From Particle Detectors to Retinal Studies

1
From Particle Detectors to Retinal Studies

• Experimental Particle Physics Group
• Detector Development
• C. Adams, A. Beattie, W. Cunningham, A. Curtis,
• D. Gunning, J. Melone, J. Morrison, V. OShea,
  K.M. Smith, CDW. Wilkinson
• A. Litke, E.J. Chichilnisky, S. Kachiguine, A.
  Sher, D. Petrescu, M. Grivich
• P.I. M. Rahman

Keith Mathieson
2
Why study the retina?

• What does the eye tell the brain?
• The retina is a part of the brain and performs
  some pre-processing of the visual scene before
  the information is sent to the visual cortex.
• Can a circuit diagram be used to represent how
  the retina processes information?
• Approachable part of the brain.
• Artificial retinas
• Is it possible to use microelectronics to replace
  damaged retina as a potential cure for blindness?

3
The Retina

• 108 photoreceptors
• rods and cones
• Parallel analogue processing in 3 layers
• horizontal, bipolar and amacrine cells
• Output from ganglion cell layer
• 106 optic nerve connections
• Area 10 cm2

4
Single cell retinal studies

• Historically retinal studies were performed on
  single cells
• More recent evidence suggests that the retina
  communicates with the brain through correlated
  signalling. Neurobiologists beginning to look at
  systems with 10s of recording electrodes
• Necessitates the need for studies with
  multi-channel arrays coupled to dedicated
  electronics
• Solution use the technology developed for
  particle physics to understand how the eye talks
  to the brain

5
Particle detectors vertex detectors

• Small signal detection
• MIP 24000 e/h pairs
• High-density detection elements
• Microstrip and pixel technology
• Semiconductor fabrication techniques
• Particle physics groups develop fabrication
  skills for investigating radiation tolerant
  detectors
• Multi-channel electronics
• Readout requirements for PPE involve
  state-ofthe-art electronics
• Data acquisition, suppression and analysis
• PPE still at forefront of data handling e.g.
  GRID

ALEPH vertex detector
Particle physics has developed expertise which is
applicable to other areas of science -
Neurobiology
6
Retinal Readout Project

• Collaboration between
• Particle physicists
• Alan Litke
• SCIPP (Santa Cruz) and CERN

• Neurobiologists
• E.J. Chichilinisky
• Salk Institute for Biological Studies, San Diego

and
Aim Use expertise gained in the field of
particle physics to develop a system capable of
recording the outputs of hundreds of ganglion
cells. Understand the language that the eye
uses to talk to the brain
7
Experimental set-up
Biological studies Salk Institute, San
Diego Readout system University of
California Santa Cruz (UCSC) Microelectrode
array development University of Glasgow
UCSC VLSI chip design University of Krakow
(CERN)
Biological studies Retina separated
from pigment epithelium Mounted ganglion side
down Colour monitor imaged onto photoreceptor
layer Continuously superfuse bicarbonate-buffe
red Ames medium at 35 oC
8
Electrode Array Geometries
1 electrode traditional
61 electrodes state-of-the-art
512 electrodes (32x16) used in pilot experiments
Input region for monkey MT neuron
1.7 mm2 60 µm
0.17 mm2 60 µm
1 mm
7.1 mm2 60 µm
0.43 mm2 30 µm
1.7 mm2 60 µm
519 electrodes fabrication underway
519 electrodes under development
2053 electrodes futuristic

• Litke
• NSS03

(Electrode diameters 5 µm area and electrode
spacing given below.)
9
Microelectrode arrays
Arrays are fabricated on a transparent conductor
called Indium Tin Oxide (ITO)
ITO coating is 300nm thick on a 1 mm thick glass
substrate

• 2 types of detector fabricated
• 61-electrode array and 500-electrode array

• ITO is patterned using photolithography
• and dry-etch processes

2. SiN passivation layer is plasma deposited
3. Vias are dry-etch through SiN layer down to ITO
4. ITO electrodes are electroplated with platinum
10
61-electrode array
Electrode Separation 60mm Sensitive area
0.17mm2
11
Readout System

• Bandpass filter 50-2000 Hz
• lt5mV rms noise
• Sampling rate 20 kHz

Chip design by W. Dabrowski and
colleagues. University of Mining and Metallurgy,
Krakow (CERN)
12
Signals from 61-electrode array

• Signals from ganglion cells (spikes)
• Amplitude 50 - 500 mV
• Width 1 - 2 ms
• Cells fire in a correlated manner
• Spike rate is thought to encode the
• information of the visual scene
• Code not well understood
• Large data sets recorded, 12 hours gives 55Gb

13
Results from Neurobiological experiments

• Results from 61-electrode array
• allow an examination of the structural
• functionality of the retina
• By stimulating an individual
• photoreceptor one ganglion cell
• was observed to signal strongly
• whilst neighbouring cells also sent
• signals to the brain but with less intensity.
• By stimulating 2 cones it was shown that response
  of ganglion cells summed linearly, despite
  non-linear elements in the intervening layers

E.J. Chichilnisky and D.A. Baylor, Nature
neuroscience, Vol. 2 (10) 1999
14
512-electrode array
512-electrode array gives 2mm2 sensitive
area 128 bondpads each side Individual
electrodes are in an hexagonal close-packed
structure but wire width does not allow an
overall hexagonal geometry
Electrode separation 60mm
Electrode diameter 5mm
15
519-electrode array
Would like to cover a larger area of the
retina 519-electrode array gives 2mm2 sensitive
area Electrode Separation 60mm Electrode diamet
er 5mm
Requires electron beam lithography
16
Readout electronics

• 8 64-channel ASICs
• Giving 512 channels
• Wire bonded to array
• Platinise electrodes using chip
• Readout rate 20kHz

Pictures courtesy of A. Litke
17
Signals from 512 array

• System operational early this year.
• Able to make maps of neural activity as a dynamic
  image is focussed on to the photoreceptors
• Large data sets from
• recordings, 8 hours
• gives 650Gb

Pictures (and movie) courtesy of E.J.
Chichilnisky and A. Litke
18
Neuron Identification (signals on electrodes ?
spikes from identified neurons)
Spike Amplitude histogram
Spike width vs. amplitude
Single electrode (electrode 1)
Multiple electrodes
1.3 ms

• Litke
• NSS03

Electrode
19
Principle component analysis

• Characteristics of each recorded waveform
  represented by a multi-dimensional vector
• E.g. Signal width, amplitude, ON or OFF cell,
  bipolar, tripolar
• Choose components which have the greatest
  variance (usually 5)
• Represent spike as a point in 5-dimensional space
• Clustering of points leads to
• neuron identification
• Large data sets 1Tb
• Must be an automated
• neuron finding code (Java)

20
Principal Components Analysis multidimensional
clustering ? 4 identified neurons
Average signal on each of the 7 electrodes for
each of the 4 identified neurons
Neuron
1
2
3
4

• Litke
• NSS03

Electrode
1
5
4
3
2
6
7
21
Some first (preliminary) results with monkey
retina
Light-sensitive regions (receptive fields) for
338 identified neurons
1.6 mm

• Litke
• NSS03

3.2 mm
22
Receptive field mosaics
On cells detect a transition from dark to
bright Off cells detect a transition from bright
to dark
Off- large
On- large
On- small
Off- small
Blue-ON
1.6 mm

• Litke
• NSS03

3.2 mm
23
Neural activity recorded with 512-electrode
system as image of vertical moving bar is focused
on a section of guinea pig retina
Electrode spike-rate

• Some cells are direction selective
• 7 directions
• 4 left-right, up down
• 3 orientated with ear canal

Spike-rate for On-off DS neurons
Spike-rate for On-off DS neurons

• Litke
• NSS03

2 mm
24
Electrophysiological Imaging
1000 ?V
?4
?4
1.6 m/s
?4
Possible to track the electrical signal as it
travels from the retina to the brain
?4

1. Litke NSS03

2 ms
25
Future studies

• 519 electrodes at 30 mm spacing
• Line width reduced to 1mm
• Can study midget cells and retinal microstructure
• 2053 electrodes at 60 mm
• Line width 1mm
• Area 7 mm2
• Offers full coverage of the MT neuron

26
Retinal prosthesis

• Can these studies, into how the retina encodes
  information, help people who are visually
  impaired?

27
Artificial Retinas

• Microelectronics for a retinal prosthesis?
• Restore partial sight to blind patients
• Several groups working on possible solutions
  around the world
• USA Retinal Implant Project
• Implanted arrays in patients
• Germany EPI-RET and SUB-RET
• Optobionics (USA) company not yet offering
  products for sale

28
Diseases of the retina

• Major cause of blindness in the western world
  macular degeneration
• Central part of the retina, responsible for
  detailed vision (reading etc.) no longer able to
  detect light.
• Cause photoreceptors (cones) no longer produce
  signals

healthy degenerated
29
The fovea

• High density area of cones in the centre of the
  retina
• Intervening cells pushed out to increase exposure
  of photoreceptors
• With macular degeneration only photoreceptors
  affected
• Ganglion cells still function but receive no
  input.

30
USA Retinal Prosthesis
Picture from http//www.irp.jhu.edu/project/

• Silicon imaging camera on glasses
• RF coupled to antenna implanted in eye
• Microelectrode array receiving signals from ASIC
  and stimulating ganglion cells

31
Artificial retina chips
M. Humanyan (USC), W. Lui (UCSC) et al.
Chip hardwired to outside world
Patient trials at Johns Hopkins Hospital,
Baltimore (USA), now at USC
32
Artificial retina chips
Fabricated chip, self-powered using solar cells
Implanted chip
Clinical trials by Optobionics Corporation (USA)
33
Design of implant

• Collaboration involving ASIC designers (RAL),
  neuroscientists and particle detector physicists
• EPSRC funded, started Dec 2002
• RAL using existing expertise developed for
  particle experiments to make imaging detector and
  retinal chip
• Glasgow to make microelectrode arrays and perform
  biological testing

34
Design of implant
Retina
Implant
Pixel detector (MAPS)
Photoreceptors (light sensitive)
Horizontal, Bipolar and amacrine cells (processing
)
Neural network chip (Si ASIC)
Microelectrode array
Ganglion cells (output)
35
Implant

• Pixel detector to receive light and create
  electrical signal
• Monolithic active pixel sensor (MAPS)
• Pixel size 5mm
• Retinal chip to pass the information from the
  pixel detector to the correct stimulating
  electrodes
• Can be programmed using the information from the
  retinal readout project
• Microelectrode array to stimulate the ganglion
  cells
• Biocompatible micro-array of electrodes

36
Prototype array

• Initially an 8-electrode array fabricated on a
  biocompatible substrate
• Polyimide substrate
• Gold wire connections
• Wires down to 10mm
• Platinum electrodes
• 25 mm diameter
• Electrode characteristics
• well behaved
• Preliminary tests with frog
• Detected signals from heart
• Retinal experiments soon
• Dr. Jim Morrison (Neuroscience and biomedical
  systems, Glasgow University)

37
Next stage

• 18 electrode array
• 9 recording electrodes
• 9 stimulating electrodes

• Bonding area for ASIC
• Using Ti wires track width down to 4mm
• Permits increased electrode densities

38
Artificial neural networks
Emulating biological neural networks by
connectionism
inputs
outputs
hidden nodes
Difficult to implement all connections in silicon
Part of biological neural network
Will a neural network with reduced connections
suffice? If so, possible to fabricate using CMOS
processes
39
Reduced Neural Network Investigations
Neural network input image - tooth
Trained network response
Untrained network response
40
Status

• 10x10 MAPS pixel detector
• under design M. Prydderch and J. Crooks (RAL)
• Retinal chip with neural network incorporated
• Simulations continuing, designs under
  consideration
• Microelectrode arrays prototypes tested
• Need tests on biocompatibility issues and
  stimulating cells

41
Other applications

• Biological neural networks
• How do nerves communicate with each other?
• Collaborate with IBLS (Centre for Cell
  Engineering) Prof. A. Curtis and Prof. C.D.W.
  Wilkinson
• Aim to culture small biological neural networks
  and understand how the neurons communicate with
  each other.
• Control the growth of cultured neurons into
  different biological networks and record signals
  using microelectrode arrays
• A. Beattie
• M. Sokolov

42
Cultured rat heart cells

• Heart cells produce large signals (1mV)
• Can observe contractions and correlate with
  recorded signals

Movie courtesy of A. Beattie
43
Biological neural networks

• Help understanding of neuronal signalling
• ITO array below cell, flexible array above cell
• Controlling cell growth and understanding
  inter-neuron communications can help with
  projects such as spinal cord regrowth

44
Conclusions

• Technologies developed for particle physics have
  a lot to offer other areas of science.
• In particular understanding how large (retina)
  and small scale neural networks process
  information
• Thanks for listening