Summary: Researchers have identified a specialized class of neurons, called IC-encoder neurons, that cause the brain to “see” illusory shapes—such as squares or triangles that aren’t present in the raw visual input. These cells receive top-down signals from higher visual areas and then complete missing contours in primary visual cortex, actively constructing the perceptual edge and shape information we experience.
In experiments, directly stimulating these IC-encoder neurons produced the same cortical activity patterns normally observed when an animal sees an illusion. The results shift our view of vision from a passive recording of the outside world to an active, constructive process with clear implications for perceptual disorders and for how the brain forms a coherent sense of reality.
Key Facts
- IC-Encoder Neurons: A distinct population in primary visual cortex that responds to illusory contours and supports pattern completion.
- Active Vision: Higher-order visual areas send feedback to lower areas to generate completed representations when sensory information is incomplete.
- Clinical Relevance: Insights into how false percepts form may help explain features of conditions such as schizophrenia and other disorders with aberrant perceptions.
Source: Allen Institute
What is an illusion? An illusion occurs when perception does not match the raw sensory input. For example, a display of four Pac Man–like black shapes can produce the vivid perception of a white square that does not exist in the incoming retinal signal. The brain fills in missing edges and surfaces to create a coherent object.
In a new study published in Nature Neuroscience, teams from the University of California, Berkeley and the Allen Institute mapped the neural circuit and identified the specific cell type that is central to detecting illusory contours—those inferred edges and borders that define perceived shapes.
Hyeyoung Shin, Ph.D. (now at Seoul University), Hillel Adesnik, Ph.D., and colleagues identified IC-encoder neurons that reliably signal the presence of illusory bars or contours even when the stimulus contains only component fragments. The researchers describe this process as recurrent pattern completion: higher visual areas form an inferred object representation and then send that information back to primary visual cortex, where IC-encoders broadcast and reinforce the completed pattern.
“Because IC-encoder neurons have this unique capacity to drive pattern completion, we think they may have specialized synaptic output connectivity that allows them to recreate the pattern very effectively,” Shin said. She added that these neurons receive strong top-down input: the representation of the illusion appears first in higher visual areas and is then communicated back to the primary visual cortex, where IC-encoders translate it into local activity.
A useful analogy is a manager giving a directive to a frontline worker: higher-level circuits interpret incomplete sensory data as a specific object and instruct lower-level neurons to implement that interpretation. In vision, this instruction can make the lower cortex “see” edges or contours that the eyes did not explicitly provide.
To test causality, the team presented mice with illusory shapes such as the Kanizsa triangle while recording brain activity across regions. They then used two-photon holographic optogenetics to selectively stimulate IC-encoder neurons in the absence of any visual stimulus. Activation of those neurons recreated the same activity patterns seen during actual perception of the illusion, demonstrating that these cells are sufficient to generate the neural signature of an inferred contour.
These results illuminate how feedback loops between higher and lower cortical areas implement perceptual inference. The Allen Institute’s OpenScope program contributed brain-wide electrophysiological measurements using multiple Neuropixels probes, allowing the team to observe feedback dynamics with millisecond precision. “OpenScope provided access to unique brain-wide recordings,” said Jerome Lecoq, Ph.D., associate investigator at the Allen Institute, enabling the detection of feedback loops in real time.
Beyond basic vision science, the study has clinical significance. Abnormal, spontaneously emerging object representations occur in some neurological and psychiatric conditions, including schizophrenia. Understanding which cells and cortical layers generate these representations is a step toward therapies that target pathological activity patterns.
Conceptually, the findings recast perception: rather than acting like a passive camera, the visual system behaves more like an active processor that combines incoming data with prior expectations to construct the most likely scene. This interpretive machinery means perception is flexible—and in some cases vulnerable to errors or manipulations—because the brain actively completes and refines incomplete sensory input.
Recognizing the neural elements that carry out pattern completion opens avenues for future research into how expectations shape what we see and how those processes break down in disease.
About this visual neuroscience and perception research news
Author: Peter Kim
Source: Allen Institute
Contact: Peter Kim – Allen Institute
Image: The image is credited to Neuroscience News
Original Research: Open access. “Recurrent pattern completion drives the neocortical representation of sensory inference” by Hyeyoung Shin et al., Nature Neuroscience.
Abstract
Recurrent pattern completion drives the neocortical representation of sensory inference
When sensory information is incomplete, the brain relies on prior expectations to infer perceptual objects. Despite the central role of this process in perception, the underlying neural mechanisms of sensory inference remain incompletely understood.
Using illusory contours (ICs), multi-Neuropixels recordings, mesoscale two-photon (2p) calcium imaging and 2p holographic optogenetics in mice, the study reveals neural codes and circuits that support sensory inference.
The authors discovered a specialized subset of neurons in primary visual cortex (V1) that respond specifically to illusory bars but not to the individual image components. Selective holographic activation of these “IC-encoders” recreated the V1 representation of illusory contours in the absence of any visual stimulus, indicating that neurons encoding sensory inference are specialized for receiving and broadcasting top-down information locally.
More broadly, pattern-completion circuits in lower cortical areas may selectively reinforce activity patterns that match prior expectations, and this selective reinforcement may be an essential step in perceptual inference.