Basic Anatomy

The macaque primary visual cortex, like that of all mammals studied, resides in the posterior pole of the occipital cortex (Fig. 8). V1 extends rostrally almost to the lunate sulcus and posterolaterally almost to the inferior occipital sulcus; the V1/V2 border is met before either sulci. Much of V1 is found within the calcarine sulci, and the limits of V1 within these folds can be marked by the distinct appearance of V1 layer 4 (Fig. 9).

Figure 8

Visual input to the brain goes from eye to LGN and then to primary visual cortex, or area V1, which is located in the posterior of the occipital lobe. Adapted from Polyak (1).

Figure 9

Nissl-stained section to show the border between area 17 (V1) and area 18 (V2).

The anatomy of the primate primary visual cortex has been worked out in great detail. The three basic organizing principles of primate V1 are the: 1) laminar and 2) columnar arrangement of excitatory and inhibitory neurons (an arrangement shared with other neocortical areas) and 3) the regular spacing of anatomical/functional compartments revealed by cytochrome oxidase (CO) labeling.

Several simple facts and rules of thumb can be stated before addressing the details of these anatomical features:

• The six cortical layers introduced by Brodmann (layer 1 most dorsal, layer 6 most ventral) have been subdivided time and again as additional aspects of the neurons and their connections have been revealed.
• Layer 1 is nearly aneuronal, composed predominantly of dendritic and axonal connections. In 1 mm³ of V1 tissue, one could expect to find approximately 4,700 neurons, 2,900 microglia, 3,400 oligodendrocytes and 49,000 astrocytes (20). Of the neurons present, over 80% are GABAergic (21).
• Approximately 20% of the neurons in layers 2–6 are inhibitory interneurons (GABAergic) that make major contributions to the function of V1 circuits but do not project axons outside of this area.
• Layers 2/3 (the "supergranular" layers) contain many excitatory projection neurons that send axons to extrastriate cortical regions (e.g., V2, V3, V4, MT, etc.).
• Layer 4 (the "granular" layer) is divided into four horizontal sublayers: 4A, 4B, 4Cα, and 4Cβ. Layers 4Cα and 4Cβ are the major recipients of LGNd innervation. The LGNd magnocellular (M) and parvocellular (P) layers project to 4Cα and the 4Cβ, respectively. Thus, the M and P streams remain segregated at this stage (but see qualification below).
• Layers 5/6 (the "infragranular" layers) contain many excitatory projection neurons that innervate the LGNd to provide feedback to this relay area.
• Labeling of V1 and V2 for CO content has revealed CO-rich and -poor regions termed "blobs" (or puff, spots, or patches) and "interblobs" in V1 and the thick, thin, and pale stripes in V2. The relevance of these compartments is still a matter of some debate.
• CO blobs do not extend uniformly from the surface of layer 1 to the base of layer 6. Instead, they concentrate in layers 3B, 4B, and 4C (Fig. 10).
• Projections to and from the CO blobs and interblobs are relatively distinct as follows (Fig. 11):
  • Pβ (for primate P or midget ganglion cells) project to the LGNd parvocellular layers (see the chapter on midget pathways of the retina), which project to V1 layer 4C and then project to the interblobs of V1 layers 2/3; and finally these layers project to the pale stripes of the V2 geniculate.
  • The blue-yellow ganglion cells (see the chapters on S-cone pathways and color vision) project to the intercalated layers and S/K (K for "koniocellular") layers of the LGNd, which project to the blobs of V1 layers 2/3, which project to the thin stripes of V2.
  • Pα (M cells or parasol ganglion cells) project to the LGNd magnocellular layers and thence to layer 4C of V1. This layer projects to the blobs of V1 layers 2/3 and also to layer 4B of V1. Layer 4B, in turn, projects to the mediotemporal area (MT) and to the thick stripes of V2.
• Layers 1 and 2 receive feedback inputs from extrastriate cortex.
• The neuroglia in the young adult macaque V1 account for 20-35% of all cells (not including pericytes). The neuroglia can be divided into oligodendrocytes (~7-10% of cells), astrocytes (~11-20% of cells), and microglial cells (~1.5-3% of cells) (20, 22).

Figure 10

Nissl (left) and cytochrome oxidase (right) labeled cross sections of the visual cortex of a macaque monkey, showing the individual layers.

Figure 11

The projections of the small (P cells) and large (M cells) ganglion cells from the two eyes to parvocellular and magnocellular layers of the LGN, respectively.

Neuronal Constituents

There are three basic types of neurons in the primate V1 (Fig. 12):

Figure 12

Basic cell types in the monkey cerebral cortex. Left: spiny neurons that include pyramidal cells and stellate cells (A). Spiny neurons use the neurotransmitter glutamate (Glu). Right: smooth cells that use the neurotransmitter GABA. B, cell with local (more...)

• spiny pyramidal cells (excitatory)
• spiny stellate cells (excitatory)
• smooth or sparsely spinous interneurons (almost all are GABAergic)

The pyramidal cells, as you might guess, have cell bodies that appear like pyramids, have apical dendrites (that is, extending toward the white matter), and their dendrites are covered with spines. Spiny stellate cells are generally smaller, and their cell bodies more resemble a star shape. They also have spiny dendrites (Fig. 12A). The interneurons have more rounded cell bodies and have little or no spines on their dendrites. Over the years, anatomists have given more or less descriptive appelations to a number of unique subtypes, which are briefly described below.

• Chandelier cell – inhibitory; axons show characteristic branching into vertical sections resembling candlesticks (Fig. 12E).
• Neurogliaform cells – inhibitory; resembling glial cells but definitively neurons; axon and dendrites branch but remain in the locale of the soma, giving rise to a dense spherical region of fibers (Fig. 12G).
• Double bouquet cell – inhibitory; axon collaterals and bitufted dendrites extend vertically in a tight bundle (Fig. 12C). Antibodies to calbindin specifically mark these cells, and by using this strategy, researchers have found that these cells are regularly spaced in layer 3, thereby forming yet another columnar unit. There are approximately 7 to 15 double bouquet cells per 10,000 μm² (23).
• Basket cells (24, 25) – inhibitory; relatively large cell bodies extend myelinated axonal branch laterally for a great distance. The axonal branches terminate on pyramidal cell bodies and dendrites in a manner resembling a basket. Dendrites extend vertically (Fig. 12, D and H).
• Cajal-Retzius cells – inhibitory; Cajal and Retzius found this cell type independently in layer 1 of neonatal animals and humans, respectively. They may or may not exist in the adult.
• Meynert cell – excitatory; large pyramidal cell first described by Meynert; the "outer" Meynert cell bodies are found in 4B and are regularly distributed relative to pyramidal cell cones (22) (see below). These cells project to area MT. The "inner" Meynert cell bodies are regularly spaced at the layer 5/6 border, lying below CO interblobs. These cells are also found in layer 6. These cells typically have asymmetric, lateral dendritic arborizations, are highly direction selective, and are also known to project to area MT.

The cell types described above represent only a few of the more unique varieties. Jennifer Lund and colleagues (26-29) have rigorously studied the presumptive interneurons of the macaque V1 using the Golgi impregnation staining method. On the basis of a range of morphological criteria (the laminar position of somata, axonal arborization, dendritic arborization, pattern and extent of branching), they have described over 40 types of interneurons alone.

Additional low-density interneurons might remain unidentified. The classes that have been described might be further divided based on chemical content and/or physiological response properties (Fig. 12). Alternatively, it should be noted that many of the anatomically distinct types may be physiologically indistinguishable from one another.

The Cortical Layers

Many numbering schemes exist for the neocortical layers in general and the layers of V1 in particular. I have adopted the basic scheme of Brodmann, who defined the cortical layers based on careful histology. I have also adopted the refinements suggested by the work of Lund and colleagues (see below) (Fig. 13).

Figure 13

Nissl stain of the visual cortex reveals the different layers quite clearly.

Layer 1 is composed of a dense network of synapses formed between the apical dendrites of layer 2–5B pyramidal cells (29) and the collected inputs from LGNd K layers (30, 31), the pulvinar, feedback pathways from extrastriate areas, nonspecific thalamic nuclei, and other subcortical regions. Thus, although layer 1 contains few neurons, it is a networking layer that has a direct, concerted affect on the firing properties of pyramidal cells in deeper layers.

The supragranular layers (2, 3A, and 3B) contain many somata and dendrites of pyramidal cells. Layers 2 and 3A receive little thalamic inputs (the K layer projection to V1 blobs is focused mainly in 3B). They do not receive direct inputs from layer 4C, in contrast to layer 3B that receives a massive 4C input. Layers 2 and 3A do contain many axons and dendrites of neurons found in all other cortical layers.

Layer 4A shows up in the Nissl stain as a dark band of small, granule-like cells. This and the lack of pyramidal cells set this layer apart from layer 3B.

Layer 4B is a cell-poor layer that contains a low density arrangement of pyramidal cells—the large outer Meynert cells. This layer receives a strong input from the underlying 4Cα, and the Meynert cells are known to project directly to area MT as well as the superior colliculus (SC) (32). Cells in layer 4B also project to the thick CO stripes of V2, which in turn project to MT. Thus, this layer plays a substantial role in conveying information in the M pathway/dorsal stream.

Layer 4C is distinguished from 4B in part due to the great density of stellate cell packing in this layer. Polyak (1) divided 4C into the α and β divisions, based solely on histology; upper 4C neurons are more dispersed. Differences in geniculate innervation patterns (i.e., the M layers of LGNd to 4Cα and the P layers to 4Cβ) (33, 34), the projection patterns of the 4C cells (35) and physiological properties (36) have overwhelmingly supported this division. Additional evidence has suggested that the upper and lower divisions of 4Cα may be distinct as well (26, 33, 37, 38).

4Cβ spiny stellate neurons provide a dense innervation of layers 4A and 3B and sparse innervation in 4C, 5, and maybe 6 (35). No direct projections from 4Cβ to 3A, 4B, or 5B have been found (26). In contrast, as already noted, 4Cα projects heavily to 4B.

Local layer 5A neurons do not project to 4B and only weakly innervate 4Cβ. They innervate all other layers. 5A neurons appear to project primarily to 3B and 1 (26). Pyramidal neurons in 5B send recurrent axons to layer 3A (39). Meynert cells at the 5/6 border and within layer 6 project to both MT and SC (40).

Layer 6: It is known that neurons in layer 6 send recurrent axons into 4Cβ (39). Other neurons (or possibly even the same ones) project axons back to the LGNd (41). Layer 6 also receives a direct input from the LGNd (42), thereby forming a neural loop.

V1 Cortical Columns

The term "cortical column" refers to the notion that cells arranged vertically from the surface of the cortex to the white matter might compose functional or anatomical units. Thus, a cortical column can be defined on the basis of anatomical features (e.g., stereotyped patterns of pyramidal cell apical dendrite bundles), functional features (e.g., columns of cortical cells, all responding to the same stimulus orientation), or both.

Many types of columns have been proposed including ocular dominance, orientation, spatial frequency, and color columns. The details of these columnar arrangements will be described in later sections dealing with the physiology of V1. However, I will here briefly describe the ocular dominance columns to provide an example of columnar arrangement in the primary visual cortex.

Ocular Dominance Columns

Visual signals from the two eyes remain segregated in the LGNd (Fig. 11) and in the geniculorecipient layers of area V1 (Fig. 14). One can observe this segregation by measuring the electrophysiological responses of the units in layer 4C. As the recording electrode is moved within layer 4C, there is an abrupt shift as to which eye drives the unit. In layer 4C, the shift from one eye to the other takes place over a distance of less than 50 μm. Signals from these bands converge on individual neurons in the superficial layers of the cortex, thereby forming columns dominated by one eye or the other in an alternating fashion (Fig. 15).

Figure 14

The signals from each eye are segregated into different ocular dominance columns within area V1, layer 4C.

Figure 15

The ocular dominance columns in area V1 can be visualized by using markers. When the marker is injected into one eye, it is transported via the LGN nucleus to the cortex. The light bands in this tangential section show the places where the marker was (more...)

Cytochrome Oxidase Labeling

CO is an integral transmembrane protein that is found in the inner mitochondrial membrane. Because it acts to catalyze the generation of adenosine triphosphate (ATP), which is itself an energy molecule used in an enormous range of cellular processes; the amount of CO in a cell or nucleus can be used as an indicator of cellular activity. In 1978, Margaret Wong-Riley discovered that, quite apart from its use as an activity indicator, CO labeling in the primary visual cortex reveals a striking array of CO-rich and -deficient regions (i.e., blobs and interblobs) (Fig. 16).

Figure 16

Diagrammatic representation of the relationships between CO "blobs", layer 4 cones, and pyramidal cell modules. A, three-dimensional representation. The "blobs" in layer 2/3I are aligned in single rows along the 400 μm-wide ocular dominance columns (more...)

Presumably, these regions reflect stable differences in energy consumption in two populations of neurons that were previously undistinguished. Assuming this were true, what other differences might exist between blobs and interblobs? It should be noted that dense CO staining does not exist throughout the entire cortical depth. Instead, CO blobs are most visible in layers 3B, 4A, and 4C. Be that as it may, if these dense regions were in fact different from their CO-deficient neighbors in some way, the cells directly above and below rich and poor regions might also show differences in connectivity and functional characteristics, correlating with the cells above and below them.

For instance, blobs in layer 3B have been shown to correspond to specific geniculate innervation in macaques (43, 44). Boyd and Casagrande (45) found recently that V1 cells projecting to MT in bush babies and owl monkeys were more numerous below the CO blobs of layer 3B than below the interblobs. Similar data in macaque monkeys also shows clustering of MT projecting cells in periodical steps through the lateral extent of V1 (46-48) but did not find a correlation between these clusters and the CO (Fig. 17 and Fig. 18).

Figure 17

Schematic drawing of the visual cortex to show neurons, layers, columns, and CO blobs.

Figure 18

Reconstructions of neurobiotin-injected V1 4C cells that have dendrites confined to layer 4C and dense axonal terminations in interblobs (A, B) or dendrites extending into layer 4C and dense axonal terminations in both blobs and interblobs. From Yabuta (more...)

Feedforward and Feedback Pathways

The signals from the two retinas are communicated to area V1 via the LGN. In the macaque monkey, after the signals are processed in V1 they are communicated via multiple pathways to the 30+ visually responsive, extrastriate cortical areas (Fig. 19).

Figure 19

Much of V1 is located in the calcarine sulci, and its relationship to other brain areas is best shown by unfolding the brain and showing it flattened open. The visually responsive areas of the macaque monkey are shown in color. From Van Essen et al. (78). (more...)

Anatomical hierarchical models place the visual cortical areas into a multi-level processing model based upon the pattern of feedforward, lateral, and feedback pathways found in each area. In general, feedforward pathways are considered to be those that project from the supragranular layers of one area and terminate in layer 4 of the target area. In contrast, feedback projections are those that arise from the infragranular layers of one area and terminate outside of layer 4 in the target area (49-51).

In the previous sections, I have described the anatomical borders of V1, the cells that exist in this area, the arrangement of those cells in layers, columns, and CO compartments, and I have hinted at the visual pathways that course into and out of V1. It is true that visual information is passed from the retina to the LGNd to V1 to the higher cortical areas and that the higher areas project back to V1 and V1 to the LGNd (there are no efferent projections to the primate retina). However, although this view of visual processing is accurate, it does not capture the wealth of upstream and downstream connections found in V1.

Direct feedforward projections to V1 originate from the pulvinar, LGNd, claustrum, nucleus paracentralis, raphe system, locus coeruleus, and the nucleus basalis of Meynert (31, 33, 43, 52-57).

Direct feedforward projections from V1 extend to V2, V3, V5 or MT, MST, and FEF (41, 46-48, 58-63).

Direct feedback projections to V1 originate from V2, V3, V4, V5 or MT, MST, FEF, LIP, and inferotemporal cortex (47-49, 57, 61, 62, 64, 65).

Direct feedback projections from V1 extend to SC, LGNd, pulvinar, and pons (32, 41, 54, 66, 67).

The connections listed above represent only a subset of the direct and indirect projections that carry signals into and out of the primary visual cortex and do not even begin to describe the extensive lateral connections within V1 itself.

Receptive Field Properties

Each cell in the visual cortex has a receptive field (RF), a discrete area in space relative to the fovea where the presentation or removal of a visual stimulus will cause cellular activation. By definition, stimuli presented outside of this receptive field will neither increase nor decrease the ongoing activity of that individual cell. The location and size of a RF can be considered among the most basic of RF properties.

Retinotopic Maps in V1

The spatial position of the ganglion cells within the retina is preserved by the spatial organization of the neurons within the LGN layers. The posterior LGN contains neurons whose RF are near the fovea. Progressing from posterior to anterior, the RF locations become increasingly peripheral in the retina (68). This spatial layout is called retinotopic organization because the topological organization of the RFs in the LGN parallels the organization of the retina.

The signals in area V1 are also retinotopically arranged. From electrophysiology in monkeys, one can measure the location of RFs with an electrode that penetrates tangentially through layer 4C, transversing through the ocular dominance columns. The RF centers of neurons along this path correspond systematically to locations from the fovea to the periphery. This trend is interrupted locally by small, abrupt jumps at the ocular dominance borders. Thus, the striate cortex retains the retinotopic map of the contralateral visual field that is developed in the LGN.

Orientation and Direction Selectivity

V1 is the first site where strong orientation and direction selectivities are observed in the macaque monkey (16). Although the vast majority of V1 cells show some degree of orientation selectivity, only approximately 25-35% of V1 cells are strongly directionally selective (69, 70). The classic method for testing orientation and direction selectivity is to measure the spike rate of a single cell in response to drifting oriented luminance bars and/or drifting luminance spots (see Fig. 21).

The concept of an orientation column can be easily appreciated by examining Fig. 20 and Fig. 21. When an electrode is lowered into V1 at an angle relatively parallel to the cortical layers (Fig. 22), the orientation selectivities of the cells encountered vary systematically, where adjacent cellular regions share approximate orientation preferences.

Figure 20

The unfolded striate cortex has a shape like a pear. It would be a quarter sphere if the visual fields were equally represented everywhere, but instead it is greatly distorted by the disproportionate representation of parts near the center of gaze (fovea), (more...)

Figure 21

A tuning curve and corresponding polar plot obtained from two macaque V1 cells in response to drifting luminance bars of systematically varied orientation and direction. The responses of one orientation selective cell and one nonselective cell are provided (more...)

Figure 22

One extensive electrode penetration in macaque V1. The short, near-vertical lines represent recording sites, and polar plots for each site are indicated. This figure shows the preferred orientation of each cell in relation to the cortical layers, CO compartments (more...)

Such recordings led Hubel and Wiesel to propose models of functional organization like the one shown in Fig. 23.

Figure 23

The ice-cube model of the cortex. It illustrates how the cortex is divided, and at the same time, into two kinds of slabs, one set of ocular dominance (left and right) and one set for orientation. The model should not be taken literally: Neither set is (more...)

Recently, the orientation columns of V1 first described by Hubel and Wiesel have also been recast into more complex geometries, such as partial columns ("slabs") and pinwheels, as dictated by the increasing volumes of evidence (71).

Binocularity and Binocular Disparity

Hubel and Wiesel (15) first demonstrated the presence of cells in primate V1 that responded preferentially to visual stimulation in both eyes as opposed to one eye alone. They classified cells using a 7-point scale where:

• 1 – cells excited by the contralateral eye alone.
• 2 – cells with a strong bias for contralateral stimulation.
• 3 – cells with a weak bias for contralateral stimulation.
• 4 – cells that responded maximally to stimulation in both eyes.
• 5 – cells with a weak bias for ipsilateral stimulation.
• 6 – cells with a strong bias for ipsilateral stimulation.
• 7 – cells excited by the ipsilateral eye alone.

The proportion of monocular cells (points 1, 2, 6, and 7) versus binocular cells (points 3, 4, and 5) was found to vary with the layer in which cells were recorded. Cells in the retinogeniculato-recipient layers 4Cα and 4Cβ were found to be exclusively monocular. Anatomical studies have since shown that these cells receive monocular inputs from the LGNd that do not converge from the alternating layers (34, 39).

What might be the usefulness of binocular cells? When an observer fixates on a visual object, the image of this object is positioned on corresponding regions of the two retinas. At the same time, however, objects in front of and behind the fixation point create images that lie on non-corresponding regions of the two retinas. The degree to which the images are non-corresponding (as measured by difference scores in retinal eccentricity for instance) is defined as binocular disparity. The ability to use binocular disparity to determine the distance of an object from oneself, and its relation to the fixation plane, is called stereopsis.

Hubel and Wiesel first described the presence of neurons in V1 that are sensitive to binocular disparity. That is, some neurons may show maximal firing rates when a stimulus has positive disparity and weak or even inhibited (below baseline) firing rates to 0 disparity stimuli (in the fixation plane) or negative disparity stimuli. These cells would be described as disparity sensitive and negative disparity tuned. If the tuning were reasonably sharp, the cell might be described as negative disparity selective, i.e., it responds soley, not simply preferentially, to negative disparity. We now know the percentage of disparity-sensitive cells and the proportion of positive, negative, and 0-disparity tuned cells. Clearly, there are no disparity tuned cells in layers 4Cα and 4Cβ because binocularity is a prerequisite for disparity sensitivity. As Livingstone (72) noted, it is safe to assume that negative and positive disparity selective cells are involved in stereopsis. Of the 0-disparity tuned cells, all may be involved in stereopsis or, alternatively, perhaps only those with very narrow tuning (i.e., selective ones) contribute to this perceptual ability (73). The work of DeAngelis, Newsome, and Cumming (74, 75) elegantly addresses this issue by attempting to alter stereopsis in trained monkeys by microstimulating subregions of disparity-tuned cells in area MT.

Response Timing in the Visual Pathways

Another feature of visually evoked, single-cell response that is currently gaining attention is timing. The order in which visual areas become active and the range of activation times within any given area are important issues. As described above, a prevailing model of primate vision proposes hierarchical stages of visual areas based on anatomical connectivity (49, 50, 76-78). There is no doubt that a single stimulus initiates a characteristic flow of neuronal discharge from the retina to the LGNd to V1 in a hierarchical manner. Unfortunately, the complexity of corticocortical connections and functional differentiation in primate visual streams renders further predictions highly speculative (51).

We could hypothesize, for instance, that the heavily myelinated path from V1 to MT would lead to fast activation of this area. However, would all MT cells be activated in this manner, or would others wait for activation via the more indirect V1 to V2 to MT route? Many studies have reported response onset latencies from a single visual area. Using these data to construct a chart of activation times across the visual system, however, reveals a host of complications (79).

To begin with, the vast majority of studies that report latencies do so for only one or two visual areas. An activation analysis in 10 visual areas, therefore, might involve cross-comparisons between 5 and 10 separate studies. In one such comparison, we might conclude that, on average, V1 cells become active long before cells in the frontal eye field area (FEF) (80, 81). In another such comparison, we would be forced to conclude the opposite (82) (for a review, see Nowak and Bullier (79)).

The cause for this seeming discrepancy, and many others like it, is undoubtedly the use of different experimental and data analysis protocols. The visual stimuli used (flashing versus moving; spot, bar, grating, or otherwise; low versus high luminance or contrast, etc.), animal preparation used (anesthetized or not), and data analysis strategy taken for marking latencies (qualitative or quantitative, etc.) all vary across the existing reports.

Consequently, in a recent study, we sought to remove these confounds by obtaining onset latencies from many primate visual areas using the same experimental and analytical techniques (83). The results demonstrated that the two major functional streams in the primate visual system (the M/dorsal stream involved in motion perception and tracking and the P/ventral stream involved in object recognition and color coding) respond with very different time courses (Fig. 24).

Figure 24

Onset latencies. Cumulative distributions of visually evoked onset response latencies in the LGNd, striate, and extrastriate areas as labeled. The percentiles of cells that have begun to respond are plotted as a function of time from stimulus presentation. (more...)

In general, M-stream areas respond rapidly and concurrently, whereas their P-stream counterparts respond slowly and sequentially. Particularly relevant to this review, the LGNd M cells become active 15 to 20 msec faster than do their P-cell counterparts (83, 84), and this temporal separation is maintained in the geniculorecipient layers 4Cα and 4Cβ (80, 82, 83). Thus, a functional separation in response timing exists in the M and P pathways of the subcortex and V1 and appears to be conveyed to the dorsal and ventral streams of the extrastriate cortex.

Illusory Contour Perception

The modern view of visual perception is one of dynamic processes that go beyond the simple replication of visual information provided to the retina. For over 80 years, Gestalt psychologists have argued that the act of perception creates a Gestalt, a figure or form that is not a property of an object observed but represents the organization of sensations by the brain (85). This dynamism is thought to be crucial for the performance of simple, everyday visual tasks, such as the recognition of an object that is partially occluded. Thus, the study of how the brain is capable of filling in the missing pieces is an important topic; one that has most often been carried out through the use of "illusory contours".

Illusory contours (ICs) are defined by their subjective appearance in the absensce of any luminance differences in the image itself. A classic example of an IC is the illusory square, where four sectored discs oriented in an appropriate manner generate the percept of a square. In this instance, it is clear that no true luminance differences exist to form a complete square, and yet a square is perceived (Fig. 25).

Figure 25

Illusory contours as seen in the illusory square.

Behavioral tests indicate that cats and monkeys are capable of perceiving such contours (86-88). Recent physiological studies have examined whether single cells in striate and extrastriate cortex respond similarly to conventional luminance contours and those defined by illusions (89-92). In each of these studies, IC-responsive cells were reported. However, these studies provide conflicting evidence about the origin of IC responsivity, arguing for or against the presence of IC-responding cells in V1. The use of alternative ICs in the different studies confuses this issue further.

We sought to approach illusory contour perception from the broader issue of contour perception in general (93). The issue here is how the brain encodes a stimulus boundary (e.g., a square) when the visual cue that defines that boundary (e.g., color, luminance, texture, motion, illusion, etc.) is varied. As Gestalt psychologists would note, the stimulus is very different in each of these cases, but the fundamental percept, the square, remains the same. We asked two fundamental questions. First, are there V1 or V2 cells in cat and monkey capable of responding in a similar fashion to a boundary, regardless of the cue defining it? Alternatives to this include the origin of cue invariance at later stages of visual processing or the separate channeling of cue information via different subpopulations of cells, each responsible for one cue. Second, if such cells exist, are they equally prevalent in V1 and V2? Our results demonstrated that a subpopulation of cells in both regions can respond to a stimulus boundary, such as an oriented bar, in a cue-invariant manner, although this property was rare in V1 while prevalent in V2 (Fig. 26).

Figure 26

Example of a cue-invariant cell in cat area 18. This cell responds with the same degree of orientation bias and to the same preferred orientation, regardless of whether the bar is defined by simple luminance (top), texture (middle), or isoluminant gratings (more...)

One conclusion from this work is that cells in the very first stage of cortical processing are already capable of responding to a single stimulus boundary in a complex manner that allows for object detection, even as the cues defining the object change or are partially occluded (93).

Understanding Vision: A Problem of Reverse Engineering

Our attempt to understand the visual pathways is very much like approaching a machine about which we know nothing but its basic functions. By way of analogy, we all know how to operate a car (more or less) and recognize its basic function—to get us from one place to another. Now let us assume that we knew absolutely nothing but this and, for some reason, decided we should learn every detail of the car's inner workings. We might start by trying to figure out the parts of the car devoted to this central function and those that are not. We remove the bumper, horn, AC, and windshields and find that the car still runs splendidly. We remove small parts of the engine, piece by piece until the car no longer starts, analogous to the lesion studies in the brain. Slowly, we begin to understand what parts of the car (brain) are involved in locomotion (vision). In vision research, we are very much at this stage of the game, still wanting to know what each part actually does, when it does it, and how all the individual parts act in concert. It is the hope of many researchers, myself included, that a careful investigation into the structure and function of the visual pathways using chemical, electrophysiological, genetic, and behavioral approaches will culminate in a true understanding of how the brain provides us with this most crucial of sensory capabilities, vision (Fig. 27).

Figure 27

Reverse engineering: the brain studying the brain.