1. Rod-Cone Convergence
In The Retina
PhD Thesis Presentation
12th
January 2014
Presented By:
Kendi Muchungi
Supervised By:
Dr. Matthew Casey
Dr. André Grüning
2. How does one see?
2Source:
University of Utah; Webvision, 2011
4. The Retina: Participating Neurons (ii)
4
1
2
CONE
3
2
4
3
Light
OuterPlexiformLayerInnerPlexiformLayer
ROD
HC
RNB CNB
AII
CFB
CNG
CFG
HC – Horizontal Cell
RNB – Rod ON Bipolar Cell
CNB – Cone On Bipolar Cell
CFB – Cone Off Bipolar Cell
AII – Amacrine Cell
CNG – Cone On Ganglion Cell
CFG – Cone Off Ganglion Cell
1 - Electric Gap Junction (Coupling)
2 - Sign Reversing Symbol
3 - Synapse
4 - Sign Conserving Symbol
Source:
Muchungi & Casey, 2012
5. A little by the way …..
5
Source:
Deakin University, 2014
6. How clearly can one see?
What can you see at 20 feet?
Visual Acuity
=> Optical and Neural Factors
Focus within the eye
State and function of the Retina
Faculty of the brain
6
7. For this Presentation
- Motivation
- Functions
- Existing Models
- Research Question
- Our Model
- Significance of Our Model
- Conclusion
- Application
- Future Work
7
8. Initial Motivation
Retinal degeneration is the cause of about 50%
of the existing cases of total blindness . (Graf et
al., 2007)
Sadly, no medical treatment exists that can help
with the regeneration of the affected retina
tissue/neurons.
Retinal prosthetics: Computational simulations
inform their design and development.
8
9. Hence ……..
Our research
–Biologically inspired model of the retina
Because
1) The retina outperforms existing man-made image
acquisition devices.
2)Inform and motivate the design and development
of retinal implants/prosthesis.
9
10. The Retina: Her Functionality
Light adaptation
Allows for visual perception even when there is a dynamic
range of light intensity
Contrast Gain Control
Allows for the perception of the presence of contrast, however
subtle; manages sensitivity to the presence of contrast
Spike Generation
Generates action potential [ON/OFF]
10
11. Existing Retina Models: Light Adaptation
Computer Retina
(Shah & Levine,
1996)
Neural Model
(Wilson, 1996)
Silicon Retina
(Zaghloul &
Boahen, 2006)
Virtual Retina
(Wohrer &
Kornprobost, 2009)
Aspects of
Light
Adaptation
1. Phototransduction
2. Cone--Cone Coupling
3. HC feedback
1. Phototransduction
2. Cone-Cone Coupling
3. HC feedback
1. Phototransduction
2. Cone--Cone Coupling
3. HC feedback
1. Phototransduction
2. Cone-Cone Coupling
3. HC feedback
How they
captured
these aspects
1. Low-pass temporal filter
2. Gaussian Operator
3. Difference of Gaussians
(D-o-G)
4. Michaelis-Menten
function
1. Low-pass temporal
filter
2. Gaussian Operator
3. D-o-G
4. Michaelis-Menten
function
1. nMOS transistors
2. Synaptic strengths
adjusted
3. Subtractive HC
feedback
1. Low-pass filters
2. Difference of Dirac
3. D-o-G
4. Gaussian Operator
Comments 1. 1st
to model dynamic
adaptation
2. Use a visual acuity
function for cone-cone
coupling
1. Single nonlinear
differential equation
for each retina neuron
2. Temporal mapping
only
1. Captured all aspects of
light adaptation and
fabricated a retina
using silicon chips
1. Linear filters throughout
model
2. Adapted for large-scale
simulations
11
12. Existing Retina Models: Contrast Gain Control
Computer Retina
(Shah & Levine,
1996)
Neural Model
(Wilson, 1996)
Silicon Retina
(Zaghloul &
Boahen, 2006)
Virtual Retina
(Wohrer &
Kornprobost, 2009)
Aspects of
Contrast Gain
Control
1. ON pathways
2. Midget and Diffuse
BCs
3. Antagonistic
morphology of retina
neurons in the IPL
and Ganglion Layer
1. ON and OFF pathways
2. Midget and Diffuse
BCs
3. Antagonistic
morphology of retina
neurons in the IPL
and Ganglion Layer
1. ON and OFF pathways
2. BCs
3. Narrow and Wide
Amacrine
1. ON and OFF pathways
2. Diffuse BCs
3. Antagonistic morphology
of retina neurons in the
IPL and Ganglion Layer
How they
captured these
aspects
1. Low-pass filters
2. Gaussian Operators
3. D-o-G
4. Arctangent Function
1. Non-linear differential
equation
2. Divisive feedback from
Amacrine Cells (A+)
1. Transistors
(Increase/Decrease
Voltage for
excitation/inhibition)
1. Low-pass filters
2. Gaussian Operators
3. D-o-G
4. Non-linear divisive
feedback loop
Comments 1. No involvement of
Amacrine cells
1. Incorporates Amacrine
cells
1. Switch between ON
and OFF pathways (not
very parallel)
2. Divisive feedback from
Narrow and Wide
Amacrine
1. Allow the level of
luminance to influence
CGC
12
13. Existing Models: Biological Aspects Established
13
Light Adaptation Contrast Gain Control
Participating Neurons and Pathways
1) Cone Specific
2) No Rod Influence
1) ON & OFF Pathways
2) Midget & Diffuse Bipolar
Cells
Biological Processes Involved
1) Phototransduction
2) Cone-cone coupling
3) Negative Feedback
1) Concentric Morphology
2) Divisive Feedback
14. Existing Models: Mathematical Approximations
14
1. Low-pass filter
1. Gaussian Operator
1. Difference of Gaussians (D-o-G)
1. Michaelis-Menten Function
Kτ t;τ( ) =
e
−t
τ
τ
Gσ x, y( ) =
1
2πσ 2
e
−( x2+y2 )
2σ 2
CS x,y,t( ) = αCK t;τC( ) G x,y;σ C( ) −αSK t − d;τ S( ) G x, y;σ S( )
P I( ) =
In
In
+
kr
kb
A +kr
÷
n
÷
÷
÷
÷
÷
Pmax
15. What is New?
Recent research (Thoreson, 2007; Trumpler, 2008;
Pang, 2010; Schiller, 2010) point to the importance
of the connections between our Rod and Cone
pathways.
1) Rods and Cones do have electrical connections
between them; Gap junctions [Rod-Cone Coupling]
2) By the use of lateral neurons; Horizontal Cells and
Amacrine cells
15
16. Research Questions
1. What effect does rod-cone convergence have
on retina functionality; light adaptation and
contrast gain control?
2. How might this convergence be exploited?
16
17. Our Retina Model: Biological Aspects
17
Light Adaptation Contrast Gain Control
Participating Neurons and Pathways
1) Cone Specific
2) With Rod Influence
1) ON & OFF Pathways
2) Diffuse Bipolar Cells
3) Rod ON Bipolar Cells
4) Amacrine Cells
Biological Processes Involved
1) Phototransduction
2) Cone-cone Coupling
3) Rod-cone Coupling
4) Negative Feedback
1) Concentric Morphology
2) Divisive Feedback
19. Spatial Extent: Light Adaptation
Incorporated Rod model developed by Lamb
and Pugh in 1992
Integration
1.Rod output gives us our spatial extent
2.Is a 10th
of the driving ambient intensity within
the OPL
19
20. Biological Evaluation: Light Adaptation (i)
Using Threshold vs Intensity (tvi) function
To steady background intensity, flash intensities given by 10×2p
td, where p is an
integer in 0 ≤ p ≤ 5 (Hood, 1998).
20
Source:
Schnapf et al., 1990
21. Biological Evaluation:
Light Adaptation (ii)
To attain results comparable to biology
-It took evaluating four variations of our model
-Determined the need for a ‘rod input factor’
(rif)
- rif is dynamic
- Pegged to existing level of light intensity
- Allowed for close replication of input intensity
- Response to the tvi function closely replicating
biology
21
23. Spatial Extent: Contrast Gain Control
Incorporated
1.Rod ONBC model developed by Shapley & Enroth-
Cugell in 1973 (Cat)
2.Amacrine (AII) model by using Michaelis-Menten
Function and low-pass filter (tau = 80ms)
Integration
1.Rod ONBC output gives the spatial extent
2.Is a 10th
of the divisive feedback Spatio-temporal
intensity.
23
25. Biological Evaluation:
Contrast Gain Control (ii)
25
- No obvious value was attributed to Rod ONBC time constant
- Determined a range
- Lower bound 0.05ms [Wohrer & Kornprobst, 2009]
- Upper bound 140ms [Wilson, 1997]
- To allow for ease of testing we narrowed down the range
- Lower bound 1ms
- Upper bound 10ms
- At 4 different time steps
- We then carried out 40 tests
- We then concluded on a time constant of 7ms
26. Rod Influence Significance (i)
26
Quantitative Approach
Signal-to-Noise Ratio (SNR)
Why?
Point to SNR being improved by Rod-Cone
Convergence [Thoreson, and by Trümpler et. al]
Compare
- Unsharp Mask [USM]
Contrast Enhancement Algorithm
- With rod influence [BioRod] [CGC]
- No rod influence [NoRod] [CGCNoRod]
27. Rod Influence Significance (ii)
27
To facilitate this analysis, we adapted the brightness of a
segment of input stimuli
1)Obtain Mean
2)Subtract Mean
1)Add back a fraction of the Mean
I ' x, y( ) = I x, y( ) −µ 1+α( )
µ =
I x, y( )
n
0.00 1.00 2.00 3.00 4.00 5.00
Lena
28. Significance of Analysis: Implications
28
SNR
LA CGC
Contrast is
perceived in spite
of varying
luminance levels
High sensitivity
to subtle
presence of
contrast
SD
LA CGC
Visual
Perceptual
Consistency
Contrast
Perceptual
Consistency
Signal-to-Noise Ratio (SNR) Standard Deviation (SD)
34. Implications for Contrast Gain Control (iv)
34
0.00 1.00 2.00 3.00 4.00 5.00
LenaUSM
CGC
ONBC
CGC
OFFBC
CGCNoRod
ONBC
CGCNoRod
OFFBC
35. Conclusion
Rod-Cone Convergence
a.Improves SNR
-Better perception of contrast [Light Adaptation]
-Increases sensitivity to contrast [Contrast Gain Control]
b.Reduces SD thereby increasing
-Visual perceptual constancy [Light Adaptation]
-Contrast perceptual constancy [Contrast Gain Control]
Rod-Cone convergence improves visual acuity.
35
36. Thesis Contributions
Posters, Papers and Publications
a. Bernstein Conference 2014: Poster presentation
b. ICANN 2012: Paper publication
c. Computing Department PhD Conference 2012:
Paper presentation
d. University of Surrey PhD Conference 2012: Paper
presentation
e. Computing Department PhD Conference 2011:
Poster presentation
36
37. Thesis Contributions
Model Contributions
a. Biologically viable model of the cone pathway: OPL
level with rod-cone integration via rod-cone coupling
b. Biologically viable model of the cone pathway: IPL
level with rod-cone integration via rod Bipolar Cells
and Amacrine Cells
c. Unique method of evaluating model performance.
d. Demonstrated the importance of rod-cone
convergence to visual acuity. 37
38. Applicability
1) Obstacle avoidance in mobile robots
Ryad Benosman and his lab developed an asynchronous
neuromorphic visual sensor.
- Bio-inspired
Incorporates parallel processing (Schiller,2010)
2) Improve visual acuity
Daniel Palanker’s Hensen Lab [Stanford University] made
a major breakthrough in the restoration of sight
- Bio-inspired
Introducing stimulation at the IPL
38
40. References
40
University of Utah; Webvision, October 8th
2011. The Organization of the Retina and
Visual System: Simple Anatomy of the Retina by Helga Kolb
Deakin University, 2014. Eye Fun; Online Eye Tests. Retrieved from
http://www.deakin.edu.au
K. Muchungi and M. Casey, 2012. Simulating light adaptation in the retina with rod-
cone coupling. ICANN 339-346, Berlin, Heidelberg, Springer-Verlag.
J. L. Schnapf, B. J. Nunn, and D. A. Baylor, 1990. Visual Transduction In Cones of
the Monkey Macaca Fascicularis. 427(1):681-713.
S. A. Baccus and Markus Meister, 2002. Fast and slow contrast adaptation in retinal
circuitry. Neuron, 36(5):909-919.
41. Rod-Cone Convergence
In The Retina
PhD Thesis Presentation
12th
January 2014
Presented By:
Kendi Muchungi || k.muchungi@surrey.ac.uk
Thank You!
42. Parallel Pathways in the Retina
42
Cone and Rod Pathways
Cones
- Processes lightness
- Processes colour
Rods
- Processes darkness
- Monochromatic
ON and OFF Pathways
ON
- Respond to light increment
OFF
- Respond to light decrement
Midget and Diffuse Pathways
Midget
- Colour selective
Diffuse
- Respond to changes in contrast
- Monochromatic
To start off, I’ll begin by
Motivating our research
Brief overview of the review done to date
Explain our simulator
Discuss our results
Propose the future direction of our research
Because the retina outperforms existing man-made acquisition devices – it therefore follows that the best way to inform the design of retinal prosthetics is by computational simulations of human biological visual systems at retinal level.
There do exist light adaptation retina simulators – however, none that capture most recent information pertaining to the mechanisms employed by the retina, which contribute to the retina’s ability to process visual information
Retinitis Pigmentosa & Macular Degeneration examples of retinal degeneration
2 kinds of retinal implants:
Epi-Retinal – this is where a complex image processing and transmission unit is introduces at the bottom of the retina and feeds electrical impulses to the optic nerves
Sub-Retinal – this is where the degenerated photoreceptors are replaced in loco and directly stimulate the subsequent retina neurons, thereby making use of their image processing abilities
Because the retina outperforms existing man-made acquisition devices – it therefore follows that the best way to inform the design of retinal prosthetics is by computational simulations of human biological visual systems at retinal level.
There do exist light adaptation retina simulators – however, none that capture most recent information pertaining to the mechanisms employed by the retina, which contribute to the retina’s ability to process visual information
Princess and the Pea -
There do exist retiina simulators that mimic light adaptation – good ones at that
Are Cone driven
however they do not encompass recent information pertaining to light adaptation, which contributes to the performance of the retina.
We however, manage to do so in our simulator.
There do exist retiina simulators that mimic light adaptation – good ones at that
Are Cone driven
however they do not encompass recent information pertaining to light adaptation, which contributes to the performance of the retina.
We however, manage to do so in our simulator.
There do exist retiina simulators that mimic light adaptation – good ones at that
Are Cone driven
however they do not encompass recent information pertaining to light adaptation, which contributes to the performance of the retina.
We however, manage to do so in our simulator.
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
There do exist retiina simulators that mimic light adaptation – good ones at that
Are Cone driven
however they do not encompass recent information pertaining to light adaptation, which contributes to the performance of the retina.
We however, manage to do so in our simulator.
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
Monkey’s cone response to the tvi function according to Schnapf, Nunn and Baylor [Visual Transduction in Monkey Macaque]
I may have given that we attained these results without much effort, which isn’t true really.
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
- Baccus and Meister measured contrast gain control in a Salamander using Linear-Non-Linear Analysis
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
These electrical connections allow for the transmission of information
At this point it might be best to delve alittle into the structure of the retina and explain why Rods and Cones
----- Meeting Notes (18/11/2011 09:05) -----
Retinitis Pigmentosa & Macular Degeneration examples of retinal degeneration
2 kinds of retinal implants:
Epi-Retinal – this is where a complex image processing and transmission unit is introduces at the bottom of the retina and feeds electrical impulses to the optic nerves
Sub-Retinal – this is where the degenerated photoreceptors are replaced in loco and directly stimulate the subsequent retina neurons, thereby making use of their image processing abilities
----- Meeting Notes (18/11/2011 09:05) -----
Retinitis Pigmentosa & Macular Degeneration examples of retinal degeneration
2 kinds of retinal implants:
Epi-Retinal – this is where a complex image processing and transmission unit is introduces at the bottom of the retina and feeds electrical impulses to the optic nerves
Sub-Retinal – this is where the degenerated photoreceptors are replaced in loco and directly stimulate the subsequent retina neurons, thereby making use of their image processing abilities
----- Meeting Notes (18/11/2011 09:05) -----
Retinitis Pigmentosa & Macular Degeneration examples of retinal degeneration
2 kinds of retinal implants:
Epi-Retinal – this is where a complex image processing and transmission unit is introduces at the bottom of the retina and feeds electrical impulses to the optic nerves
Sub-Retinal – this is where the degenerated photoreceptors are replaced in loco and directly stimulate the subsequent retina neurons, thereby making use of their image processing abilities
----- Meeting Notes (18/11/2011 09:05) -----
Retinitis Pigmentosa & Macular Degeneration examples of retinal degeneration
2 kinds of retinal implants:
Epi-Retinal – this is where a complex image processing and transmission unit is introduces at the bottom of the retina and feeds electrical impulses to the optic nerves
Sub-Retinal – this is where the degenerated photoreceptors are replaced in loco and directly stimulate the subsequent retina neurons, thereby making use of their image processing abilities
We love to be able to incorporate the following aspects to our current model
----- Meeting Notes (18/11/2011 09:05) -----
Retinitis Pigmentosa & Macular Degeneration examples of retinal degeneration
2 kinds of retinal implants:
Epi-Retinal – this is where a complex image processing and transmission unit is introduces at the bottom of the retina and feeds electrical impulses to the optic nerves
Sub-Retinal – this is where the degenerated photoreceptors are replaced in loco and directly stimulate the subsequent retina neurons, thereby making use of their image processing abilities
----- Meeting Notes (18/11/2011 09:05) -----
Retinitis Pigmentosa & Macular Degeneration examples of retinal degeneration
2 kinds of retinal implants:
Epi-Retinal – this is where a complex image processing and transmission unit is introduces at the bottom of the retina and feeds electrical impulses to the optic nerves
Sub-Retinal – this is where the degenerated photoreceptors are replaced in loco and directly stimulate the subsequent retina neurons, thereby making use of their image processing abilities
This gives us our Spatio-Temporal Intensity, which is the driving intensity of the visual intensity
These weightings help us manipulate the participation of Rod-Cone coupling in our system
This gives us our Spatio-Temporal Intensity, which is the driving intensity of the visual intensity
These weightings help us manipulate the participation of Rod-Cone coupling in our system
This gives us our Spatio-Temporal Intensity, which is the driving intensity of the visual intensity
These weightings help us manipulate the participation of Rod-Cone coupling in our system
* refers to the convolution – blending two intensities – with one of the intensities
100 because cones respond to a single photon of light about 100 times more than rods
Enhance local adaptation – works to synchronize information within cones in close proximity to each other and thereby improve the clarity of the image – in other words visual acuity
The constants used here keep the coupling of cones bound to the level of light illumination
Here is should flip between slide 9 – which has the diagram of the outer nuclear and then
Final cone output is given when Cone Output without HC feedback is weighted to represent HC output that is negatively feeding back into the Cone to give is our final cone output
Visual Pigment in photoreceptors (Rhodopsin & Cone Opsin) is activated, activated opsins then activate the G protein, Transducin, which in turn activates the subunits of the Inhibitory effector enzyme cGMP, PDE. Activated PDE cause cGMP to dissociate from the cGMP gated channels, which leads to the closure of these channels and stops the influx of Na+ and Ca2+.
The hyperpolarization of Photoreceptors causes a reduction in the release of the neurotransmitter glutamate, which causes two different responses in BCs. OffBCs, which happen to be sign conserving hyperpolarize and OnBCs, which are sign inverting depolarize. HCs also depolarize, because they are also sign inverting.