Transcript Puzzel Talk - School of Computer Science

Optical Neural System Imaging
Survey
November 15, 1999
Andreas G. Nowatzyk
Outline
• Background and Motivation
• System Overview
• Backscatter Imager
• Fluorescent Imager
• Microtome
• Staining Unit
• Computational Aspects
• Summary
Context
• Enabling technologies
–
–
–
–
–
–
Moore’s Law: Billions of Billions of cycles
Interconnect networks: Multi-Gbytes/s I/O
Optics: confocal microscopy, lasers
Biochemistry: selective staining
Computer science: Image processing
Mechanics: Chip fabrication technologies
• Interdisciplinary research: Combining all
these new tools to open up new research
areas
Ultimate Goal: Scan a Mouse in 3D
• Eventually,
reverse
engineer a
mouse brain,
including the
entire rodent
nervous system
Computational Challenge
• 2 x 2 x 4 cm3 specimen volume
• 0.2 mm resolution
• = 100,000 x 100,000 x 200,000 voxels
• 6 x 12 bit per voxel
• 16,763,806 Gbytes of raw data
• 200:1 compression (lossy)
• => 1700 tapes (8mm, 50Gbytes/tape)
Computational Challenge
• 20 Msamples/s per channel (12 bits
each)
• 6 x 64 channels
• ~50% scan duty cycle
• 2x for complex phase sensing
• = 11,520 Mbytes/s average data rate
System Overview
Basic approach, instrument
architecture, and computing
infrastructure
Functional Imaging
• Selective labeling with Fluorescent Dyes
• Genetically engineered Labels (GFP)
VS.
Basic Approach
• Use of light microscopy plus selective
functional imaging
• Five step process
– Bulk imaging into freshly cut sample
– Mechanical sectioning via integrated microtome
– Automated, continuous staining
– Functional, fluorescent imaging
– Data fusion, compression and archival storage
Confocal Light Microscopy
• Using one objective lens twice
• Point-spread function squared
Instrument Overview
Backscatter
Imager
Carrier Tape
Transfer Unit
Specimen
Holder
Microtom
Staining Unit
Fluorescent
Imager
Backscatter Imager
Optical system, mechanical
components, scan path
Design Objectives
• Need to maximize throughput
• Maximize practical resolution
• Optical sectioning into the exposed
sample surface
XZ Detection Function
High NA (0.9) objective with pinhole detector
PMT
XZ Detection Function
Confocal illumination
PMT
XZ Detection Function
Michelson interferometer
PMT
Acousto-Optical Modulator
• Exploit Doppler shift to split laser into
two components with differing frequency
RF
w-e
w+e
XZ Detection Function
Heterodyne detection
X
PMT
f
w+e
w-e
Heterodyne Detection
• Elimination of black-level drift through ACcoupled amplifiers
• Contrast independent of reference beam
intensity (within dynamic range of detector)
• Optical phase transferred to electrical domain
• Simultaneous capture of phase and magnitude
• Enables wave-front reconstruction / holography
Solid State Detector
• Better quantum efficiency (~ 85 vs 30 %)
• Better dynamic range (> 120 db)
y = 2.0328x + 24.826
y = 2.0469x + 7.2512
y = 2.036x - 13.267
Signal Out [dbm]
HP SS-detector
-20
-30
-40
-50
-60
-70
-80
-90
-100
-110
-120
-130
-70
-60
-50
-40
-30
Optical Input Power [dbm]
Sig 0db [dbm]
Noise floor [dbm/Hz]
PMT(700V), -40db
Linear (Sig -20db [dbm])
Sig -20db [dbm]
PMT(700V),0db
PMT(700V) noise floor
Linear (Sig -40db [dbm])
Sig -40db [dbm]
PMT(700V),-20db
Linear (Sig 0db [dbm])
-20
Qualitative Verification
First Tests
Z-Axis Selection Vs. Bandwidth
Backscatter Imager Summary
• Optimized for single purpose
– less relay optic, no eye piece, no turret, etc.
• Scan system optimized for bulk operation
• Heterodyne detection for improved Z-axis
resolution (full sampling of the optical
phase)
• Multiple wavelengths
• Requires computer controlled alignment
Fluorescent Imager
Optical system, mechanical
components, coherent detection
Design Objectives
• Need to match throughput of backscatter
imager
• Film based input media, accessible from
both sides with symmetric, optical
properties
• Support for optical sectioning
Problems
• Dye saturation
• Signal to noise ratio
• Incoherent signal
Need to maximize light gathering ability
and quantum efficiency of detector
Scan System
• Sorry, the few slides were removed due
to pending (but incomplete) patent
applications
Microtom
Functions, mechanical components
Design Objectives
• Integrated with optical system
• Fully automatic
• Precise control of cutting plane and
related parameters
Transferring fragile Objects to Film
•
•
•
•
•
Microtom with stationary knife
Floating pick-up
Match surface velocity to cutting speed
Minimize surface tension
Electrostatic transfer
Knife Assembly
far end of
cutting area
6.292
4.694
3.017
2.103
2.051
1.82 in.
1.719
1.300
2.700
5.200
5.700
1.300
Instrument Base
Microscope Environment
• -20 to -30 degree C, controlled operating
temperature
• dust free, controlled flow, dry nitrogen
atmosphere
• vibration isolation
• issue: sublimation (prevent by index
matching fluid)
Mechanical components
• Linear air bearings
• Voice-coil motor direct drive
• Laser interferometer position sensing
with 10nm resolution
• Piezo-actuators for knife positioning
Specimen chamber
10.04 in.
3.69 in.
Staining Unit
Outline, Challenges
Film as Sample Slice Carrier
• Candidate: DuPont Cronar 410 polyester
flim, gelatin coated, 100mm
• Need to maintain sample adhesion during
stating process
• Optically clear, substitute for cover-slips
(need to be witting correction range)
• Chemically inert
Continuous Staining Process
• Tight control of process parameters
(temperature, flow rate, chemical
concentrations, etc.)
• Clean-room environment: dust-free, high
purity, no manual steps
• Uniform, predictable distortions
Computational Aspects
Where does all the money go?
Computational Infrastructure
• Instrument control
• Front-end signal processing
Control System Overview
• Linux based
• 3 Functional Blocks:
– System Controler
– Signal Processing
– File system /
Database
Image processing
• Deconvolution (CAT scan, MRI)
• Super-Resolution [Cheeseman et al]
• 3D reconstruction
• Image fusion
• 3D Compression
Circuit Extraction Algorithm
• Linear scan of the data set
• Connectivity function integrated with
Bayesian super-resolution algorithm
Summary of Instrument Capabilities
• 6 band (3 backscatter + 3 fluorescent
labeled) confocal microscope
• 0.2mm (or better) 3D resolution
• Integrated sectioning and staining
• Image acquisition and processing speed
sufficient for large volume scanning
Summary
• Large scale, high resolution scanning of
biological specimen will become practical
• Automated tracing of neurological
systems is conceivable
• Provides a clear, long term research focus
• Significant Research potential
• Intermediate, practical spin-off potential