From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On
Behalf Of Versace, Amelia
Sent: Wednesday, April 06, 2016 12:35 PM
To: [log in to unmask]
Subject: Re: [FSL] principal diffusion direction, dyads1 and mean_f1?
Dear FSL experts,
For my study, I am using a triple tensor model (bedpostx –n 3) and I am looking at the dyads1 and mean_f1 images to find a way to select those( and only those) brain regions where I do expect to find the highest/higher colinearity (i.e. just one primary diffusion
direction) of the fibers.
I thought of using mean_f1 between 0.9-1, but could not find any voxel in the brain that has the 90% of its volume occupied by ‘dyads1’. Even in the genu or splenium of the corpus callosum there are at least 2 fibers in each voxel. Is this what I should expect?
It makes sense to me that in the large majority of the brain voxels, the sum of mean_f1, mean_2 and mean_f3 constitutes less than 20-40% of the volume. Of course, the most of these voxels fall ‘outside’ of white matter areas. However, even in brain regions
where I would expect to find 100% of white matter (eg., genu of the CC), the mean_fsum doesn’t go higher than 80% of the volume of these voxels. What’s the other 20%??
The partial volume fractions (f1,f2,f3) should not be interpreted as the fraction of the voxel occupied by a tract. Rather, they are equal to the fraction of the signal that can be explained by a stick response function in the ball and stick model (see
Behrens et al. MRM 2003, Neuroimage 2007).
In this triple tensor model, a reasonable lower threshold of mean_f1 should be .33, which, in my mind, should be the lowest possible value of mean_f1 in the context of a 100% white matter region (100%
of mean_fsum; … assuming it exists). However, what would be the best way to select only those voxels where I do expect to find ‘only
one’ principal diffusion direction of the fibers based on the multiple-tensor information (files in the bedpostx folder)?
A similar question (what would be the best way to select only those voxels where I do expect to find a triple crossing
of the fibers?) applies for mean_f3, where, for the same reasoning, the highest possible value, in my mind, should be should .33.
You may also find it useful to calculate the fraction of a fibre orientation relative to the others (e.g. f1/(f1+f2+f3)) which you can easily calculate with fslmaths.
The approach we mostly follow is to use model selection (via an ARD prior - see Behrens et al 2007) to determine the complexity at any given voxel. For example, if f2 and f3 are <0.01 or something like that, then we assume a single fibre orientation in
a voxel, as supported by the data and given all our model assumptions. (these last two points are important!)
PS: I noticed that in the axial view of the 1st volume
of the dyads1 images there is a clear laterality effect where the anterior portion of the corpus callosum shows positive value in one side and negative values in the contralateral. This trend is inverted in the posterior regions. I wasn’t able to identify
such a pattern in volume2 and volume 3 with respect to superior-to-inferior or anterior-to-posterior axes. Can you explain why?
I also looked at the dyads2 and dyads3 and couldn’t see any laterality effect even in the 1st volume. Should I expect something like this in dyads2 and dyads3?
You should visualise the dyads using “vector mode” in fslview. Looking at them as images is not very helpful as you can only see one of the x,y,z components of the vectors, and there is a sign ambiguity on the vectors.
Cheers,
Saad
Amelia Versace, MD
Loeffler Building, room 215
121 Meyran Avenue
Phone: 412.383.8131
