Visualizing electron microscope 3D reconstructions using Teragrid resources

1
Visualizing electron microscope 3D
reconstructions using Teragrid resources

• Christopher J Gilpin1 and Kelly P. Gaither2
• 1 UT Southwestern Medical Center at Dallas
• 2 Texas Advanced Computer Center

2
(No Transcript)
3
Cancer trends progress report 2005 update

• The use of screening tests to detect cancers
  early may allow patients to obtain more effective
  treatment with fewer side effects. Patients whose
  cancers are found early and treated in a timely
  manner are more likely to survive these cancers
  than are those whose cancers are not found until
  symptoms appear.
• http//progressreport.cancer.gov/doc.asp?pid1did
  2005midvcolchid22

4
How can we improve detection?

• Use targeted markers
• Use markers that give high resolution
• Non-toxic
• Non-invasive

5
A candidate marker systemNano Test Tubes!
Silicon dioxide tubes Containing super
paramagnetic iron oxide particles (SPIO) Tubes
are 500nm x 100nm Particles are 11nm
diameter Surface coated with cancer-cell
targeting molecules
6
Schematic of SPIO-NTT Synthesis
7
Transmission Electron Micrograph of SPIO-Loaded
NTTs
8
Electron microscope
9
(No Transcript)
10
(No Transcript)
11
Cellular tomography VisualisationThe current
method
12
Electron TomographyBack Projection
13
(No Transcript)
14
VisualisationThe current method

• Tomogram
• Modeling
• Result

15
Visualisation

• Maverick at TACC
• Remote access via vnc
• Real-time
• Paraview
• Ensight

16
Nano Test Tubes!
Silicon dioxide tubes Super paramagnetic iron
oxide particles (SPIO) Tubes are 500nm x
100nm Particles are 11nm diameter Particle size,
shape and clustering is important
17
(No Transcript)
18
(No Transcript)
19
Data Issues

• Data is converted to unstructured during
  processing
• E.g. 30MB raw data occupies 6GB memory after an
  isosurface operation
• 30MB raw data cannot be processed on a 32 bit
  workstation
• Raw data sets up to 7GB are already being
  produced.

20
2D and 3D slices
21
Measurements

• The size, distribution and packing of particles
  influences MRI signal
• It may be possible to use tubes with different
  particle packing simultaneously to highlight
  multiple targets.

22
Feature Extraction Techniques

• 3D Sobel Edge Detection
• High-pass filter on image stack (volume)
• Performs a 3D spatial gradient measurement
• Uses convolution masks to perform the filtering
• 3D Hough Transformation
• Used to detect shapes in 3D volumes
• Uses a voting process to determine areas of high
  probability for containing a given feature
• Works best on filtered data (e.g., Sobel
  Transform)
• Computation and memory intensive operation

23
Detecting Features in Electron Microscopy Data
3D Sobel Edge Detector
3D Generalized Hough Transform
Histogram Thresholding
Histogram Thresholding
24
(No Transcript)
25
(No Transcript)
26
Conclusions

• Available software converts data to unstructured
• This causes a significant increase in file size
  and memory requirements
• Software specifically tailored for structured
  data needs to be written
• We have a system to develop into a useful tumour
  marker!

27
Acknowledgements

• Dr. Jinming Gao UT Southwestern
• Dr Heather Hillebrenner UT Southwestern
• Paul Navratil - TACC
• Greg P. Johnson - TACC
• Kevin Colburn - Ensight

28
Final conclusion

• Novel science applications should be encouraged
  and cultivated.
• Use the Teragrid and live longer!