Development of Maps of Simple and Complex Cells in the Primary Visual Cortex

Ján Antolík, James A. BednarFrontiers Comput. Neurosci 2011 · DOI · OpenReview

In a study published in Frontiers in Computational Neuroscience, researchers propose a biologically plausible model for the self-organization of the primary visual cortex (V1). By shifting the developmental "engine" from the simple-cell layer to the complex-cell layer, the authors resolve a long-standing paradox: how simple cells with diverse phase preferences can coexist within smooth, globally organized orientation maps.

The Paradox of V1 Organization

In the visual cortex, neurons are categorized into two main types: "simple cells" and "complex cells." Simple cells respond to specific orientations at precise spatial positions (phases). Complex cells also respond to specific orientations but are "phase-invariant," meaning they fire regardless of the stimulus's exact position within their receptive field.

To construct a complex cell, the brain must pool inputs from nearby simple cells that share the same orientation preference but possess different phase preferences. However, standard computational models of cortical development often rely on "Mexican-hat" lateral interactions. These interactions force neighboring neurons to become highly correlated. If nearby simple cells are forced to share the same phase, they cannot be pooled effectively to create a phase-invariant complex cell. Conversely, if phase preferences are forced to be disordered to allow for complex cell construction, standard models typically fail to produce the smooth, large-scale topographic orientation maps observed in biological imaging.

Schematic of pooling mechanisms from simple cells to complex cells.

A Two-Layer Solution

The authors address this conflict by introducing a two-layer cortical model that more accurately reflects primate anatomy. Most existing models place the strongest lateral connections in the simple-cell layer (Layer 4). However, anatomical evidence suggests these connections are actually strongest in the complex-cell layer (Layer 2/3).

The proposed model utilizes this hierarchy:

  • Layer 4 (Simple Cells): Receives direct input from the lateral geniculate nucleus (LGN) and has weak lateral connections.
  • Layer 2/3 (Complex Cells): Receives feed-forward input from Layer 4 and possesses strong, long-range lateral inhibitory connections.
  • Feedback Mechanism: Crucially, the model includes feedback from Layer 2/3 back to Layer 4.

Because Layer 4 has weak internal lateral connections, it does not face the "pressure" to make neighboring cells identical in phase. This allows for the local phase diversity necessary for complex cell development. However, the feedback from the highly organized Layer 2/3 forces Layer 4 to align its orientation preferences with the global map, ensuring large-scale order.

Methodology and Simulation

The researchers used the Topographica simulator to implement this architecture. The model employs unsupervised Hebbian learning with divisive normalization. Development was simulated in two distinct stages:

  1. Pre-natal: Driven by "retinal waves"—spatially correlated spontaneous activity modeled as expanding rings.
  2. Post-natal: Driven by natural image patches extracted from movies of natural environments.

A key technical component was the use of homeostatic plasticity to adjust neuron thresholds. This ensured stable activity levels as the model transitioned from spontaneous pre-natal activity to complex natural stimuli.

Key Results and Biological Alignment

The model successfully produced several key features of the primary visual cortex:

  • Functional Maps: It developed smooth orientation preference maps featuring biological hallmarks such as "pinwheels" and "linear zones" in both layers.
  • Cell Dichotomy: Units in Layer 4 emerged as simple cells with high phase selectivity, while Layer 2/3 units became phase-invariant complex cells.
  • Statistical Match: The distribution of the Modulation Ratio (MR)—a metric for phase sensitivity—showed a bimodal peak that closely matched experimental data from macaque V1.

New Predictions and Future Work

The model offers several testable predictions for future neurobiological research. It suggests that complex cells may possess a weak, clustered phase preference detectable with high-resolution imaging. Furthermore, it predicts that neurons located near orientation map "singularities" (pinwheels) should exhibit higher phase selectivity than those in the center of orientation domains.

While the model is a significant step forward, limitations remain. The computational cost of the simulations (48 hours per run in 2011) limited parameter exploration. Future iterations could increase neuron density to further improve phase invariance and incorporate multiple spatial frequency channels to model the development of color and motion preferences.

Summary updated

Abstract

p>Hubel and Wiesel (<xref ref-type="bibr" rid="B25">1962</xref>) classified primary visual cortex (V1) neurons as either simple, with responses modulated by the spatial phase of a sine grating, or complex, i.e., largely phase invariant. Much progress has been made in understanding how simple-cells develop, and there are now detailed computational models establishing how they can form topographic maps ordered by orientation preference. There are also models of how complex cells can develop using outputs from simple cells with different phase preferences, but no model of how a topographic orientation map of complex cells could be formed based on the actual connectivity patterns found in V1. Addressing this question is important, because the majority of existing developmental models of simple-cell maps group neurons selective to similar spatial phases together, which is contrary to experimental evidence, and makes it difficult to construct complex cells. Overcoming this limitation is not trivial, because mechanisms responsible for map development drive receptive fields (RF) of nearby neurons to be highly correlated, while co-oriented RFs of opposite phases are anti-correlated. In this work, we model V1 as two topographically organized sheets representing cortical layer 4 and 2/3. Only layer 4 receives direct thalamic input. Both sheets are connected with narrow feed-forward and feedback connectivity. Only layer 2/3 contains strong long-range lateral connectivity, in line with current anatomical findings. Initially all weights in the model are random, and each is modified via a Hebbian learning rule. The model develops smooth, matching, orientation preference maps in both sheets. Layer 4 units become simple cells, with phase preference arranged randomly, while those in layer 2/3 are primarily complex cells. To our knowledge this model is the first explaining how simple cells can develop with random phase preference, and how maps of complex cells can develop, using only realistic patterns of connectivity.</p>
BibTeX
@article{antolik2011development,
  author = {Ján Antolík and James A. Bednar},
  title = {{Development of Maps of Simple and Complex Cells in the Primary Visual Cortex}},
  journal = {Frontiers Comput. Neurosci.},
  year = {2011},
  doi = {10.3389/fncom.2011.00017},
}
Figure