Algorithms:Core1Visit May24:DTI

Comments from Casey's Talk:

• One way to validate registration techniques is to assess how well primary eigenvectors align.
• Reluctant to optimize the registration on the features for which you desire to statistically analyze.

Comments from Tom's Talk:

• Presentation of an argument for higher angular diffusion imaging (via the analysis of aligned v.s. unaligned tensors).

Comments from Bruce's Talk:

• Comments on ex-vivo imagery
• How accessible is this nice data?
  • It would be nice to use this high resolution data as a ground truth for diffusion studies.

Comments from Polina's Talk:

• Is there any guarantee that these clusters will be compact?
  • There is some hope for this due to the distance measure that is used. It's not obvious.
• Is this sensitive to the choice of sigma?
  • Look to Lauren's thesis which will be published next week. Every aspect was evaluated and shown to not be too sensitive and to be in a reasonable range.
• What if you limited the tracts are constrained to have a limited length?
  • Then you might be able to get better results at fiber crossings.

Comments from Allen's Talk:

• The property of the Finsler metric ensures that the curve evolution does not prefer the tangent direction over the curve shortening.
• Discussion on the minimum of curves in the Finsler geometry.
• Use of particle systems in image segmentation - resolved the issue in the non-convex case.