Perception Lecture Notes: The Retina

Professor David Heeger

What you should know from this lecture

The retina is the most extensively studied piece of the central nervous system.

Purkinje tree

Light comes in through the pupil, focused by cornea and lens onto the retina, passes through the aqueous and vitrious humors and several layers of neural tissue (ganglion cells, bipolar cells) before it reaches the photoreceptors. All this stuff is mostly transparent so that light passes right through.

Apart from the neural tissue between the lens and the light receptors of your eye, there is also a bunch of tiny little blood vessels that branch all over the surface of the retina and provide it with food and nourishment. The pattern of blood vessels is called the Purkinje tree (after the Bohemian physiologist who first described it). The Purkinje tree branches out from the optic disk (blind spot) where the blood supply comes in and the optic nerve goes out. Notice also that there are few, if any, blood vessels in the the fovea, the center of your visual field.

Why don't you see the blood vessels when you look into somebody's eyes? If the light source is behind you, so that your head is between the light source and the eye you are studying, then your head will cast a shadow that interferes with the light from the point source arriving at your friend's eye. As a result, when you look in to measure the retinal image you see nothing beyond what is in your heart. If you move to the side of the light path, the image at the back of your friend's eye will be reflected towards the light source, following a reversible path. Since you are now on the side, out of the path of the light source, no light will be sent towards your eye.

Helmholtz built the first ophthalmoscope. One needs to arrange a light path using half-silvered mirrors so that the examiner's eye does not cast a shadow. A bright source of light is required since the back of the human eye is not very reflective.

Stabilized vision: Why can't you see your own blood vessels in normal everyday vision? From what I've described to you, they are directly in the light path between the lens and the retina. So shouldn't you be seeing the shadow cast by these blood vessels on the surface of your retina? The reason you can't see thse blood vessels reveals a very important phenomenon about the way your visual system works. Notice that these blood vessels are fixed with respect to your retina. That is, as you move your eye about, since the blood vessels are attached to your retina, they move about with it, forming a fixed relationship. The pattern of retinal blood vessels does not change as you move your eye. Such a pattern - one that is fixed with respect to your retina even as the position of your eye changes - is called a stabilized image. The surprising fact about your eye is that it is insensitive to stabilized images.

Special devices have been built that stabilize images. These devices are used for doing certain kinds of vision experiments that require careful control over eye movements. The device works by sensing the position of your eye. Each time you move your eye (even a little bit), it moves the visual stimulus by the same amount. In this way, the visual stimulus is projected onto your retina in such a way that even if you move your eyes the stimulus remains fixed with respect to your retina. Under such viewing conditions - called stabilized viewing conditions - the image of the world that is projected onto your retina does not make it to your conscious awareness. After a few seconds of stabilization, the entire scene seems to fade and disappear. The stabilized scene of the world disappears just like the shadows of your blood vessels disappear.

Stabilized image demo

One can't really demonstrate this in class without one of these expensive eye-trackers, but we can get close. Above is a small dot with a blurry gray circle around it. Since the transition is very gradual (from gray circle to white background), if you hold your eyes as steady as possible while fixating the central dot, the boundary of the blurry gray circle seems to disappear.

Diseases of the Retina

Diabetic retinopathy:  swelling and abnormal growth of capillaries, insufficient oxygen supply to the retina, may cause bleeding into the vitreous.  Treated by using a laser to seal off blood vessels.

Detached retina:  the retina separates from the pigment epithelium, scotoma (blind spot) in the area that detaches.  Caused by glaucoma or by injury.  Use laser, heat, or cold probe to cause scar tissue to form and hold the retina in place.

Macular degeneration: degeneration of the part of the retina called the macula that includes the fovea and surrounding region.  Leads to loss of central vision.  The most common form is age related.

Retinitis pigmentosa: a hereditary disease, initially affects rods and peripheral vision, then gradually starts to affect cones and eventually leads to complete blindness.

Retinal implants:  Because there are no good treatments for some retinal diseases (e.g., retinitis pigmentosa) there is interest in developing technology that would help alleviate these diseases.  Some ideas involve redirecting the image to a healthy region of the retina. For example, one might have a mini video camera and a mini flat-panel display all mounted on eye glasses, with a computer to process the incoming images and special optics for projecting the processed images into the eye to a healthy region of the retina.  This is not so far in the future.

Conceptual design for a retinal implant

Other plans involve implanting a microchip with a digital camera on one side and stimulating electrodes on the other side, analogous to a cochlear implant.  Obviously, it is very difficult to properly stimulate a large number of optic nerve fibers so it will be quite a number of years before this will be possible.

At present, only one vision-enhancement system that exploits advanced optoelectronics is on the market: the Low Vision Enhancement System (LVES), developed in the Wilmer Eye Institute at the Johns Hopkins University School of Medicine. This system provides variable focus and contrast rather than neuronal stimulation.

A company called Second Sight was founded in 1998 to create a retinal prosthesis to provide sight to patients blinded from retinal diseases. They have a prototype, experimental device that has been tested in a few individuals.

This 62 year old man, who has been blind since the age of 36, has been provided with limited vision by using a tiny camera wired directly to the primary visual cortex of his brain.  The reports are that he can read large letters and navigate around big objects. He does not see an image. Rather, he perceives up to 100 specks of light that appear and disappear, like stars that come and go behind passing clouds, as his field of vision shifts.

Current research is aimed at extending the success of these experimental systems, to make them as successful and ubiquitous as cochlear implants.  For example, there was a recent conference on Replacement Parts for the Brain.  This conference brought together leading researchers throughout the country who focus on one of the newest frontiers of neuroscientific and bioengineering research: the intracranial implantation of computer chip models of brain function as neural prosthetics to replace damaged or dysfunctional brain tissue.

Further information about research and development of retinal implants.

Meanwhile, good electronic resources, links, and texts relating to technology and services for the blind can be found at The American Foundation for the Blind and at The Royal National Institute for the Blind.

Retinal Circuitry (Parallel Pathways)

The retina is a highly organized, layered structure. The information begins at the receptors and flows through a second layer of cells, called bipolar cells, and then continues through a third layer of cells, called ganglion cells. In addition, at the level of the receptor to bipolar connections, there are cells called horizontal cells that connect from receptors to other receptors. And at the level of interconnections between the bipolar cells and the ganglion cells there are cells called amacrine cells that take as their input signals from bipolar cells and feed their output back onto other bipolar cells.

In the living eye, the neurons in the retina are all quite transparent so as not to interfere with the light on its way to the photoreceptors. In this picture, the neurons are stained so that you can see them. From bottom to top there are: clear ganglion cell axons that make up the optic nerve, stained ganglion cell bodies, bipolar cells, photoreceptors, and the black layer of pigment epithelium (special skin cells that absorb stray light and that help in the process of regenerating bleached photopigment).

None of the peripheral cells in the retina - the receptors, horizontal cells, bipolars, or amacrine cells - generate action poentials. Horizontal cells and amacrine cells don't even have proper axons. Rather these neurons respond with graded potentials. If the resulting current is large enough at their synapses, then they release neurotransmitter. It is only when you get to the level of the ganglion cells - with axons that reach a rather long distance into the brain - that you get action potentials.

A big theme in sensory neuroscience is that information is split apart and processed by separate subparts of the visual system. We have already seen a good example of parallel pathways: rods and cones. Typical properties of parallel pathways include:

  1. Physiologically/functionally distinct. Rods give you very high sensitivity so you can see at night. But if rods were all you had, you would be in trouble. A very small percentage of people, called rod monochromats, have no cones. They see a wash of brightness during daylight conditions and are functionally blind at high light levels. So both the rods and the cones are needed.
  2. Anatomically distinct. Rods and cones have different shapes. For many parallel pathways, neurons in the separate streams are separated from one another, e.g., localized in different layers. This is not true of rod vs cones, but is true of many other parallel pathways.
  3. Complete coverage (or nearly complete coverage) of the visual field by each stream. There are no rods in central fovea but otherwise the rods cover entire visual field. Likewise, the cone mosaic covers the entire visual field.
  4. Recombine: parallel streams often (but not always) remerge eventually. Rod-cone pathways recombine at the ganglion cells. We need the different sensitivities of the rods and cones at the very front-end of the visual system but once the light has been transduced (with proper sensitivity) into a neural signal, there's no need to keep the separate processing.

Midget, parasol, and bistratified retinal ganglion cells: Another example of parallel pathways. I'll discuss the midget and parasol cells here. The small bistratified retinal ganglion cells play a special role in color vision (see color lecture).

Anatomically distinct: Parasol dendritic trees are big, combine inputs from many cones. Midget trees are small. In the fovea, midget ganglion cells receive synapses from only one bipolar cell each which in turn receive synapses from only one cone. In the periphery, the dendritic fields of both ganglion cell types get bigger. It is easy to learn how to distinguish the two types. You need to pay attention to: (1) where you are in the retina, (2) how big is the tree.

Complete coverage: A. Tracings of dendrites of 42 midget ganglion cells intracellularly stained. B. Cell body positions, regular distribution. C. Tracings around dendritic trees, perfect coverage with no overlap.

Recombine: Midget and parasol streams stay segregated in lateral geniculate nucleus (LGN), recombine to a large extent in visual cortex.

Physiologically distinct:

Parasol ganglion cells (magnocellular LGN cells):

Midget ganglion cells (parvocellular LGN cells): Magnocellular pathway and dyslexia:  There is a hypothesis that dyslexia (reading disability) is associated with a deficit in the magnocellular pathway. A graduate student in Prof. Heeger's lab used fMRI to test this hypothesis. A specially-designed visual stimulus (fast moving and dim light levels) was used to isolate the magnocellular pathway to visual cortex. Brain activity in visual cortex (using fMRI) and reading rate (using a speed-accuracy-trade-off test) were measured. Results: (1) there were weaker magnocellular pathway responses in dyslexic subjects; (2) one can predict reading rate from the strength of the magnoccellular pathway responses. Conclusion: dyslexia is associated with a deficit in the magnocellular pathway, a pathway that is critical for fast visual processing. This idea is analogous to the hypothesis that language learning disabilities are caused by a deficit in fast auditory processing (see speech lecture notes). But, we do not know if a magnocellular pathway deficit causes dyslexia. The possible link between visual processing speed and dyslexia is summarized in an American Psychological Association article.

Other parallel pathways initiated in the retina: There are about 40 ganglion cell types. About 90% of them project (send their axons to) the LGN and then on to visual cortex. The other 10% project to: superior colliculus in midbrain (involved in control of eye movements), pretectum in midbrain (pupil size, focus), superchiasmatic nucleus above optic chiasm (circadian rhythms).

Sound localization revisited (another example of a parallel pathway): We have seen an example of parallel pathways in audition as well.  The two parts of the superior olive (MSO and LSO) are involved in sound localization.  These pathways satisfy an analogous set of properties.  They are functionally distinct: one extracts inter-aural timing differences (ITD) and the other  extracts inter-aural intensity differences (IID).  They are anatomically distinct: in 2 separate nuclei within the superior olive.  There is complete coverage: in this case it's not coverage of the visual field, rather each nucleus has a full tonotopic map (covers all frequencies).  They recombine: signals from MSO and LSO about ITD and IID information is combined later in the auditory pathways.

Information Processing in the Retina

The retina performs five important jobs:
  1. transduction
  2. data compression
  3. light adaptation
  4. spatial filtering
  5. wavelength encoding
Transduction and data compression are discussed below. The others are covered in the next two lectures.

Photoreceptors and Transduction

There are roughly a hundred million photoreceptor cells in each eye. There are two types of photoreceptors, rods and cones, with different shapes. Rods are specialized for low light levels (night vision), and cones are for high light levels and color vision.

Electron microscope picture of some rods and cones followed by electron microscope picture (at higher magnification) of the folded membrane in the outer segment of a rod.

In this electron micrograph of a rod outer segment, you can see that the cellular membrane folds in and out. These internal membranes are highly structured and filled with photopigment molecules. In the rods, the photopigment is called rhodopsin and it is a complex molecule that is made up of two parts: one is called opsin and the other is a derivative of vitamin A.

When a quantum of light is absorbed by a molecule of rhodopsin, it changes the chemical state of the photopigment. The two parts of the molecule split. This change in state is called the isomerization of the photopigment. The isomerization sets off a biochemical chain reaction that eventually leads to an electrical current flowing in the rod. If the current is big enough, this neural signal is transmitted to the bipolar cells in the retina because neurotransmitter is released at the rod-bipolar synapse. Absorbing a single photon of light in a rod is enough to evoke a regular/reliable photocurrent. In fact, Hecht, Schlaer and Pirenne demonstrated psychophysically in the 40's that human subjects can reliably detect single photons.

You can see the physical consequences of the isomerization in a chemistry lab. Each vial above contains a solution of rhodopsin that has been exposed to a different amount of light. The rhodopsin pigment changes color when exposed to light, called bleaching.

You can also see this phenomena directly in the living eye of certain animals (like cats and alligators) because the back of their eyes contains a white reflective surface, called the tapetum. That's why cats eyes appear to glow so brightly when they catch the beam of a car's headlight.

Alligator with red eyes; Alligator with no red eye

The above photographs show (1) what an alligator eye looks like after keeping in the dark for 24 hours so that the photopigment was fully regenerated and (2) what it looks like when exposed to light so that most of its rhodopsin is bleached.

In the butterfly eye, you can see different colors corresponding to photoreceptors with different photopigments.

The rods all have the same photopigment, rhodopsin. But there are three different types of cones in the human retina, each with a slightly different photopigment. New optical techniques have made it possible to take pictures of the living human retina at very high magnification, and visualize the layout of the 3 different cone classes. The 3 photopigments absorb and reflect different wavelengths of light, giving rise to the different colors in the picture.

Retinal Inhomogeneity (Data Compression)

There are roughly a hundred million light receptors in each eye, but only half a million ganglion cells. This is a design problem. One can't have one ganglion cell for every photoreceptor (or the optic nerve would be bigger than your head). This means that in order to take the original image on the photoreceptors, and send it to the more central portions of the brain, a great deal of compression takes place. A large part of the compression is accomplished by having blurry vision in the periphery. Note: this isn't optical blur (the optical quality in the periphery isn't that bad), but by analyzing that retinal image using fewer receptors, fewer bipolar cells, fewer ganglion cells, etc.

These two images are simulations of the high spatial resolution in central vision coupled with the blurry low spatial resolution in the periphery.  The image on the left simulates fixating on the woman while the image on the right simulates fixating on the man.

Rods and cones are not distributed uniformly across the retina. Most of the cones are in or near the fovea, the central part of the visual field. Most of the rods are in the peripheral part of the visual field. As mentioned above, rods are used for night vision and cones subserve daylight vision. Want to see a dim star on a moonless night? Cones are worthless. You need to use your rods. There are few rods in the center. So, you need to look slightly off to the side.

This pictures shows a cross-section of the foveal retina through the layer of the photoreceptor outer segments. The cones are packed very tightly, in nearly perfect hexagons, in the fovea.

This picture shows a cross-section of the peripheral retina. The large brown circles are cones and the small yellow ones are rods. In the periphery, there are a lot more rods than cones, but the cones are big to absorb a lot of light. They are not packed in a perfect, orderly way as they are in the fovea.

Peripheral ganglion cells pool inputs from many receptors, have bigger dendritic trees, have bigger receptive fields, and there are fewer of them. This allows you to have high spatial resolution for foveal viewing, but still have some vision in the periphery, and a small optic nerve.


Copyright © 2006, Department of Psychology, New York University
David Heeger