One way to approach the question of how the brain breaks up visual information is to ask what types of processing are anatomically separated. Over 50% of the rhesus monkey cerebral cortex, a primate frequently used in vision studies as a model for many human abilities, appears to be involved in processing visual information(1). On the basis of anatomy and physiology, over thirty distinct "ar eas" involved in vision have been defined. Based on features such as the differing inputs and outputs of each area and the whole or partial representations of the visual scene (map) found in many areas, it has been assumed that each area is involved in only some small function of the overall information processing that the visual system does. Thus, to determine the functions of an area is to determine some of the separate processing tasks, or computations, into which visual processing is broken. In t his article I discuss the advantages and the problems of studying animals with selective lesions of the cortex as a way of revealing how processing tasks are separated. Also discussed is the primate cortical area V4, an area of some controversy.
Methods for Studying Information Processing
Several approaches have been used to study the types of processing taking place in various anatomical areas. These have included studying of brain damaged human patients, imaging of brains involved in various cognitive tasks, recording the activity of individual neurons, and creating specific lesions of animal brains followed by analysis of deficits in visual ability. It is unlikely that any one approach will be sufficient to fully describe any one a rea. Each approach has various benefits and limitations. A detailed analysis of all of these approaches is beyond the scope of the current presentation, but the following brief observations will be made: Patients with disorders like prosopagnosia can pr ovide fascinating clues into how different abilities segregate, but for the purposes of systematic experimentation, there are ethical and technical limitations. Technically problematic is the tendency of brain injuries to affect more than one area and to affect both grey and white matter. Imaging of brain activity during cognitive function is becoming more powerful as temporal and spatial resolution increases. By comparing activity during different tasks, imaging attempts to narrow possible roles for a n area to determine what types of processing may be occurring there. Recording the activity of single neurons in the cortex has also proven to be technically feasible and scientifically informative. Single cell recordings are very limited in scale, for cells can only be recorded for a few hours at best, and recording more than a few cells simultaneously is very difficult or impossible.
Advantages and Problems of Studying Lesions
Systematic testing of visual capabilities after s elective lesioning of animal cortex allows for the observation of the capabilities of the visual system as a whole. The observation of a lesion induced deficit is usually taken to mean that the lesioned area was necessary for the impaired ability. Knowin g what capabilities an area subserves allows inferences to be made about what role that area is playing in those capabilities, what sorts of visual information that area uses, and what kinds of different abilities are segregated. Single cell recordings a ttempt to determine what functions an area is performing by describing the activity of its component parts. Unlike single cell recordings, studies of selective lesions can reveal what an area is doing without necessarily having to deal with the much hard er problem of how it does it. As with observations of human lesion effects, this approach can reveal intriguing and unexpectedly clustered or isolated deficits (like prosopagnosia).
It often remains unclear, however, whether the lesioned area was si mply a way-station for visual information needed from the lesion by areas higher up in the cortex, or whether the area was sufficient for the now impaired ability, or whether it is something in between. Furthermore, the detectable deficit may not provide information on the normal function of that area (since removing that area may have left the system somehow unstable, or at least unlike the normal system in some important way). In analogous terms, just because a radio howls upon removal of a resistor does not imply that the purpose of that resistor was to act as a "howl suppressor"(2). Finally, some abilities may be more susceptible to disruption than others, and therefore are more likely to be observed. However, even with these limitations, selectiv e lesions are a powerful tool for investigating how visual capabilities are segregated in the visual cortex.
Single cell recording versus lesion studies: Is V4 a color area?
Early single cell recordings made by Hubel and Wiesel(3) , Zeki(4), and others found cells in V2 sensitive to binocular disparity of stimuli, cells in V4 sensitive to color, and cells in MT sensitive to motion. These and similar discoveries led to the theory that each area was responsible for analyzing one att ribute of a stimulus in parallel, often referred to as the "one-area-one-function" model(5). Attributes were thought of as features like stereoscopic disparity, color, and motion. The model considered these to be fundamental segregations of tasks in the visual system. Investigations in the twenty years since Hubel, Wiesel, and Zeki on visual cortical areas in general and V4 in particular have put this model in doubt, however(5). The division of labor in the visual cortex does not seem to be done exact ly along the conceptual lines incorporated in the model. At the very least, it appears that multiple types of processing are often present in the same area, and that the segregation of different types of processing is not as complete as the model might s uggest6. Schiller(5)(7) has proposed a more evolutionarily plausible framework with which to view the purpose of the many higher cortical areas: each evolved to extend or enhance visual ability in some fitness-enhancing way. This led to enhan cements such as faster processing, less reliance on reflex action, greater temporal, spatial or other kinds of resolution, and improved learning. The evolutionary pressures on the development of sophisticated information processing abilities is an area w orthy of further investigation.
The color selectivity of area V4 was central to the original formation of the "one-area-one-function" hypothesis, and has always been an area of debate(5). In 1973, Zeki(4) first reported highly selective sensitivity to color when recording from single cells in V4. Later investigations of V4 found widely disparate levels of color selectivity. Then, Schein and Desimone(8) reported that V4 cells responded very broadly to colored stimuli, and that this broad response was probably responsible for the widely varying reports of the extent of color selectivity in V4.
Analysis of color may yet prove to be separate from other kinds of processing. Achromatopsia is a known deficit in humans that is correlated to injury to an area in the human brain thought to be approximately anatomically analogous to V4(9). These patients describe the world as appearing gray and are unable to select colored stimuli from achromatic ones. But upon bilateral ablation of V4 in monkeys, Heywoo d and coworkers(9) did not find similar deficits. Heywood and Cowey(6) found only moderate deficits in color discrimination in monkeys with bilateral lesions of V4. Schiller(7) also found only mild deficits in color detection and discrimination in monke ys with selective lesions of V4. Thus, while V4 plays some role in analysis of color information, recent studies of lesioned animals imply that it is not as specialized or as important a center for color as was once thought.
Lesions studies sugg est other roles for V4
Work in the early 1980s by Mishkin and Ungerleider identified two major streams of visual processing by lesioning large areas of the temporal or parietal lobes(10). They found that the temporal lobe is crucially invo lved in the discrimination of form, while the parietal lobe is necessary to determine the spatial relations of stimuli. V4 is an early stop in the pathways leading to the temporal lobe, and thus the role of V4 in the processing of form information came u nder scrutiny(11).
Desimone and Schein(12) found that single cells in V4 responded to oriented lines placed within their receptive fields, and that these cells were tuned to certain orientations and sizes. Desimone and coworkers(11) also reported som e responses to stimulus form similar to that of complex cells of V1. This indicates that selectivity for stimulus form is present in some V4 cells. However, what this selectivity implies about the function of V4 is unclear. Issues like stimulus salience confuse the matter even further: Haenny and Schiller(13) found that the orientation selectivity of V4 cells narrowed considerably and responded more vigorously when recognition of the stimulus led to a reward, for example.
More direct evidence that V4 is involved in discrimination of form has come from several studies conducted on animals with whole or partial lesions of this area. Heywood and Cowey(6), studying monkeys with complete bilateral ablations of V4, found moderate deficits when animals ha d to discriminate between simple geometric shapes (a cross versus a square), between horizontally and vertically oriented gratings, and between gray-scale photos of faces from two monkey species. Walsh and coworkers(14), using methods similar to those of Heywood and Cowey, found severe deficits in V4 lesioned animals in discriminating between identical stimuli with different orientations.
In an extensive study, Schiller(7) tested a wide array of visual abilities in animals with partial lesions of V4, controlling the animals' eye movements so as to present stimuli within or without the area affected by the lesion. Schiller found mild to moderate deficits in discrimination of different frequency checkerboard patterns and in discrimination of simple geo metric shapes. No deficit was observed for detection of areas of differing texture. This all suggests that V4 plays some role in discrimination between forms, but that it is not crucial to these tasks.
To better understand V4's role, perhaps one shou ld ask the question more narrowly: are there certain conditions under which V4 is crucially important for shape discrimination? Are there specific types of visual information that are processed by V4 in discriminating shapes and others that are not? Schi ller7 serendipitously found one intriguing answer to these questions, though more may exist. Monkeys with V4 lesions had severe deficits in selecting a stimulus from a group of similar distractors when that stimulus was somehow less prominent than the d istractors. This "lesser than" deficit affected stimuli that were dimmer or smaller than the distractors. The deficit was observed even when stimuli were made visible by a variety of different means: luminance information only, chrominance only, motion only, or stereo only. The animals had no deficit in detecting identical stimuli when the stimuli were presented singly. What results from this observation is that V4 plays a role in shape discrimination, but that its role depends at least in part on con text.
Schiller suggests the intriguing idea that one function of V4 is to detect "camouflaged" objects that elicit less activity, and hence are possibly not focused upon, in other parts of the visual system. This ability to defeat camouflage by going beyond the evolutionarily older response of orienting to a prominent stimulus could enhance an animal's fitness, and so could have been evolutionarily selected. Regardless of its origins, the dimension of prominence is an unexpected, yet clearly useful, dimension along which the visual system breaks down a scene. A result such as this one shows the power of area-specific lesions to expose the division of labor in cortical visual analysis.
Lesion studies in a larger context
Iden tifying in what ways processing is segregated is an important part of our continuing effort to uncover the axes along which the brain breaks down visual information. Lesion studies allow us to determine what some of the subtasks of visual processing are , and so what some of these axes may be. As noted above, determination of these axes can reveal possible evolutionary pressures in the formation of the primate visual system. Knowing which features are processed explicitly also helps define what informa tional building blocks are used in the formation of a percept.
If we are ever to comprehend the computations performed in the brain's analysis of the visual scene, we must first determine what these computations are and which dimensions of the visual information present they are using as inputs. The identification of segregated, relatively simple processing tasks, and the kinds of information important to those tasks, will begin to define those computations.
Ultimately, we would like to understan d what computations are being done, what algorithms are being used, and how the brain implements those algorithms. The power of studying the effects of selective lesions is that it allows one to break down the vast computation known as "vision" into smal ler, less ill-defined computations such as "use motion cues to discriminate shapes" (which, admittedly, is still quite vague), and to localize such computations anatomically. That is, this approach can reveal what computations are being done, while avoid ing the much harder problem of how these computations are implemented. Yet this may also be its downfall. We may learn much about what the visual system is doing, and yet be unable to generate a useful model, because without knowing how the computations are implemented, we may not be able to understand how they relate. Perhaps, like naturalists before Darwin, we are merely collecting and cataloging observations that we will be unable to explain and connect until the next great conceptual leap is made. This leap may come with an understanding of how individual neurons encode and process information, and how neural structures larger than individual cells (but smaller than the square millimeters of cortex removed in lesion studies) compute. Currently the re are few intellectual constructs within these approximately six orders of magnitude(15). More are needed.
In the nineteenth century, Herman von Helmholtz was able to theorize that there must be three color receptors by studying only the extent and limitations of human visual performance, lacking, as he clearly states, any "anatomical basis for this theory of colors"(16). Later, when three cone types were discovered, his model was used to relate structure to function. Similarly, to understand how visual information processing is implemented in the cortex, we must know much about what computations the processing must support. Perhaps Helmholtz' analysis was possible only because the system he was studying had relatively few "moving" parts. But I would like to think that our improving knowledge of information processing and of biological systems will allow us to someday make as clever and useful an intellectual leap as he did.