Near Real-Time Deconvolution for 3D Fluorescence Microscopy Marc - - PowerPoint PPT Presentation
Near Real-Time Deconvolution for 3D Fluorescence Microscopy Marc - - PowerPoint PPT Presentation
Near Real-Time Deconvolution for 3D Fluorescence Microscopy Marc Bruce, PhD CEO, Microvolution GPU Accelerated Deconvolution Software After 3-D Deconvolution Other Deconvolution 16 Seconds 10.7 Min Thin Filaments Are Missing Preserves Thin
GPU Accelerated Deconvolution Software
Microvolution offers faster and more accurate deconvolution
Thin Filaments Are Missing Preserves Thin Filaments: 16 sec
Image Courtesy of Molecular Devices
Other Deconvolution 10.7 Min After 3-D Deconvolution 16 Seconds
Basics of f Fluorescence Microscopy
Sample Objective
Basics of f Fluorescence Microscopy
Illumination Dichroic mirror
Basics of f Fluorescence Microscopy
Emission
Basics of f Fluorescence Microscopy
- Damage to sample from high
intensity light
- Haze from out-of-plane
fluorescence
- Blurriness due to diffraction
reduces resolution of small
- bjects
Derived from Wikipedia, “Light sheet fluorescence microscopy”
Diffraction Limits 3D Resolution
XY Z X Y
Point Spread Function
The Solution: Deconvolution Software
http://www.leica-microsystems.com/science-lab/deconvolution
XY Z
Richardson Lucy Algorithm
Iterative algorithm operating under the assumptions of Poisson- distributed noise
Real-time Deconvolution is Now Possible
- GPU yields speeds up to 200X
faster than old methods
- Allows for real-time
deconvolution
- Optimize image capture
145 215 571 900 0.42 1.5 4.8 7.6
100 200 300 400 500 600 700 800 900 1000
10 MB 253x210x50 190 MB 1080x1080x50 890 MB 2160x2160x50 1246 MB 2160x2160x70
Seconds
Old methods GPU accelerated
Higher Signal-to to-Noise Ratio Yields Better Data
- Analysis of raw FRET images
showed low contrast
- 4x effective increase in contrast
post-deconvolution
- Expected cell features can now be
seen
Miskolci, V., Hodgson, L. & Cox, D. Using Fluorescence Resonance Energy Transfer-Based Biosensors to Probe Rho GTPase Activation During Phagocytosis. Methods in Molecular Biology 1519, 125-43 (2017)
Diffraction-Limited Resolution Limit
In Increased Resolution to 180 nm
20 40 60 80 100 120 140 160 180 0.5 1 1.5 2 2.5 3 20 40 60 80 100 120 140 0.5 1 1.5 2 2.5 3
Tong Zhang and Puifai Santisakultarm, Salk Institute
Microscopy in Dim Light
Be Before 2048 x x 2048 x x 51 Aft fter 3D D De Deconvolution 8.5 .5 sec econds
Image Courtesy of Technical Instruments
Widefield Microscopy
Dennis Hughes, MD, PhD, Assoc Prof of Pediatrics, Jared Mortus, Lead Researcher, Laurence Cooper, MD, PhD, Professor of Pediatrics, George McNamara, PhD, Senior Research Scientist, The Children’s Cancer Hospital, MD Anderson Cancer Center, Houston, TX
Wid idefie ield Im Image Be Before 1024 x x 1024 x x 34 Aft fter 3D D De Deconvolution 1.6 .6 sec econds
Light Sheet Microscopy
Derived from Wikipedia, “Light sheet fluorescence microscopy”
Light Sheet Microscopy
Ligh Light Shee eet Im Image Be Before 946 x x 768 x x 321 Aft fter 3D D De Deconvolution 22.5 .5 sec econds
Dong-Yuan Chen in the Bilder lab, UC Berkeley
Lattice Light Sheet Microscopy
- Dr. Boyd Butler using 3i Lattice LightSheet
LLS LLSM Be Before 1167 x x 512 x x 401 Aft fter 3D D De Deconvolution 10.2 .2 sec econds
In Instant SIM IM
VisiTech
In Instant SIM IM Im Image Be Before 1446 x x 1131 x x 14 Aft fter 3D D De Deconvolution 4.9 .9 sec econds
FINCH (Fresnel in
incoherent corr rrelation holography)
Siegel, N., Lupashin, V., Storrie, B. & Brooker, G. High-magnification super-resolution FINCH microscopy using birefringent crystal lens interferometers. Nature Photonics 10, 802-808 (2016)
Aft fter Propagation and 3D D De Deconvolu lution 10 sec econ
- nds
Wid idefie ield Im Image 2048 x x 2048 x x 21
Usable on All Tiers of f GPUs
- HPC clusters
- Split large multi-channel, multi-timepoint images onto independent nodes
- Let peer GPUs collaborate
- Desktops
- Run alongside acquisition
- Embedded systems
- Currently developing for TX1
Usable on All Tiers of f GPUs
5 10 15 20 25 30 35 Seconds
TX1 K2200 970M K40 P5000 GP100