Damien Rajon
SCIA 2009
Siemens Corporate Research _ Feb ...
Presentation Outline
Company presentation
Internship topic
Technical aspect
Background knowledge
Actual work
Co...
The company
The company
Siemens AG
Founded in 1847 by Werner Von Siemens
One of Europe's largest conglomerate
Over 480 000 employees i...
The company
Siemens Corporate Research
Founded in 1977
Located in Princeton, New Jersey
Research and development center...
The company
Imaging and Visualization
60 Research scientists
Medical image analysis and visualization solutions
Improv...
The company
Organization hierarchy
Internship Topic
Internship topic
Context
Locate time consuming parts of the image
reconstruction algorithm
Optimize them
Create a sho...
Internship topic
Organization
Informal and formal
status reports
Meetings with
image
reconstruction
experts
Technical aspect
Technical aspect
Magnetic Resonance Imaging
Why MRI ?
How does it work ?
k-space vs. image space
Limitations and solu...
Technical aspect
GRAPPA
pMRI technique
Reconstruction occurs in k-space
Under-sampling according to a Reduction factor
Technical aspect
GRAPPA
Image data
ACS data
Technical aspect
GRAPPA
Reduction factor (R) R 1x 2x 3x 4x
relative to SNR SNR 1 0.7 0.5 ...
Technical aspect
GRAPPA - Weights computation
Technical aspect
GRAPPA - Weights computation
How to combine all the source data
in order to obtain the target ...
Technical aspect
GRAPPA - Weights computation
Technical aspect
GRAPPA - Weights computation
Source data
Technical aspect
GRAPPA - Weights computation
Target data
Technical aspect
GRAPPA - Weights computation
Weights
Technical aspect
GRAPPA - Missing data estimation
Shift the same kernel
in the image data
Use the computed
weights to es...
Technical aspect
GRAPPA - Image output
All the k-space data is available
Inverse Fourier Tranform
Images from each co...
Technical aspect
GRAPPA - Image output
Technical aspect
Image Calculation Environment
Classified as confidential
Framework used within MRI systems
Complex serv...
Technical aspect
Optimization - Profiling
Log profiling
Source code analysis
Creation of log parsing scripts (AWK, shel...
Technical aspect
Optimization - Weights computation
Cholesky decomposition used to solve the Aw=B
system
8210937600 a...
Technical aspect
Optimization - Weights computation
3 implementations
to meet different
needs
Huge performance
gain
Technical aspect
Optimization - Weights computation
Technical aspect
Showcase
Build a standalone application
Further optimizations
Conclusion
Conclusion
Work on bleeding edge pMRI techniques
Work on an actual product
Knowledge in medical imaging
Difficulties
Reve...
Conclusion
Professional outcome
Improved medical imaging knowledge
Real scale project
Understand and understood
Pers...
Thank you for listening
Any questions ?
of 34

Prez Test

ttest
Published on: Mar 4, 2016
Published in: Education      Business      
Source: www.slideshare.net


Transcripts - Prez Test

  • 1. Damien Rajon SCIA 2009 Siemens Corporate Research _ Feb 09 - Aug 09 Final year internship presentation Optimization of image reconstruction algorithms
  • 2. Presentation Outline Company presentation Internship topic Technical aspect Background knowledge Actual work Conclusion
  • 3. The company
  • 4. The company Siemens AG Founded in 1847 by Werner Von Siemens One of Europe's largest conglomerate Over 480 000 employees in 190 countries Focused on three main sectors Net income of 4, 038 billion euros
  • 5. The company Siemens Corporate Research Founded in 1977 Located in Princeton, New Jersey Research and development center Provides technological solutions to other Siemens businesses Experts and scientifics from around the world
  • 6. The company Imaging and Visualization 60 Research scientists Medical image analysis and visualization solutions Improve vision and rendering methods of the inside of the human body Improve current imaging and analysis techniques
  • 7. The company Organization hierarchy
  • 8. Internship Topic
  • 9. Internship topic Context Locate time consuming parts of the image reconstruction algorithm Optimize them Create a showcase for these optimizations and present it to the client
  • 10. Internship topic Organization Informal and formal status reports Meetings with image reconstruction experts
  • 11. Technical aspect
  • 12. Technical aspect Magnetic Resonance Imaging Why MRI ? How does it work ? k-space vs. image space Limitations and solutions
  • 13. Technical aspect GRAPPA pMRI technique Reconstruction occurs in k-space Under-sampling according to a Reduction factor
  • 14. Technical aspect GRAPPA Image data ACS data
  • 15. Technical aspect GRAPPA Reduction factor (R) R 1x 2x 3x 4x relative to SNR SNR 1 0.7 0.5 0.4
  • 16. Technical aspect GRAPPA - Weights computation
  • 17. Technical aspect GRAPPA - Weights computation How to combine all the source data in order to obtain the target data ?
  • 18. Technical aspect GRAPPA - Weights computation
  • 19. Technical aspect GRAPPA - Weights computation Source data
  • 20. Technical aspect GRAPPA - Weights computation Target data
  • 21. Technical aspect GRAPPA - Weights computation Weights
  • 22. Technical aspect GRAPPA - Missing data estimation Shift the same kernel in the image data Use the computed weights to estimate missing k-space data
  • 23. Technical aspect GRAPPA - Image output All the k-space data is available Inverse Fourier Tranform Images from each coil are combined by taking the root of the Sum Of Squares
  • 24. Technical aspect GRAPPA - Image output
  • 25. Technical aspect Image Calculation Environment Classified as confidential Framework used within MRI systems Complex server-client architecture Image calculation dedicated environment Reverse engineering
  • 26. Technical aspect Optimization - Profiling Log profiling Source code analysis Creation of log parsing scripts (AWK, shell script, python)
  • 27. Technical aspect Optimization - Weights computation Cholesky decomposition used to solve the Aw=B system 8210937600 additions and multiplications to build matrix A Reference application uses naive implementation Use of Intel’s MKL
  • 28. Technical aspect Optimization - Weights computation 3 implementations to meet different needs Huge performance gain
  • 29. Technical aspect Optimization - Weights computation
  • 30. Technical aspect Showcase Build a standalone application Further optimizations
  • 31. Conclusion
  • 32. Conclusion Work on bleeding edge pMRI techniques Work on an actual product Knowledge in medical imaging Difficulties Reverse engineering a massive project Lack of medical imaging knowledge
  • 33. Conclusion Professional outcome Improved medical imaging knowledge Real scale project Understand and understood Personal outcome Improvement in English Work in the USA Work in a renown company Meet interesting people from around the world
  • 34. Thank you for listening Any questions ?