2008 Engineering review at Utah:Summary

From NAMIC
Revision as of 20:53, 31 March 2008 by Gerig (talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search
Home < 2008 Engineering review at Utah:Summary

Minutes and Summary of Utah Core-1/Core-2 Brainstorming Meeting:

  • back to agenda [1]
  • Monday 24
  • Kick-off presentations and definition of open issues to be discussed. Main theme: a) Putting modules into workflow pipelines, running such pipelines on a large number of datasets using a batch processing and eventually grid capabilities. b) New methods from the Utah groups to be integrated into Slicer-3. Presentations (see agenda) by the different teams on:
    • Slicer-3 architecture
    • Segmentation
    • DTI analysis
    • Shape analysis
  • Tuesday 25: Topics discussed in details:
    • Dicom to nrrd conversion, in particular with DWI data with different coordinate systems, gradient direction schemes, variable b-values and more. Agreed to run a "hard test" by taking new Siemens mosaic-format, multiple b-value DICOM data (Core-3 UNC autism project) and testing the existing conversion tool.
    • Quality control / Validation: Discussion of need to QC the nrrd DTI data (conversion correct?, tensors in correct orientation?, outlier detection?, artifact detection?). Group defined a need to work on QC methodology for data quality checking, to be run within Slicer-3 but maybe also as a stand-alone command-line as part of the data conversion and correction pipelines.
    • Workflow: Concepts of a workflow environment was discussed, using the example of DTI preprocessing and population-based analysis of DTI data. Need for two concepts, one for chaining of NAMIC modules, and one for running such chains/pipelines in a batch mode. Discussion of Batchmake versus capabilities of Grid computing.
    • Workflow in DWI preprocessing: DICOM to nrrd conversion, motion correction, Eddy current correction, linear registration to anatomical MRI or atlas space, robust calculation of tensor etc.: Would be unique NAMIC contribution to help all researchers. All pieces exist (nrrd Core-2, linear registration with scalar images by Jim Miller and itk modules, transformation of DWI and tensor fields via Francois and Casey, concatenation of transformations as planned (see Summary). Motion and Eddy current correction might be made available as a pipeline in collaboration of Tom Fletcher with Sylvain Bouix and Alex.
    • Workflow in population-based DTI analysis: Discussion of population-based DWI analysis of white matter fiber bundles. Casey Goodlett has all modules in cvs repository, ready to be delivered as NAMIC modules, ready to be used in a NAMIC chaining architecture. Group registration of large set of images is using the MIT atlas-building code (Polina Golland, Sardar), preliminary tests are successful. The prototype of a fully automatic workflow for processing of DWI study is planned for the summer programming week end of June 2008 (see summary). Statistics is so far done with Matlab. Discussion on using "R" as statistics modules in Slicer-3.
    • Statistics in Slicer-3: Discussion shows that there might be an interest to integrate basic statistical analysis into pipelines (DTI, shape, etc.). A candidate statistics package might be "R". Also, communication between files might be best if results would be written as csv files. These files can then be read by Excel, SAS or any type of statistics software.
    • Workflow in Segmentation: Brain tissue segmentation developed by Marcel Prastawa has a very different focus than Kilian's EMS. The package by Marcel is already designed as a workflow, integrating a one-level fully automatic full brain analysis, with arbitrary #of MRI channels, nonparametric classification, skull stripping and bias correction. The package forms the basis for new lesion modeling and segmentation as currently in development (rules on intensity and spatial priors). The shape pipeline can be presented to users via special packages, like a brain-stripping module, a bias correction module, a full brain tissue classification module, etc. Resulting output files (probabilistic segmentation maps, label map, bias-corrected multi-channel MRI, registered atlas) can be exchanged across other packages. Marcel Prastawa will further refine the modularized software and make the modules available via the NAMIC repository, chaining is already prepared and planned for summer (see Summary).
    • Workflow in Shape Analysis: The existing workflow, also developed in close collaboration with MartinStyner UNC, is composed of well-defined ITK-ready modules that need to be put into a pipeline with the new chaining environment (see Summary).
    • NAMIC-kit Repository: Discussions converged on the need for a dedicated NAMIC module repository, which would only accept NAMIC-kit modules that are documented and ready to be integrated into workflows. This is different from the Sandbox which is collecting all possible pieces of software. The group discussed the need for Name-Space for modules, where names would explain the category and main function of modules but would also differ for different version of algorithms. Core-2 will take responsibility over a consistent assignment of categories and names, based on names proposed by Core-1 developers.
    • Slicer-3 proper versus modules: Modules and methods that are ready to go and fit into the Slicer-3 architecture can be integrated after talking to Core-2 (Steve Pieper and Jim Miller.) Core-2 needs to be informed if something new is fully ready.
    • HIGHLIGHTS / SUCCESS STORIES:
      • Deformable Registration: Ross Whitakers new code on b-spline registration could get compiled and was tested on a pair of image data (Steve Pieper).
      • Dicom to nrrd DTI reformatting: The existing, recently developed Dicom to nrrd conversion was tested (Xiaodong) with a most recent Siemens mosaic variable b-value DTI dataset with 25 directions, developed by Bob McKinstry, WU. The tool could successfully ready the gradient directions, b-values and volume orientation. One out of 25 gradient images got wrongly transformed, but it will have to be checked if this is an error of the Dicom data.
      • Chaining of NAMIC Modules / Pipelines within Slicer-3: Marco did early tests with shell scripts that create a proxy to command line modules. this comes close to xml pipelines as e.g. used in Marcel Prastawa's segmentation pipeline.
      • Utah-SCI Dashboard: Will talked to the systems manager at SCI, Nick Rathke, and there are plans to setup a most recent version of an ITK dashboard for testing and regression tests on the Utah NAMIC software.


  • back to agenda [2]