Retina Tues. Jan. 23, 2018 1 Layers of the Retina (rods and - - PowerPoint PPT Presentation

1

COMP 546

Lecture 4

Retina

• Tues. Jan. 23, 2018

Layers of the Retina

light signals To the brain

2

(rods and cones)

Photoreceptor (rod and cone) density

3

This is the left eye. Why?

Cone density is very high in the center of the field of view. This area of the retina is called the fovea.

continuous discrete (“spikes”)

Responses of cells in the Retina

4

5

ASIDE: neural coding using spikes

(retinal ganglion cells) I mentioned this in lecture 0.

Response of neuron

(measured by experimenter)

6

Membrane potential (mV)

• 70

depolarized hyperpolarized

time

average

pre-synaptic cell post-synaptic cell

Signalling between cells at synapse

(not measured by experimenter)

7

Release rate of neurotransmitters depends on the membrane potential. Neurotransmitters can be either excitatory (depolarizing) or inhibitory (hyperpolarizing).

Q: How do nerve cells signal over long distances ?

8

input synapses

• utput synapses

Axon can be quite long (cm, or up to 1 m).

A: Spikes (“Action Potentials”)

9

http://www.youtube.com/watch?v=ifD1YG07fB8

10

• shape
• speed
• frequency (“firing rate”)
• information (see book)

https://mitpress.mit.edu/books/spikes

Receptive Field of a Retinal Cell

x y

11

Receptive field sizes increase with eccentricity

12

Receptive field diameter

• f retinal ganglion cells

6 arcmin =

1 10 degree

13

2 arcmin =

1 30 degree

NOTE: Log scale

Retinal ganglion cells encode image sums and differences :

• spectral (wavelength l) , “chromatic”
• spatial (x,y)
• temporal (t)
• spectral-spatio-temporal (l, x, y, t)

14

Spectral sums and differences

“L + M” red + green = yellow “L - M” red – green “(L + M) - S” yellow – blue

15

L – M (L + M) - S L + M S L M

Spectral sums and differences

Photoreceptors (cones) Retinal ganglion cells

16

red yellow white

17

Orange is reddish-yellow. Purple is blueish-red. Cyan is greenish-blue. Colors cannot appear reddish-green, blueish-yellow, blackish-white.

“Color Opponency” (Hering, 19th century)

black blue green L – M (L + M) - S L + M

Neural Mechanism (modern theory)

ASIDE: Classical Color Wheel

Art class ROYGBV theory of primary, secondary, and complementary colors is based on mixing pigments, not mixing lights.

18

Polar coordinates for color

19

angle = “hue” radius = “saturation”

20

'color‘ name purity intensity

(for HSV)

RGB and HSL (similar to HSV)

21

Retinal ganglion cells encode image differences :

• spectral (wavelength l) , “chromatic”
• spatial (x,y)
• temporal (t)
• spectral-spatio-temporal (l, x, y, t)

22

Spatial differences:

“center-surround receptive fields”

OFF center, ON surround

+ + + + + +

• +
• ON center,

OFF surround

23

+ and - indicate where the cell is excited or inhibited (depolarized or polarized) by bright image spot in its receptive field.

e.g. Retinal ganglion cells

(first experiments on cats done in 1953)

24

+

• -
• +
• -
• +
• -
• Shine light only

in center. (ON center) Shine light only in surround. (OFF surround) Shine light in center and surround.

time time time

Proposed Mechanism (Rodieck, 1965)

25

+

• -
• +

+ + + + +

• +
• +

Gaussian model

26

1 2𝜌 𝜏 𝑓− 𝑦2

2𝜏2

G(𝑦) =

Difference of Gaussians (DOG) model

27

1 2𝜌𝜏1 𝑓

− 𝑦2 2𝜏12

𝐸𝑃𝐻 𝑦, 𝜏1, 𝜏2 = − 1 2𝜌𝜏2 𝑓

− 𝑦2 2𝜏22

+

• =

2D Gaussian and 2D DOG

28

= 1 2𝜌𝜏 𝑓− 𝑦2

2𝜏2

𝐻 𝑦, 𝑧, 𝜏 ≡ 𝐻 𝑦, 𝜏 𝐻 𝑧, 𝜏 ∗ 1 2𝜌𝜏 𝑓− 𝑧2

2𝜏2

= 1 2𝜌𝜏2 𝑓− 𝑦2+𝑧2

2𝜏2

𝐸𝑃𝐻 𝑦, 𝑧, 𝜏1, 𝜏2 = 𝐻 𝑦, 𝑧, 𝜏1 - 𝐻 𝑦, 𝑧, 𝜏2

Response of a cell (DOG)

29

𝑀 = 𝐽 𝑦, 𝑧 𝐸𝑃𝐻 𝑦 − 𝑦0 , 𝑧 − 𝑧0 , 𝜏1, 𝜏2 𝑒𝑦 𝑒𝑧

Response depends on:

+

• -
• DOG cell centered at (𝑦0 , 𝑧0 )

Here I am ignoring temporal properties for simplicity.

Linear response model

30

𝑀 = 𝐸𝑃𝐻 𝑦 − 𝑦0 , 𝑧 − 𝑧0 , 𝜏1, 𝜏2 𝐽 𝑦, 𝑧 𝑒𝑦 𝑒𝑧

Alternatively we can write it as a sum:

𝑀 =

𝑦,𝑧

𝐸𝑃𝐻 𝑦 − 𝑦0 , 𝑧 − 𝑧

0 , 𝜏 1, 𝜏2

𝐽 𝑦, 𝑧

“Static Non-linearity”

31

Spike firing rate (spikes per second)

𝑀

~200 10

𝑢

Spike train

Half-wave rectification model

32

Spike “firing rate”

𝑀

Max firing rate: in practice there is a cutoff

𝑀 = max( 0,

𝑦,𝑧

𝐸𝑃𝐻 𝑦 − 𝑦0 , 𝑧 − 𝑧

0 , 𝜏 1, 𝜏2

𝐽 𝑦, 𝑧 )

Responses of a population of DOGs

33

+

• -
• +
• -
• +
• -
• +
• -
• +
• -
• +
• -
• +
• -
• +
• -
• +
• -
• +
• -
• +
• -
• +
• -
• … and many overlapping ones which I am not

showing because it would be too messy

“Cross correlation”

34

+

• -
• image

DOG

Responses of a population of DOGs

35

Cross correlation operator

𝑀 𝑦0, 𝑧0 ≡ 𝐸𝑃𝐻 𝑦 , 𝑧 , 𝜏1, 𝜏2 ⨂ 𝐽 𝑦, 𝑧

≡

𝑦,𝑧

𝐸𝑃𝐻 𝑦, 𝑧 𝐽 𝑦0 + 𝑦, 𝑧

0 + 𝑧

≡

𝑣,𝑤

𝐸𝑃𝐻 𝑣 − 𝑦0, 𝑤 − 𝑧 𝐽 𝑣, 𝑤

change of variables

Cross correlation

36

≡

𝑣,𝑤

𝑔 𝑣 − 𝑦, 𝑤 − 𝑧 𝐽 𝑣, 𝑤

Convolution (to be discussed later)

𝑔 𝑦, 𝑧 ⨂ 𝐽 𝑦, 𝑧 ≡

𝑣,𝑤

𝑔 𝑦 − 𝑣, 𝑧 − 𝑤 𝐽 𝑣, 𝑤 𝑔 𝑦, 𝑧 ∗ 𝐽 𝑦, 𝑧

Technical detail (boundary effects)

37

+

• -
• image (photoreceptors)

+

• -
• +
• -
• +
• -
• Not well defined

responses Well defined response