Construction of an atlas

preprocessing

#1

Hello MRtrixers,

I would like to create an atlas of TWIs. I also used the population_template script on FODs data and it works well.

However I have two questions on the pipeline:

  1. population_template can works with FA images, FODs,TWIs. Which one is the better to create an atlas? Is there any documentation on the different pipelines?

  2. To obtain the TWI in my template space, I think I have 3 possibilities: a) estimate the tractograms directly on the FOD in the template space then estimate the TWI, b) estimate the tractograms and the TWIs in the patient space then apply the transform to the template space, c) estimate the tractograms on the patient space then apply the transform to the template space and estimate the TWI.
    Which one is the better or suggested?

I saw this recommendations for the HCP in previous post on the forum:
“The general recommendation would be to normalize the streamlines to template space, and then generate the ‘number of connections’ image in template space. This is preferable to warping a density image, as the orientation dependence of density modulation based on expansion / contraction in the warp field will be performed correctly.”

Thanks by advance
Best regards
Félix


#2

Hi Félix,

  1. It is important to differentiate between the image information used to drive registration, versus the image information transformed to population template space in order to generate the group mean image. For instance, you can generate a population FOD template driven by FOD information, but then use the estimated non-linear transformations to generate a population mean FA template image. I think generally the recommendation would be to use FOD images to drive the registration, since it provides better alignment of white matter structures. There is some research on using different image contrasts to drive the registration, but that has an implicit assumption that aligning subjects based on that contrast is preferable to a purely anatomical alignment; that has not yet been proven for TWIs.

  2. Personally I believe that the “best” approach is to:

  • Estimate tractograms in subject spaces.
  • Estimate TWI factor per streamline in subject space.
  • Transform subject tractogram to template space.
  • Generate TWI in template space.
    This however requires the ability to pre-compute the TWI factor per streamline, and provide those to tckmap, which is part of a body of work that is not yet ready for release (ISMRM2017 abstract 64).
    In the absence of availability of this feature, it’s hard to tell what the best compromise is. If your TWIs involve a voxel-wise mean statistic, then the differential modulation of streamlines density during non-linear transformation gets regressed out anyway, so that’s not a major concern, and you could probably just compute your TWIs in subject space and then do an image transformation to the template. That, and warping the streamlines & quantitative image to template space before computing the TWI in template space, would probably perform similarly.
    You definitely do not want to compute tractograms on subject FODs after warping them to template space: you will make @dchristiaens very unhappy. :stuck_out_tongue_closed_eyes:

Cheers
Rob