by Igor SHEVELYOV, Academy Member, Head, Laboratory of Physiology of the Sensory Systems, Institute of Higher Nervous Activity and Neurophysiology, Russian Academy of Sciences

Recognition of visual images is one of the most complex functions of the brain. How and why does it differentiate between the images projected on the retina of the eye? Significant progress in understanding of these operations was made in recent years, though much more things still remain unknown. However, due to new methods of investigation we now can see the contours of a harmonious system of analysis of image signs and their synthesis in a single visual image.

What do we know about the neuronal mechanisms of the object shape perception in different compartments of the visual system? Let us start from the retina of the eye, whose structure and functions are studied most amply; vision is impossible without it. But aren't the previous concepts on its complexity and significance of its functions, linked with visual information processing, and not reception*, exaggerated? Let us take, the traditional assumption "the retina is part of the brain presented at the periphery". Is it so? Comparison of the signal processing and transformation in the retina and in the exterior geniculate body (subcortical visual center),

* Reception is perception of the stimulant's energy by receptors and its transformation into nervous stimulation. - Ed.

стр. 19

In each of the three blocks: upper series: images presented to the cat's eye; lower series: computer-reconstructed relief of neuronal activity in the external geniculate body for charges of 177 neurons.

Scheme of horizontal inhibition (A) and its effect (B).

on the one hand, and in the brain cortex, on the other, demonstrates their obvious difference.

It is well known that transduction (photoenergy transformation into nervous stimulation energy, paralleled by tremendous amplification of the signal) takes place in photoreceptors (rods and cones). This signal is transferred into the neuronal layers of the retina, where it undergoes a series of spatial and time transformations, the most important of which are adaptation restructuring of the receptive fields of bipolar and ganglionar* cells, regulated by two horizontal inhibition systems (mind that the above receptive fields of the sensory neurons are a totality of receptors sending direct or indirect signals to it). Importantly, the ganglionar cells provide an appreciable part of rapid visual adaptation of the eye to changing basal illumination. When the light is bright, these round receptive fields maximally shrink due to the inhibitory ring, providing high spatial resolution of the eye, while in the dusk they sharply dilate. Naturally, this reduces the visual acuity, but increases the photosensitivity of the system in general.

However, let us discuss the spatial transformations of nerve signals, most important for recognition of the shape of visual objects. What is the role of the retina in this process? Here it is responsible for just two, though very important, operations. The provision of nervous phase of adaptation is briefly described above; the "underlining of contours" deserves a detailed description. It is based on mutual inhibition between the neighboring nerve elements of the eye, detected in the 1930s by the American physiologist Holden Hartline in a Xiphosura (sea crustacean). This scientist won the Nobel Prize of 1944 for discovery of the universal mechanism of distinguishing the most significant communications from the nerve signal flow. These communications include those informing the brain about changes in the visual space. And really, is the information about unchanging (even) sites of objects worthy of note? If it were so, all computing reserves of the brain, the entire memory would be soon over-saturated and it would be unable to function normally. But the Nature found a brilliant simple and effective solution to the problem.

If an element of the visual system is stimulated by light, it inhibits the neighbors, the more it is stimulated and the closer are the neighbors, the stronger. As a result, the contrast between the cell charges is great at the end of activated cell chain: the stimulated terminal element responds to the light more intensely than its neighbors, which are in the same state, while the neighbor which is in darkness responds weaker than those not stimulated. If

* Retinal bipolar cells connect its outer and inner synaptic layers. They transfer information from photoreceptors to ganglionar cells (retinal nervous cells transmitting signals into the subcortical visual center). - Ed.

стр. 20

Cat's visual cortex neuron sensitivity to orientation (A - D) of a light strip. Right: pulse response of cell to emergence of a strip.

Left: stimulatory reliefs of receptive fields of the cat's visual cortex neuron at different levels of light background. Right: inhibition in the same receptive field. PA: optical, MA: transitional, and DA: dusk adaptation.

this operation is repeated at several successive levels of the visual system, the brain receives general information about the image contours.

It is filtration, in other words, important communications are selectively distinguished at the expense of suppression of less significant ones. As a result, mainly the data on changes in the environment are transmitted to the brain from the eye, while the "excessive" information is suppressed. Mutual inhibition also selectively discriminates between signals about changes in time. Hence, the brain is informed mainly about new events, while the information about continuing stimulation, in other words, about static "picture", is rapidly blocked. But then, how can we see the visual world continuously? Evolution (or the Designer General - as you like) found a way to overcome this difficulty by making the eyeballs move and tremble permanently. Their tremor, jumps, and drive modify the brightness distribution on the retina, activating the neurons suppressed by inhibition.

These actions are duplicated and amplified at the next (after the retina) level of the visual system - in the external geniculate body. It is noteworthy that no appreciable neuronal operations for analysis of the image shape are realized either in the retina, or in the external geniculate body. This "preprocessing" is rather a series of important preparatory steps for further analysis of the signals in the cortex. In 1999 a group of American scientists, fixing the activities of 177 neurons in the external geniculate body, obtained a good illustration of this process. It was found that the neuronal activity relief in it virtually reproduced the images presented to the eye. This experiment confirmed once more that the retinal and external geniculate body neurons with round or concentrical receptive fields realized just the so-called punctate description of the image, including no analysis of its shape.

Axons* of the external geniculate body neurons are "stretched" to the primary projection zone of the visual cortex. The first basic operations on analysis of the objects' shape-detection of their essential signs - take place here. One of them is orientation of fragments of lines or strips, constituting the elements of the image contours. A cortical neuron can be sensitive only to, for example, a vertical strip. What is the structure of these detectors? It is one of the priority problems for a scientist. Vision physiologists David Hubel and Torsten Wiesel (USA), Nobel Prize winners of 1981, put forward a so-called convergence hypothesis, which became classical and remained virtually not modified up to the present time. According to this hypothesis, a cortical neuron detects a certain orientation of a light or dark strip due to convergence, i. e. delivery of signals from a group of neurons in the external geniculate body, arranged in a line, with the same orientation of their receptive fields in the visual field.

The first contradiction in this seemingly simple and elegant concept was detected at the beginning of the 1970s by groups of scientists in Moscow and Boston during

* Axon is a neuronal process conducting the nerve pulses from cell body to innervated organs or other nerve cells. - Ed.

стр. 21

Time course of receptive field of the cat's visual cortex neuron. 1 - 6: successive time sections of receptive field with a 20-msec step.

Time course of orientation set-up of visual cortex neuron. A: forms of responses to strips of different orientation; B: schemes of set-up to orientation in successive time fragments of responses at 10-msec step; C: the same as B, but for the entire response; D: dependence of orientation preferred by the neuron upon the time after the stimulus beginning.

modification of the visual system adaptation level. Our experiments showed that with reduction of the light background the cortical receptive fields were sharply extended, while the neurons either lost or drastically deteriorated their detector characteristics.

These data cannot be explained by the above scheme but correspond to a well-known adaptation restructuring of round receptive fields in the retinal ganglionar cells. The difference consists just in the shape of the cortical receptive fields: they are not concentrical, but elongated; but the philosophy of their inhibitory modification by the respective cortical cells ("inhibitory modeling", according to the formulation of an Australian neurophysiologist, Nobel Prize winner of 1963 John Eccles) seemed to explain it fairly well. It removed the contradictions of the convergence hypothesis and was in good agreement with the properties of the neurons realizing the intracortical inhibition. Detection of sharp restructuring of the cortical receptive fields during alteration of the level of consciousness (sleep-calm consciousness-active consciousness) became an additional proof of the correctness of this mechanism and its functional significance. All this indicates high adaptive potential and pliability of the cortical neuronal systems involved in analysis of essential signs of image. Restructuring of the receptive fields during readaptation to the background or level of consciousness is usually realized within fractures of a second.

The natural question is: are the detector characteristics of these neurons as rapid? It has been found that if the setup to the line orientation were evaluated not by the number of pulses in the entire response, but by its successive time fragments, clear-cut changes in it would be seen in two-thirds of cells. At the beginning of response this cortical neuron detects a strip of vertical orientation (90°) and then gradually (within 120 msec) moves to the horizontal line (0°). Hence, we have proven that the visual cortical neurons are not only highly adaptive, but also dynamic filters of the visual image signs.

So far we discussed the fact that cortical neurons detect the orientation of fragments of the line from which any visual image largely consists, after the contour distinguishing operation. After discovery of detectors it was assumed for a long time that they were the only workers in the primary visual cortex. However, at the beginning of the 1990s our scientists and later a British physiologist Adam Sillito discovered cortical detector neurons set up not for the line orientation, but at line crossing. We called them "plus-figure detectors". They constitute about half of neurons in the primary visual cortex, react to the cross more rapidly and, on the average, three-fold more intensely than to a strip and are located mainly in the middle layers of the cortex. In addition to the plus detectors, we found angle and Y-figure detectors. Importantly, about 70 percent of neurons sensitive to these figures are selective, that

стр. 22

Optical map of a small site of the cat's visual cortex. Cortical sites activated during presentation of strips of different orientation (on the right) are colored; dark points show the zones of activation by plus-figures.

Selective (A) and invariant (B) sensitivity of two neurons of the cat's visual cortex to the strip orientation and to the shape and orientation of a plus-figure. Vertical line: the value of neuronal pulse response to a respective stimulus.

is, set up to certain orientations and shapes, while the rest are invariant (similarly sensitive to figures of any shape or orientation).

It was also found that for a quarter of neurons the sensitivity to figures is determined by convergence of signals from two detectors with different orientation set-up. For the rest of neurons the sensitivity to a figure is determined by the presence of two zones in their receptive field: the so-called "front" inhibitory and lateral disinhibitory. The former reduces the neuronal response to a sufficiently long strip, the latter blocks this inhibition, if the disinhibitory zone is activated by the second line of the plus. Simulation experiments showed that by modulating the width and weight of these two zones it is possible to attain both selectivity and different types of invariant set-up for a figure.

Detectors with similar preference are united ("packed") in "orientation columns" in the cortical representation of each local site of visual field, penetrating through the entire thickness of the cortex from its surface to the white matter, consisting of numerous nerve fibers. A set of these columns, responding to all orientations in a certain site of visual field, constitutes a "supercolumn" - a neuron modulus of a higher order. What is the place of line crossing detectors in this orderly structure? In order to answer this question, we carried out optical mapping of the primary visual cortex by a new method for optical signal recording, depending on the hemoglobin/deoxyhemoglobin* ratio in this or that site of nerve tissue, that is, on the degree of its activation. It was found that the cortical activation zones are cross-overlapped by the map of orientation columns. This means that line crossing detectors are incorporated in the latter.

When a neurophysiologist describes this or that characteristic of neurons, he always has to evaluate the real behavioral significance of the studied effect. The same is true for detection of plus-figures by cortical neurons. Psychophysiological experiments were carried out on volunteers, who were to recognize geometrical figures lacking fragments of sides or angles. All volunteers significantly better recognized the former than the latter. Hence, the activity of angle detector neurons is of higher behavioral significance for recognition of a figure than that of the line orientation detectors.

We spoke here about the characteristics of neuronal systems of primary projection region of the cortex, detecting the individual signs of a visual image. However, it is just one compartment of the cortical part of the visual system; analysis of signs and their synthesis into a single neuronal presentation of the image is carried out in many zones (up to 36 in humans). Usually two main cortical flows of information transmission and processing are distinguished. The first of them is called "ventral" and includes

* Deoxyhemoglobin is hemoglobin devoid of oxygen. - Ed.

стр. 23

Dependence of correct response of geometrical figures by a man upon the sides (light circles) and angles (dark circles) being masked to a different degree.

Scheme of dorsal and ventral visual cortical route in monkey brain.

regions V1, V2, V4, and IT (inferior temporal cortex) and provides a response to the question "what do we see?", that is, recognition of the object shape. The second, "dorsal" route, includes regions V1, V2, MT, and the parietal cortex. It produces an answer to the question "where is the object?", the ratio of several objects in a visual field is evaluated, and the entire scene is analyzed. Transition to successive levels of each of the two above flows is associated with complication and consolidation of the neuron-detected image signs as a result of their unification.

Let us omit the details of the processes and discuss the data and concepts on the final stage of neuronal reflection of the visual objects' shape in the inferior temporal cortex. The data were obtained in recent years as a result of pioneer investigations of the Japanese and British scientists using the traditional (microelectrode) and two novel methods of the cortex neurovisualization or mapping. It was shown that individual small locuses (areas), differently located in the inferior temporal cortex of the monkey, selectively responded to known and unknown faces, their aspect, buildings, and other objects. This information was supplemented by neurological data on cases of extremely selective prosopagnosia* (for example, loss of capacity to differentiate between car types in a car expert after a local brain stroke).

The new data are principally important, as they indicate the positional neuronal coding of integral visual images in the inferior temporal cortex. Here we must remember the hypothesis of Jerzy Konorski, a well-known Polish neuro-physiologist, who postulated the existence of "superneurons" responsible for reception of integral images, as early as in the 1950s. These cells were ironically called "my grandma's detectors" for a long time, but this skepticism is out of place in the light of modern data on the inferior temporal cortex neurons.

One critically important circumstance impedes the understanding of neuronal operations of the visual image recognition. We mean synchronization of activities of neurons responsible for detection of different signs of an image. Their responses should be somehow united (the binding or fixation effect), though they can be often located in the cortex at an appreciable distance from each other. The probability of their charges synchronization at rather high (40 - 70 pulse/sec) frequencies is usually regarded as a respective mechanism.

However, we should remember about the significance of rather low frequency wave processes for this effect (for example, 10 Hz reflected by the electroencephalogram alpha rhythm), demonstrated for the first time by Mikhail Livanov, Russian physiologist, Academy Member, in the 1950s - 1970s. Previously (in 1947) Walter Pitts, mathematician, and Warren MacCullogh, neurologist (USA) splendidly predicted that the electroencephalogram alpha rhythm reflects the wave process, periodically moving along the visual cortex and providing successive reading of sensory information from it for its subsequent transmission to higher visual centers. For many years we had indirect evidence of this wave process movement in the cortex, and recently we obtained a direct evidence. We showed that the alpha rhythm current dipole successively shifts in human brain visual cortex with every alpha wave.

We shall note in conclusion that the amazing harmony and perfection of brain structures and functional mechanisms responsible for visual perception cannot but wake the enthusiasm and eagerness of a scientist working in this rapidly developing sphere of modern science.

* Prosopagnosia is a neurological disorder associated with failure to recognize previously known faces, objects. - Ed.