TubeTK/Events/2010.07.26: Difference between revisions

From KitwarePublic
< TubeTK‎ | Events
Jump to navigationJump to search
(Created page with '= Andinet = * Primary goal: Data from Duke for BWH * Accomplishments ** Attend NAMIC AHM ** Determine what is necessary to record data sent to OpenIGTLink from VectorVision syst…')
 
No edit summary
Line 19: Line 19:
* Primary goal: Bump and dent identification on IC images
* Primary goal: Bump and dent identification on IC images
* Accomplishments
* Accomplishments
** Traveled to SSRL to view the acquisition and meet with Greg and Mike.
** Modified the GenerateFeatures application to handle the input of arbitrary feature images
** Using the previous features and Casey's new patch-based features, was able to achieve 95% pixel level accuracy in Weka and 23/25 defects found with 0 false positives in image space (after morphology).
** Explore new features
** Explore new features
*** z-score values from three different mean/stdDev joint histograms: add, subtract, and unchanged
*** evaluate a variety of standard deviations for intensity and ridge computations
*** evaluate a variety of standard deviations for intensity and ridge computations
** New centerline method (skeletonization)
* Near Term (Aug 2)
** GenerateFeaturesForWeka
** Receiving code to simulate the tomography directly on GDS Layers
* Near Term (Aug 1)
** Compute dot-product between line (hessian) tangent and normal directions in ES and GDS images
** Get registered data from Greg
** Subselect features
** compute dot-product between line (hessian) tangent and normal directions in ES and GDS images
** write program that goes from Weka output to image and computes TPR/FPR scores on that image
** Collaborate with Casey
*** Choose classification scheme
*** Implement in C++ or python - in tubetk
**** Use neuralnets / parzenWindowing in ITK
*** Subselect features
** Product: ~ 5 slides / report to USC illustrating path chosen, strengths, and weaknesses.
** Product: ~ 5 slides / report to USC illustrating path chosen, strengths, and weaknesses.
*** Real-world tests / workflow
*** Real-world tests / workflow
Line 39: Line 34:
**** i.e., do we need to insert modifications for training on every slice / acquisition / ?
**** i.e., do we need to insert modifications for training on every slice / acquisition / ?
**** Normalizing for inter-acquisition (or inter-slice) variations?
**** Normalizing for inter-acquisition (or inter-slice) variations?
** Go to Synchrotron
** Work with new collaborator at Kitware.
* Medium term (1 months, August 15)
* Medium term (August 15)
** Delivery and education
** Delivery and education
** Can we get better in simulation?
** Can we get better in simulation?

Revision as of 16:20, 26 July 2010

Andinet

  • Primary goal: Data from Duke for BWH
  • Accomplishments
    • Attend NAMIC AHM
    • Determine what is necessary to record data sent to OpenIGTLink from VectorVision system
      • Define data workflow and software architecture
      • Begin implementation
      • Product: powerpoint presentation: 5 slides
  • Near term (August 1)
    • Install VV at Duke
    • Determine if we can get US data from Duke machine
    • Write IJ article
      • Cite grant proposal in article
  • Medium term (1.5 months, August 15)
    • Investigate simulation of ultrasound from MR/CT (Talk to Stephen First :) )
    • Code review of vessel segmentation method from Stephen

Patrick

  • Primary goal: Bump and dent identification on IC images
  • Accomplishments
    • Traveled to SSRL to view the acquisition and meet with Greg and Mike.
    • Modified the GenerateFeatures application to handle the input of arbitrary feature images
    • Using the previous features and Casey's new patch-based features, was able to achieve 95% pixel level accuracy in Weka and 23/25 defects found with 0 false positives in image space (after morphology).
    • Explore new features
      • evaluate a variety of standard deviations for intensity and ridge computations
  • Near Term (Aug 2)
    • Receiving code to simulate the tomography directly on GDS Layers
    • Compute dot-product between line (hessian) tangent and normal directions in ES and GDS images
    • Subselect features
    • Product: ~ 5 slides / report to USC illustrating path chosen, strengths, and weaknesses.
      • Real-world tests / workflow
      • Does a trained classifier work on other layers?
      • Does a trained classifier work on other acquisitions?
        • i.e., do we need to insert modifications for training on every slice / acquisition / ?
        • Normalizing for inter-acquisition (or inter-slice) variations?
    • Work with new collaborator at Kitware.
  • Medium term (August 15)
    • Delivery and education
    • Can we get better in simulation?
    • Connectivity analysis

Casey

  • Primary goal: Compare populations of vascular networks
  • Near term (0.5, Aug 1)
    • Collaborate with Patrick
      • Feature extraction
        • Patch-based features (max, median, quantiles)
        • Scales / neighborhood
      • Feature research
        • Width estimate from Stephen (concern, time)
        • Location of local max in ridgeness
        • Subsample centerlines
        • Optimal match filter
      • Classifiers
        • Comparison
        • Implementation for transfer to USC
        • Hierarchy (Good/bad. If bad, then add/sub.)
    • Data for Retinas
  • Medium term (1 months, Aug 15)
    • Review previous processing pipeline with Stephen
    • Research on methods for comparing spatial graphs / adjacency matrices
    • Begin Port and test existing adjacency code
    • Process retinal data
    • Complete port and test of existing adjacency code
      • Prepare IJ article

Hua

  • Primary goal: ultrasound image processing
  • Accomplished
    • Verify Andinet's code: add tests and help with IJ publication
      • Creating tests (sin pattern with known derivatives)
  • Near term (August 1)
  • Medium term (1.5 months
    • Investigate use of speckle in ultrasound registration
    • Model-based deformation field interpolation
    • 2D-3D registration (data)
    • Simulating ultrasound from MR/CT