Vistas for hardware implementation of SOFI:
High speed imaging for rapid biological processes
Imaging fast, user-friendly and integrate with ease
Dirk Hähnel
III. Institute of Physics – Biophysics
Georg-August-University Göttingen
SOFI Developer Meeting
Göttingen 28th March 2015
Göttingen 28th March 2015
2
Making it easy and user-friendly?
Göttingen 28th March 2015
6. price < 10thd. USD
1. physicist
2. chemicist
3. artifacts
4. image stacks
5. dynamical biosystems
CSDISMRequirements
Palm
Storm
SIM
SSIM
Sted Tirf SOFI
3
Why speed is important?
Göttingen 28th March 2015
atomic scale
0.1 - 1.0 nm
dynamic data
0.1 - 10 ns
molecular dynamics
molecular scale
1.0 - 10 nm
interaction data
Kon, Koff, Kd
10 ns - 10 ms
interactions
cellular scale
10 - 100 nm
concentrations
diffusion rates
10 ms - 1000 s
fluid dynamics
4
• tier 1: interatome
– which molecules talk to each other in networks?
• tier 2: deterministic
– what is the average case behavior?
• tier 3: stochastic
– what is the variance of the system?
Why integration is important?
Göttingen 28th March 2015
6
Fast SOFI: crucial challenges
• subpixel resolution
• linearize brightness
• multiplane imaging
Göttingen 28th March 2015
• timing / speed
• memory allocation
• cumulants computation
• integration
• scaling
physical and experimental challenges: implementation challenges
7
Implementation: timing challenge
Göttingen 28th March 2015
acquisition
image reconstruction
final SOFI image
acquisition image reconstruction
Final SOFI Image
Imaging today: no dynamics Imaging dynamical biological processes
live imaging
real time reconstruction
𝑡 𝑎𝑐𝑞𝑢𝑖𝑠𝑖𝑡𝑖𝑜𝑛 ≥ 𝑡 𝑟𝑒𝑐𝑜𝑛𝑠𝑡𝑟𝑢𝑐𝑡𝑖𝑜𝑛
8
Implementation: memory challenge
Göttingen 28th March 2015
subpixel = more gates
subpixel = more time
cumulants => data swapping
linearization => very complex
start
input frame
SCMOS input:
3 Gpixel => SOFI image
1,5 GByte => SOFI image
SCMOS subpixel:
12 Gpixel => SOFI image
6 GByte => SOFI image
…
end
9
Implementation: cumulants computation challenge
Göttingen 28th March 2015
challenge:
memory space < data 3D stack
𝑛 𝑡ℎ
cumulants are 𝑛 𝑡ℎ
moments
corrected by lower order moments
Tremendous matrix operations
10
Implementation: integration challenge
Göttingen 28th March 2015
features:
• no intermediate data
• open API
• micromanager integration
11
Implementation: scaling system architecture challenge
Göttingen 28th March 2015
hardware schematic software schematic
cam n
cam n-1
hardware bus
cam 0
cam 1
data layer
Buffer
Reconstruction
computation
original
data
intermediate
constant
cam(s)
stream
Intermediate
data
application layer interface
12
Integration and development roadmap
Imaging of dynamical biological systems:
acquisition time > image reconstruction
Göttingen 28th March 2015
camera link
FFT
cumulants
linearization
micromanager
integration
IIIQ 2015
integration
• standard FPGA card
• beta software mid 2015
• setup recipe
• complexity blackbox
13
Acknowledgements
Göttingen 28th March 2015

SOFI Developer Meeting Göttingen 28th March 2015

  • 1.
    Vistas for hardwareimplementation of SOFI: High speed imaging for rapid biological processes Imaging fast, user-friendly and integrate with ease Dirk Hähnel III. Institute of Physics – Biophysics Georg-August-University Göttingen SOFI Developer Meeting Göttingen 28th March 2015 Göttingen 28th March 2015
  • 2.
    2 Making it easyand user-friendly? Göttingen 28th March 2015 6. price < 10thd. USD 1. physicist 2. chemicist 3. artifacts 4. image stacks 5. dynamical biosystems CSDISMRequirements Palm Storm SIM SSIM Sted Tirf SOFI
  • 3.
    3 Why speed isimportant? Göttingen 28th March 2015 atomic scale 0.1 - 1.0 nm dynamic data 0.1 - 10 ns molecular dynamics molecular scale 1.0 - 10 nm interaction data Kon, Koff, Kd 10 ns - 10 ms interactions cellular scale 10 - 100 nm concentrations diffusion rates 10 ms - 1000 s fluid dynamics
  • 4.
    4 • tier 1:interatome – which molecules talk to each other in networks? • tier 2: deterministic – what is the average case behavior? • tier 3: stochastic – what is the variance of the system? Why integration is important? Göttingen 28th March 2015
  • 5.
    6 Fast SOFI: crucialchallenges • subpixel resolution • linearize brightness • multiplane imaging Göttingen 28th March 2015 • timing / speed • memory allocation • cumulants computation • integration • scaling physical and experimental challenges: implementation challenges
  • 6.
    7 Implementation: timing challenge Göttingen28th March 2015 acquisition image reconstruction final SOFI image acquisition image reconstruction Final SOFI Image Imaging today: no dynamics Imaging dynamical biological processes live imaging real time reconstruction 𝑡 𝑎𝑐𝑞𝑢𝑖𝑠𝑖𝑡𝑖𝑜𝑛 ≥ 𝑡 𝑟𝑒𝑐𝑜𝑛𝑠𝑡𝑟𝑢𝑐𝑡𝑖𝑜𝑛
  • 7.
    8 Implementation: memory challenge Göttingen28th March 2015 subpixel = more gates subpixel = more time cumulants => data swapping linearization => very complex start input frame SCMOS input: 3 Gpixel => SOFI image 1,5 GByte => SOFI image SCMOS subpixel: 12 Gpixel => SOFI image 6 GByte => SOFI image … end
  • 8.
    9 Implementation: cumulants computationchallenge Göttingen 28th March 2015 challenge: memory space < data 3D stack 𝑛 𝑡ℎ cumulants are 𝑛 𝑡ℎ moments corrected by lower order moments Tremendous matrix operations
  • 9.
    10 Implementation: integration challenge Göttingen28th March 2015 features: • no intermediate data • open API • micromanager integration
  • 10.
    11 Implementation: scaling systemarchitecture challenge Göttingen 28th March 2015 hardware schematic software schematic cam n cam n-1 hardware bus cam 0 cam 1 data layer Buffer Reconstruction computation original data intermediate constant cam(s) stream Intermediate data application layer interface
  • 11.
    12 Integration and developmentroadmap Imaging of dynamical biological systems: acquisition time > image reconstruction Göttingen 28th March 2015 camera link FFT cumulants linearization micromanager integration IIIQ 2015 integration • standard FPGA card • beta software mid 2015 • setup recipe • complexity blackbox
  • 12.