Dear Martin,
what you see is, as you suspect, a sampling artefact. FA changes quite rapidly (in space) and there are sheets with high FA that are literally just a couple of mm thick sitting next to areas of low FA. When we then sample that at resolutions of ~2mm^3 we severly undersample the signal. When we then interpolate this signal with a kernel that has negative side lobes the problem is further aggravated.
On top of this you also have a non-linear warping in TBSS that can further increase the effective upsampling, e.g. in cases where the CC is particularly thin in a given subject and then gets upsampled to match the CC of the template/target.
My guess is that if you look at your subjects you will find that this problem is most severe in subjects who's CC is thin to start with, and in the areas where the original planes cut the CC approximately diagonally (i.e. if you look at a sagittal reformatting of the original image volume the CC will run ~diagonally in the yz-plane.
I don't think resampling it a little less (1.5mm^3) is going to make any real difference (over and above cosmetically). What we ultimately need is higher resolution diffusion data. I know you have very good physics support for the diffusion there so hopefully in a not to far off future.
Jesper
On 21 Aug 2013, at 09:43, SUBSCRIBE FSL Martin Uppman <[log in to unmask]>
wrote:
> Dear all users,
>
> I use the TBSS toolbox in order to analyse FA maps. My raw FA maps are sampled at (2x2x3) mm, and they look rather OK. However, after running the tbss_3_postreg command to perform registration to the target image (no mather if it's the FMRIB58_FA or a study specific target), and upsampling to (1x1x1) mm, many of the subjects' FA maps gets affected in a strange way (please see attached print screen). Mainly, the corpus callosum gets chopped off (FA decreases towards 0).
>
> I have gone through the archieve but hasn't found anyone posting this issue before.
>
> I guess it's mostly related to the upsampling. Would there be any idea to downsample the FMRIB58_FA target to (1.5x1.5x1.5) mm instead? Is there anyone else experience the same?
>
> All comments are highly welcome and appreciated!
>
> Best regards,
> Martin
> <tbss_artefact_upsampling.png>
|