How the retina works

Hi,

Sorry for the bit of cross-posting, I've had a couple of problems working out the best
place for this post.

I've done quite a lot of googling to find out how the retina works in simplistic terms
but
to no avail.

I have been working at Carver Mead's Silicon Retina in May 91's Scientific American and
have worked out so far:

Light hits the cones and rods, and passed on to the horizonal cells.
The horizontal cells smooths or spacially averages the signal from the rods and cones,
there is also some feedback to them.
The rods and cones along with the horizontal cells send a signal to the bipolar cells
through the triad synapse, which basically takes the difference between them.

So far the effect of this basically detects edges and motion from the original image. Is
this correct so far?

Now I can't find much information on the amacrine and ganglion cells, except that the
amacrine cells perform a similar role to the horizontal cells and the ganglion act as a
conduit to the lateral geniculate in the thalamus.

Could someone kindly expand on that a bit or point me to a site that can explain it in
layman terms?

Any help appreciated.

Wil

 

I suggest a decent med school textbook like Kandel and Schwartz. It
may help you beyond a simplistic understanding.

Kal

 

I probably should get a book like that.

What I'm trying to do is make a simulation of essentially the essence of how the various
layers work, and then to later build on the output of that work.

I'm currently at the stage of researching whether my plans make ant sense!
Basically I don't want to be spiralling off doing research on retinas at least until I
know there's merit in building a simplistic model.
To that end I currently need an overview of the various layers.

Wil

 

What you really need is a book like "The Retina: An Approachable Part
of the Brain" by John Dowling, Harvard U Press, 1987. That work is,
in fact, referenced in Mahowald and Mead's "Silicon Retina" paper.
(Incidentally, it is not polite to ignore coauthors when citing work!"

R. Masland has a brief introduction to "The Functional Architecture of
the Retina" in the Dec. 1986 Scientific American.

Just about every book you can find on neurobiology or neurophysiology
or neuroscience or physiological psychology should have a large
section, perhaps a whole chapter, on retinal information processing.
these books are really essential to get a firm understanding of the
anatomy and physiology behind the information processing.

This seems to be a nice web site on the retina
http://retina.umh.es/Webvision/

 

*blush* sorry, was being lazy!

Thanks for the leads, off down to the library for me then!

Wil

 

In that case, what you need is far more than the introductory type
things I outlined in another response.

If you want your model to be anything other than simplistic, you need
to understand what real neurons do, as opposed to "cartoon" neurons
used in most "neural" models. These are quite fine for research in
information processing, even for research in visual information or
image processing. They are not at all suitable for research in how
the eye actually works.

You need to understand the details of signal conduction down branching
dendrites, about transmitter release by graded signals, about synaptic
plasticity. You need to understand the details of microcircuit
anatomy of the retina. You need to understand the details of
neurophysiology of the retina derived from electrophysiological
studies.

In short, you need to go through some hefty neuroscience text as just
a start and then work your way into the primary literature. I
strongly recommend
Kandel, Schwartz, and Jessel,
Principles of Neural Science
McGraw-Hill, 2000

Squire et al.
Fundamental Neuroscience, 2nd ed
Academic Press 2002

Shepherd
Synaptic Organization of the Brain, 5th Ed
Oxford U. Press, 2003

You also must read
Koch
Biophysics of Computation: Information Processing in Single Neurons
Oxford U Press, 1999

 

I think you need more than an overview; you need an intimate
understanding of the processing of each of the interacting components.
By this, I mean that a 'simplistic' definition by cell or by layer
will not permit you to design a useful model.

Kal

 

Eeek!

What I'm after is something a lot more akin to Mahowald and Mead's Silicon Retina except
done in software.
Would I still need to do into such detail?

Wil

 

I think you could be right, it might be time to have a rethink!

Wil

 

That seems to be different from what you described earlier.

Do you want to model the real retina or the "Silicon retina"? There
is a big difference. To model the real retina, you must know really
how the cells work. To model the silicon retina, all you have to know
is how that works. And there, I can't help you.

 

FWIW, I'm not sure you need to go into each layer, so much as the
effective output of the ganglion cells and center-surround antagonism.
It depends on why you want to model the retina in the first place.

Even more pertinent to your work might be Marr, "Vision", Freeman, 1982.
In all your searches, add the phrase "lateral inhibition"-- its my
favorite network architecture. It retina, it serves as a spatial
differentiator, but in brainstem and spinal cord is serves a temporal
integrator. Lateral inhibition-- is there anything it can't do?

Scott

 

You are correct, I am really after something more aligned to the silicon retina, but I
also want to understand it's design in relation to a real retina and to try to understand
the myriad of amacrine cells. To be honest though my research will not be going in to PhD
detail!

I have to say though I went though the site you suggested and it is extremely good, there
is a pdf download that gives a fantastic overview of the retina and is exactly the sort of
thing I was after.

I also have a collection of papers about neurons on order it includes "The Functional
Architecture of the Retina" in the Dec. 1986 Scientific American, though I just found out
I think you can get those old papers from the sciam site, anyway definitely worth a read.

All in all thanks for your responses, they have definitely helped me refine what I am
looking for.

Wil

 

That is exactly what I am trying to model, I have more of an interest in edge detection
than in the make up of the retina in a physiological sense.

Do you know of any models I could look at?

Wil

 

Lateral inhibition looks promising I must say. It resembles a self organising map in
neural networks and seems to be very useful!
I'm presuming that the horizontal and amacrine cells exhibit lateral inhibition, so
there's no getting away from it!

Wil

 

Lateral inhibition is the mechanism for the whole center-surround
antagonism operation of the retina. If you think about it, it not only
enhances edge detection, it increases the dynamic range of the entire
retina. The horizontal cells implement lateral inhibition at the
photoreceptor level, and the amacrine cells at the ganglion cell level.
The bipolar cells essentially define whether a ganglion cell is on-
center/off-surround, or off-center/on-surround.

An interesting tidbit-- the light has to pass through the ganglion,
bipolar, amacrine, and horizontal cells, as well as the cell body of the
photoreceptors, before hitting the light sensitive parts of the
photoreceptor. At the fovea, this isn't as true-- the upper layers are
sort of pushed aside

Scott

 

I thought this happens thought feedback from the horizontla cells and from the
photoreceptors themsleves.
When this happens is it not just a relative shift in sensitivity?

I picked up on the the light passing though all the layers as well, what is the relevance
of this not happening at the fovea?
Or to be more precise, do you know why light passes through all the layers first?

Wil

 

The fovea does our highest acuity work, so the most direct path for light
seems to have evolved there. As for why light passes through all the other
layers, noones come up with a good reason for that. Maybe it has some
protective function for the receptors, maybe the retina is mechanically
more stable that way, maybe its just an evolutionary quick. FWIW, I seem
to recall being told that the octopus retina is not layered this way, but I
can't vouch for the info

Scott

 

Lateral inhibition doesn't just change the sensitivity, it is what is
primarily responsible for edge detection, contrast enhancement, etc.
The center-surround organization is just another way to describe
lateral inhibition. Yes, it involves connections through the
horizontal cells.

I don't think the retinal layers are different in the fovea -- light
still does pass through all the neural cell layers. What is different
is that in the rest of the retina light also has to dodge the bundles
of axons and the blood vessels supplying the retina. These tend to
skirt around the fovea.

Why the eye is built that way is another story. One idea is that it
is just an example of "incredibly stupid" design, in opposition to the
"intelligent design" notion. It probably was simply a bad design
"choice" early in the evolution of the vertebrate eye that was frozen
in place by succeeding evolutionary steps. It certainly does work
well enough, so no big deal. Cephalopod molluscs (squid and octopus)
have complex camera eyes with layered retinas that are built the
"right way", with the photoreceptors on the surface so they are the
first cells that light hits.

 

At the fovea, the bipolar cells go off at an angle insted of coming
straight off from the receptor, essentially getting out of the way. Kandel
and Schwarz have a pretty good picture of it

Scott

 

It would seem odd that light has to pass thought all the other layers first, you would
imagine that light is being muted going thought those other layers. Seeing as we a so
extremely dependant on sight this would be a very evolutionary big disadvantage.