2008 Core 1 Core 3 mtg:notes

Core 3 Presentations

9:30 B&W/Harvard Marek Kubicki, Core 3 Talk. Slides. Stochastic Tractography

Additional discussion:

• Slicer3 release date? Some question regarding proper handling of coordinate frames in Slicer3.
• Discussion of registration and existing/needed tools in Slicer 3. Discussion postponed to registration session.
• Discussion re bias field correction and EM seg -> perhaps moreCore3/Core1 interaction needed

9:50 Queen's U not here yet -> talk postponed

9:50 UNC Rachel Smith talk. Slides. Longitudinal study of brain devel. in autism. Additional discussion:

• Brief discussion of segmentation methods for brain substructure segmentations
• Additional clarification/specifics of the status of the processing of datasets
• Status of Marcel segmentation modules and issues re Core2/1 interaction issues -> Utah (Guido, Marcel) waiting for resolution of some critical Core2 software infrastructure issues e.g. chaining infrastructure design before

they can finalize tools

• Cortical thickness measurment algorithms-> Advantages/drawbacks of different methods. Guido: we should compare methods and see difference in simpler vs. more complex algorithms. Martin: Core 3 intends to look at this issue
• Cortical correspondence-> Martin gives update on local correspondence methods status (particle method with freesurfer preprocessing). Issue: Freesurfer not working for pediatric data. Critical Freesurfer functionality will be reimplemented as NAMIC tool

10:20 break

Jeremy Bockholt (DPB lupus)

• collecting new data/ ongoing clinical study (old data)
• discussion about how DTI sequences are selected
• discussion about analysis of lesion growth and identification in follow up images
  • ability to find follow up lesions
  • growth may merge lesions. lesions may disappear. not expected for lesions to move
• question to test - does lesion information correlate with clinical measures/outcomes
• Discussion of classification comparison for lesions
• Comment on the MICCAI workshop on MS lesions segmentation
• Discussion on methods for segmentation comparison
• Discussion of commercial tool JIM

Core 1 Presentations

John Melonakos & Yi Gao (segmentation)

• prostate segmentation (Yi Gao)
  • Extraction of concave valley of prostate for use as landmark in registration
  • Discussion of comparison to UNC work (Steve Pizer's group) on prostate segmentation
  • 3D Ultrasound (raw vs reconstructed)
• lesion segmentation (John Melonakos)
  • Question on how lesions manifest in DTI images (reduced FA; increased MD)
  • Boundary of CSF around ventricles appears similar to lesion
    • lesions can occur in same region
    • Radiologist heuristic has to do with symmetry to distinguish partial voluming from real lesions
• tissue classification
• label space segmentation
  • Question on meaning of binary vectors representation
  • Explanation of LogOdds representation
  • Comment on work on label spaces in volume rendering
• discussion of tissue labeling versus boundary segmentation
  • itk-snap
  • freesurfer
  • parametrization not available in tissue labeling approaches
  • cortical analysis - thickness and alignment - do we want a surface based approach? hybrid approaches?
  • Integrating DTI into cortical-based analysis frameworks
  • Discussion of how to approach external freely available tools. reimplement in slicer? link to external tools? develop new algorithms?
    • relation to renewal.
    • how core 1 and core 2 interact on this
    • innovation in new algorithm improvement as well as engineering. need to show novelty.
    • validation
      • comparison of methods
      • ground truth - cortical thickness validated by dissection and histology (MGH)
      • reproducibility by making software and data available
• clinical choice of MR sequences - tools which need special sequences
  • UNC - special sequences needed for pediatric imaging. strong time limitations. optimized contrast for pediatric populations
  • BWH - DTI imaging resolution and effect on distortion (high resolutions needed for smaller sequences)
  • multi-site studies some data had to be rejected due to inconsistent protocols across sites. freesurfer sequence requires special agreement with MGH
  • differences from scanner advancement - 1.5 T vs 3 T and bias field inhomogeneity
  • comment - tools should be independent of choice of pulse sequence.
  • retrospective data from clinical scans. need tools that can handle this
  • arguments need to be made by DBPs with regard to tools being independent of scanner sequences
  • open source tools which can be reengineered for different purposes. namic tools should be be reconfigurable. freesurfer cannot be modified.
  • comment that freesurfer can be used with other pulse sequences with some preprocessing; freesurfer has two components 1) tissue classification 2) surfaces
  • want to be able to mix and match classification, bias field correction, surfaces
  • matching across surfaces

Guido Gerig (DTI)

• Utah - DTI DICOM issues
  • Oblique slices not recommended
  • Distortion of EPI may lead to desire for oblique slices
• GA Tech - Tensor reorientation
  • question - tensor invariants instead of reorientation

MIT - stochastic tractography

• • visualization of monte-carlo sampled streamlines
  • what is the level of interest in engineering effort in stochastic tractography? consensus seems to be that it is important
  • what is the analysis on top of stochastic tractography?
    • mask for statistics
  • probability of connection to ROIs
  • multiple connections
  • 5 minute runtime for stochastic tractography
  • not truly MCMC - parameters of tensor are fixed
• MIT - DTI clustering
  • some registration involved
    • distortion not accounted
    • registration of fibers
• Utah 1 - EPI correction
  • currently software not available
  • GA tech working on EPI correction with optimal mass transport
• Utah 1 - volumetric pathway
  • how to determine if path is real? Guaranteed to produce a tract even if none exists. Pathway cost is way of measuring whether patch is meaningful. Similary problem exists in stochastic tractography and GA tech method.
  • probabilistic interpretation of these methods
  • along tract statistics becoming common
• Population-based analysis
  • MIT/Harvard - arc-length ; pointwise statistics
  • Utah 1 - Arcuate in autism
• Utah 2 - population based analysis
  • replacement with open source b-spline approach from Serdard & Polina
  • comparison of linear to non-linear alignment
• Tractography validation
  • What is the outcome?
  • Validation a little early - tools still in development
  • DTI research requires substantial knowledge of DTI and data
  • Project increased discussion among PIs
  • manpower issue
  • NIH pressure to show value added of methods
    • How do we recomend tools for specific problem?
    • Are we close to validation challenge like MICCAI workshop
    • Can we generate synthetic data?
    • need to develop recommendations for usage of methods
    • this is needed for renewal
    • would like input from DBPs on goals for analysis. what would they like to measure?
    • input from animal imaging would be helpful
• Conclusions
  • Core 2 help needed to integrate tools which work for developers. Would like to transition to users.
  • Do users want multiple tools? How do we decide what to expose to users.
  • NITRC and Slicer as repository of tools
  • Overarching design for recommended default pipeline.
  • Tutorials for specific tools versus tutorial on what methods are available for DTI
  • need to develop pipeline for users
  • Good news - all tools needed for pipeline are available
  • Need way to develop integration