency and its resonance properties without altering the physical connectivity.
Psychophysical Experiments: Four human subjects performed a two-alternative forced-choice task to detect spatially periodic luminance gratings presented at varying locations. Contrast sensitivity functions (CSFs) were measured across a finely sampled range of mean luminance levels (starting at 10 cd/m²).
Data Analysis: Preferred spatial frequency ( $f_{pref}$ ) was defined as the frequency at which the fitted phenomenological CSF reached its maximum. To account for uncertainty, the authors employed a resampling procedure (10,000 iterations) on staircase reversal data to generate distributions of preferred frequency estimates. These empirical results were compared against two model-derived quantities: the resonance spatial frequency ( $f_r$ ), defined by the peak of the excitatory response coefficient, and the intrinsic spatial frequency ( $f_n$ ), derived from the network's Green's function in the absence of external stimulation.

Key Results

Model Predictions: The recurrent E–I network model predicts that as mean luminance increases, the baseline activity of excitatory and inhibitory populations changes non-linearly. This alters the effective E–I balance, causing the intrinsic spatial frequency of the circuit to shift. Consequently, the resonance frequency ( $f_r$ ) exhibits a characteristic S-shaped transition: it remains approximately constant at low luminance, rises sharply over a narrow luminance range, and then approaches a higher plateau.
Psychophysical Findings: Human contrast sensitivity measurements confirmed the model's predictions. Preferred spatial frequency remained constant at low luminance levels but underwent a sharp, non-monotonic increase over a restricted luminance interval, forming an S-shaped curve. This transition was reproducible across all four subjects. At very high luminance levels, the preferred frequency did not increase monotonically in all subjects, suggesting the relationship is not a simple scaling but a reorganization of spatial selectivity.
Consistency with Physiology: The framework was also fitted to existing physiological data, including LGN neuron responses, cortical hyperpolarization, and optogenetic stimulation of mouse visual cortex. The model successfully reproduced normalization-like response profiles and luminance-dependent shifts in E–I balance observed in these datasets, supporting the biological plausibility of the proposed mechanism.

Key Contributions

Mechanistic Explanation for Spatial Reorganization: The paper provides a circuit-based account for why preferred spatial frequency changes with luminance, moving beyond simple gain-control explanations. It demonstrates that luminance-dependent changes in the E–I balance shift the network's intrinsic spatial frequency, effectively moving the system between dynamical regimes.
Intrinsic Spatial Frequency: The authors identify an "intrinsic spatial frequency" as a fundamental property of recurrent cortical circuits that shapes spatial resonance. They show that this frequency is state-dependent, modulated by the operating point of the network rather than being a fixed property of receptive fields.
Unified Framework: The study links psychophysical observations of contrast sensitivity directly to the dynamics of recurrent cortical computation, suggesting that sensory feature selectivity emerges from the interaction between circuit architecture and the current dynamical state of the system.

Significance
The paper claims that these findings provide a mechanistic link between recurrent cortical dynamics and luminance-dependent spatial tuning. By demonstrating that preferred spatial frequency is not merely rescaled but reorganized via shifts in the intrinsic spatial frequency of the circuit, the work suggests that sensory selectivity is a state-dependent property of recurrent networks. This challenges the view of feature detectors as static entities and proposes that stability and selectivity arise from the same recurrent interactions that stabilize cortical activity. The results support a unified account where luminance-dependent changes in preferred spatial frequency emerge naturally from the dynamics of excitatory–inhibitory interactions, offering a new perspective on how the visual system adapts to changing environmental conditions.

More like this