Feature Extraction By The Frogs Retina 621 Early Work

Back in 1938, Hartline reported on results obtained recording from single frog optic nerve fibers (GC axons) when the eye was stimulated with simple spots of light. In this pioneering work, Hartline coined the term receptive field (RF) to describe the area of the retina over which any change in illumination (ON, OFF, and ON/OFF) would affect the base rate firing of the optic nerve fiber under study. A GC RF was often found to be surrounded with an annular peripheral region in which changes in illumination affected the firing caused by illumination in the center of the RF. Hartline found three types of RF in the frog optic nerve: ON, ON/OFF, and OFF. If a small spot of light appears in the RF of an ON fiber (a pulse of light intensity), the instantaneous spike frequency increases to a peak, then falls off to a lower, steady-state value. At off, the discharge is abolished. The response of an ON/OFF fiber to a pulse of light in its RF is a burst at ON, falling off to zero frequency, then another short burst at OFF. OFF units fire briskly at OFF; then their frequency decreases slowly in the dark. ON silences the OFF unit.

In 1953 Barlow added to Hartline's (1938) observations. He noted that OFF cells have adding RFs; i.e., the response to OFF occurs at both center and periphery of the RF. The effect of turning off the light in the periphery adds to the effect of reducing the light in the center of the RF, with a weight decreasing with radial distance from the center. He found that the ON/OFF cells have differencing RFs, where a discharge caused by ON at the center is diminished by a simultaneous ON in the periphery. The same reduction happens for OFF in the center plus OFF in the periphery. No results were given for ON cells.

Thus, the early work of Hartline and Barlow suggested that each optic nerve fiber (and its associated RF) sends to the frog's brain basic information about where dimming has occurred on the retina (OFF operation); ON/OFF fibers signal where the contrasting boundaries are moving, or where local inequalities of contrast are forming. The ON fibers signal where brightening has occurred. If there is no change in time of a contrasting image projected on the retina, the three types of fibers firing rates will decrease to low, basic levels.

In 1959, Lettvin and colleagues published the results of a pioneering neurophys-iological study on visual feature extraction, entitled, "What the Frog's Eye Tells the Frog's Brain." Frog GC responses were examined when the eye was stimulated with simple, contrasting, two-dimensional objects presented on the inside surface of a 14-in.-diameter, matte gray hemisphere centered over the eye. The thrust of their research was to find natural "features" of simple objects that elicited maximum GC responses, rather than simply to shine lights into the frog's eye.

What Lettvin et al., (1959) demonstrated is that the frog's retina can send more complex information to the brain than simple ON, ON/OFF, and OFF operations on RFs. Instead of using spots of light as stimuli, Lettvin et al., used simple, contrasting, geometric objects such as a 1°-diameter black spot, and a 12° x 30° black rectangle moved with magnets on the inside of the 14-in.-diameter, gray hemisphere centered over the eye. They were able to identify four consistent, separate operations on images projected on the frog's retina. These operations were found to be substantially independent of the overall illumination. The operations were named (1) sustained contrast detection, (2) net convexity detection, (3) moving edge detection, and (4) net dimming detection. The details of these operations are described below:

1. Sustained contrast detection (SCD): The axons of these ganglion cells are unmyelinated. They are probably Hartline and Barlow's ON fibers. The SCD fibers do not respond to ON or OFF of general illumination. The SCD unit RFs are from 2° to 4° in diameter. These units fire when a contrasting, 1° or 3°-diameter black disk is moved into the RF. Apparently the SCD fibers are directionally sensitive; i.e., they have a preferred direction that gives maximum response, and if motion is in the opposite direction (antipreferred direction), firing is suppressed. Curiously, SCD fibers have "memory." That is, if a spot moves into the RF in the preferred direction, the unit fires. It continues to fire (at a reduced rate) if the spot stops in the RF. If general illumination is turned off, the firing ceases, or in some units is greatly reduced. When the general illumination is again turned on, and the spot remains stationary in the RF, the unit again begins to fire, demonstrating "memory." It was claimed that SCD fibers responded to white as well as black test objects. They also responded to linear edges (as on the 12 x 30° black rectangle). Unfortunately, no quantitative information was given on directional preferences or optimum object speeds in the paper.

2. Net convexity detection (NCD): NCD fibers are also unmyelinated. They, too, are unresponsive to ON and OFF of general illumination. The RFs are from 3° to 7° in diameter. NCD units respond only to dark contrasting spots moved into the RF. If the spot is stopped in the RF, the unit continues to fire. If general illumination is turned off, the unit is quiet. At ON, the unit continues to be silent until the spot is again moved. There appears to be no memory. There appears to be a broad size optimality for response to spots moved into the RF. Responses were noted down to spot diameters of 0.05°. When the spot is about one half the diameter of the RF, a maximum response is noted. Response begins to fall off for spots less that 1° diameter or greater than half the RF diameter. There is no response to long, moving, dark edges that overlap the RF. Also, a large, black and white checkerboard with a repeat distance of one half the RF diameter gave little, if any, response when moved over the RF. Jerky motions appeared to be more effective in eliciting spikes from moving spots than smooth motions. No quantitative data were given on object speed or directional preference; presumably NCD units respond to object motion in any direction.

It is tempting to call NCD units the frog's "fly detectors"; perhaps they send information to the CNS alerting the frog that prey may be near; get ready to strike.

3. Moving edge detection (MED): These are myelinated fibers conducting at about 2 m/s. The RFs of MEDs are about 12° diameter. They are the same as Barlow's ON/OFF units. They respond to long contrasting edges moving through their RF. Their frequency is proportional to edge speed, and is said to fall off at high speeds. No quantitative information was given on directional sensitivity, if any, or the speed that gave maximum firing frequency. The MED fibers project into the third layer of the tectum.

4. Net dimming detection (NDD): NDD unit fibers are myelinated and conduct at 10 m/s. They are the same as Hartline and Barlow's OFF fibers. They have large, about 15°, RFs; they respond to OFF or general dimming by a prolonged burst. The amount of light at ON required to interrupt the OFF burst gets less and less the longer the eye sits in the dark. If the general illumination is dimmed, firing occurs. Now if a dark object is moved through the RF, the firing is suppressed. Lettvin et al., thought this effect was due to the relative brightening in the RF as the black object passes through it.

To summarize the operations performed by the frog's retina, we quote Lettvin et al.:

Let us compress all of these findings in the following description. Consider that we have four [ganglion cell] fibers, one from each group, which are concentric in their receptive fields. Suppose that an object is moved about in this concentric array:

1) The contrast detector tells, in the smallest area of all, the presence of a sharp boundary, moving or still, with much or little contrast.

2) The convexity detector informs us in a somewhat larger area whether or not the object has a [sharply] curved boundary, if it is darker than the background and moving on it; it remembers the object when it has stopped, providing the boundary lie totally within that area and is sharp; it shows most activity if the enclosed object moves intermittently with respect to a background. The memory of the object is abolished if a shadow obscures the object for a moment.

3) The moving edge detector tells whether or not there is a moving [straight] boundary in a yet larger area within the field.

4) The dimming detector tells us how much dimming occurs in the largest area, weighted by distance from the center [of the RF] and by how fast it happens.

All of the operations are independent of general illumination. There are 30 times as many of the first two detectors as of the last two, and the sensitivity to sharpness of edge or increments of movement in the first two are higher than in the last two.

It is interesting to note that the shape of the dendritic tree structure of a frog GC can be associated with the feature extraction operation performed. The GCs of edge detectors (also known as boundary or sustained contrast detectors) have constricted dendritic fields (<100 |im wide); convex edge detectors (also known as moving spot detectors) have "E-tree" dendritic fields (~200 |im wide); moving contrast detector GCs (also known as moving edge detectors) have "H-tree" dendritic fields (~300 |im wide); and dimness detector (also known as dimming detector) GCs have broad dendritic fields (> 500 |im wide). Function and form are intimately associated in the frog retina not only for GCs, but also for HCs, amacrine cells, and BCs.

Was this article helpful?

0 0
Peripheral Neuropathy Natural Treatment Options

Peripheral Neuropathy Natural Treatment Options

This guide will help millions of people understand this condition so that they can take control of their lives and make informed decisions. The ebook covers information on a vast number of different types of neuropathy. In addition, it will be a useful resource for their families, caregivers, and health care providers.

Get My Free Ebook

Post a comment