Hi,
If I am getting it right, you have registered raw diffusion data
across subjects by means of a deformable registration method? I guess
that this mandates "rotation" of each voxel's data (e.g. reconstructed
ODFs) based on the local warp field, rather than rotating the bvecs
which will have a unique global effect for each diffusion direction
image.
For the second question, I guess that scaling the data is not a good
option, as this can confound the calculation of distribution of
diffusion parameters in the following steps.
But, this is a question that I really like to hear an expert comment
as much as you do!
Cheers,
Sourena
On 7/10/12, Frederick Damen <[log in to unmask]> wrote:
> Hello FSL Users!
>
> I've been working on producing a population based DTI template, and was
> hoping to get some feedback on two particular portions of my processing. So
> far, I have projected the raw diffusion data from a set of subjects into a
> population specific template space, and rotated the bvecs for each subject
> accordingly. All of the data/bvecs/bvals in template space were concatenated
> and then processed using dtifit (bedpostx is currently running to help
> validate the calculation of crossing-fibers and probabilistic
> tractography).
> I can explain the template production process more in depth if needed, but
> in general I used a iterative set of non-linear registrations on T1w images
> with a convergence algorithm to avoid any bias towards one subject. The
> results right now look very promising; however, I want to make sure I am
> covering all my bases.
>
> 1) When I project the raw data for each subject into the template space
> (non-linear warpings), what would be the most appropriate interpolation
> method? I am currently employing sinc, but have not specified a sinc window
> or width.
>
> 2) When the data from each subject is projected into template space, do the
> signal intensities of the raw data have to be normalized across the set of
> subjects before tensor calculations? I do not mean that the b=0 and b=1000
> images should be on the same level, but rather multiply each subjects set of
> diffusion data be a respective ratio that normalizes the b=0 images across
> subjects.
>
> Just as a side note, I am fully aware that trying to do template based
> diffusion analysis/tractography is a new and frequently disputed field;
> however, I believe that I have produced reasonable validation measures so
> far and for the larger project at hand, population-based tractography would
> be the most appropriate method for analysis (we are working on cross-species
> network analysis). That being said, any questions/criticisms regarding the
> methods I have presented so far are welcome, as well as any answers to my
> questions, as they will help me to produce the best results possible for my
> overarching project.
>
> Thank you,
>
> Frederick "Freddy" Damen
>
|