2010 Winter Project Week WM ATLAS
Input from example subject: Automatically segmented fiber tracts and fMRI peak locations.
Output: Search region for arcuate based on fMRI locations. Yellow region is prediction of arcuate location in this patient.
Key Investigators
- Lauren O'Donnell (Instructor, Harvard Medical School (BWH))
- Carl-Fredrik Westin (Professor, Harvard Medical School (BWH) )
- William Wells (Professor, Harvard Medical School (BWH) )
- Alexandra J Golby (Professor of Neurosurgery, Harvard Medical School (BWH) )
- Gordon L. Kindlmann (Professor University of Chicago) (teem help/docs/implementation)
- James Malcolm and Yogesh Rathi (2-tensor tractography)
- Steve Pieper (advice on ipython and module implementation)
Objective
Eventual goal: Implement slicer module for neurosurgical planning using fMRI+DTI atlas.
Intermediate goals: Investigate re-implementation of tract-based morphometry pipeline in 3D Slicer, including clustering for atlas creation, atlas-based WM (white matter) segmentation and whole-brain WM coordinate system generation. Investigate integration of FMRI+DTI into atlas pipeline, for neurosurgical planning applications.
Approach, Plan
The original WM atlas pipeline consists of (over a group):
- Preprocessing
- DTI and FA image calculation (slicer)
- congealing registration of FA images (Lilla's code)
- whole-brain tract seeding (slicer2)
- application of registration to tractography (matlab)
- measurement of fiber-fMRI model (slicer2/matlab)
- Atlas generation (matlab)
- clustering of (sample from) whole-brain tractography in group
- expert labeling of clusters to give anatomical structures
- the above produces an atlas that can be used to segment
- WM tract segmentation (matlab)
- based on atlas, plus fMRI in future for IGT
- analysis of segmented WM (matlab)
- optimal generation of tract-based coordinate systems per anatomical structure
- export of cluster/structure based measurements (FA, etc)
- some statistical analysis
We propose re-implementation of much of the above in python and/or Slicer3 framework. This will enable use of the pipeline by others and our investigation of DTI+fMRI atlases for neurosurgery.
Progress
Our current goal is to decide what dependencies the software should have and whether re-implementation in python is feasible.
- Preprocessing
- initial implementation of BatchMake modules for tensor calc, but found ipython is preferable for overall pipeline
- tested congealing implementation in ITK from NAMIC SandBox to replace our current use of original code by Lilla Zollei
- unable to output mean images due to code crash
- discussions with Sandy, Lilla indicate ITK congealing broke with new ITK version, unclear when this was
- implementation of initial seeding pipeline script in ipython
- current pipeline uses slicer command line modules that call teem/vtk
- Gordon has implemented tractography methods and some probing capability in teem, may allow more direct teem/itk/vtk dependencies
- Clustering
- looking into available packages in python
- IGT fMRI+DTI atlas
- Initial methods development: fiber-fMRI distance model (MICCAI Diffusion workshop 2009)
References
This work is based on the papers
Defining Spatial Relationships Between fMRI and DTI Fiber Tracts Lauren J. O'Donnell, Laura E. Rigolo, Isaiah Norton, Carl-Fredrik Westin, and Alexandra J. Golby. MICCAI Workshop on Diffusion Modeling and the Fibre Cup, 2009
Tract-Based Morphometry for White Matter Group Analysis. Lauren J. O'Donnell, Carl-Fredrik Westin, and Alexandra J. Golby. NeuroImage 45(3):832-844, 2009
Automatic Tractography Segmentation Using a High-Dimensional White Matter Atlas. Lauren O'Donnell and Carl-Fredrik Westin. IEEE Transactions in Medical Imaging 26(11):1562-1575, 2007
