Interaction Among Ocularity, Retinotopy and On-center/Off-center Pathways During Development

NeurIPS 1990

TL;DR

This paper uses a computational model to explore how the visual cortex develops its intricate structure, specifically focusing on how inputs from each eye (ocularity), spatial mapping (retinotopy), and different types of retinal cells (on-center/off-center) interact. The model successfully reproduces distinct patterns of ocular dominance columns observed in cats and monkeys, suggesting that the interplay of these factors is crucial for shaping the visual cortex.

What Problem Does This Paper Address?

The visual cortex, the part of the brain responsible for processing visual information, exhibits a highly organized structure. A key feature is the presence of ocular dominance columns (ODCs), which are regions where neurons preferentially respond to input from either the left or right eye. The patterns of these ODCs, as well as the way neurons are spatially mapped (retinotopy) and how they process visual signals (on-center/off-center pathways), differ significantly between species like cats and monkeys. For instance, cats have "beaded" ODCs and a single peak in their ocular dominance histograms (ODHs), indicating balanced input from both eyes. Monkeys, on the other hand, have more striped ODCs and a dip in their ODHs, suggesting a different pattern of binocular integration. This paper aims to understand the developmental mechanisms that lead to these species-specific differences by modeling the interactions between ocularity, retinotopy, and the on-center/off-center pathways.

Methods

The authors employ a computational approach based on a previously proposed thermodynamic formulation for the self-organization of neural networks. They develop two models:

  • Model (A): This model considers the development of ocular dominance and retinotopy.
  • Model (B): This model extends Model (A) by also incorporating the development of on-center/off-center pathway terminals.

The core of the model involves simulating the synaptic connection density of afferent fibers from the lateral geniculate nucleus (LGN) to the visual cortex using Potts spin variables. A Hamiltonian function describes the interactions between these spins, representing the effective temperature and the correlations in activity between different retinal inputs and cell types. The simulations are performed using the Metropolis algorithm.

Datasets

Not discussed in the paper. The study relies on simulated data generated by their computational models.

Key Insights and Ideas

  • Interplay of Factors is Crucial: The study highlights that the distinct visual cortex structures in cats and monkeys are not solely due to one factor but arise from the complex interactions between ocularity, retinotopy, and the specific types of retinal pathways (on-center/off-center).
  • Model (A) vs. Model (B) Differences: Model (A), which focuses on ocularity and retinotopy, successfully reproduces the striped ODCs and dip in ODH seen in monkeys. Model (B), which adds on-center/off-center pathways, generates more irregular, "beaded" ODC patterns and single-peak ODHs, aligning with observations in cats.
  • On-center/Off-center Segregation: The inclusion of on-center/off-center pathways in Model (B) leads to their segregation in the cortex, which appears to interfere with the development of ODCs and retinotopic organization, potentially explaining the differences observed in cats.
  • Receptive Field Complexity: Model (A) primarily produces simple, concentric receptive fields (RFs), while Model (B) generates more complex RF structures. This suggests that the integration of different pathway types is necessary for the development of complex RFs.
  • Cooperative Binocular Input: The authors hypothesize that cooperative binocular input is essential for the formation of normal receptive fields, as demonstrated by simulations with binocular deprivation in kittens.

Key Results

  • Model (A) Results: Simulated striped ocular dominance spatial patterns and ODHs with a dip in the binocular bin, closely matching observations in monkeys.
  • Model (B) Results: Simulated spatially modulated, irregular ODC patterns and ODHs with a single peak, closely matching observations in cats.
  • Retinotopic Order: Model (A) showed better preservation of retinotopic order compared to Model (B), where it seemed to be broken.
  • Receptive Fields: Model (A) produced simple, concentric RFs. Model (B) produced more complex RFs, and further simulations showed the emergence of simple-cell-like RFs with orientation and phase differences.

Limitations and Future Work

  • Color-Related Cell Types: The study did not consider the effect of color-related cell types on ODC formation, which could be a factor in monkey visual cortex development.
  • Sublayer Segregation: The authors suggest that for monkeys, on-center and off-center cells might need to project to different sublayers within layer 4Cβ to explain the observed ODC patterns, a detail not explicitly modeled in the current framework.
  • Future Directions: The paper implies that further investigation into the role of specific sublayer terminations and the integration of color information could refine the models and provide a more complete understanding of visual cortex development. The authors also suggest that cooperative binocular input is crucial for normal RF formation.

Summary updated

BibTeX
@inproceedings{1990interaction,
  title = {{Interaction Among Ocularity, Retinotopy and On-center/Off-center Pathways During Development}},
  booktitle = {Advances in Neural Information Processing Systems},
  year = {1990},
}
Figure