Hi Josue,
my impression is that seeding from pure grey matter areas of the cortex is more problematic than seeding from the same anatomical region but including white matter areas (grey-white matter interface)
Personally I would consider “the grey-white matter interface” to be a different thing to “pure grey matter areas including white matter areas”. The latter implies a volume, whereas the former is typically assumed (in the absence of pathology) to be an infinitely-thin surface. So it’s not clear here whether you are trying to make an argument that is wholly independent of ACT and associated mechanisms, which makes answering slightly difficult.
Maybe the presence of U fibers produce some kind of “noise” that prevent easy propagation to the subcortical regions?
For simplicity, yes; but more below:
My question is what is the physical reason that seeding from grey matter in the cortex is more problematic for the propagation of the fibers to the subcortical areas?
Particularly in the absence of a three-tissue deconvolution, there is a lot of DWI signal from the grey matter that contributes to FOD anisotropy and affects determination of orientations in the white matter due to partial volume (even if the signal contribution is relatively isotropic, with CSD this tends to lead to many spurious peaks rather than an isotropic increase in FOD amplitude). This can cause a lot of very short connections, whether due to “noise” or due to streamlines “in the WM” following fibre orientations that actually “originate from the GM” but are sampled due to interpolation.
Where streamlines coming from deeper WM into such a region will tend to stick to the incoming orientation and/or anything within a narrow range of angles, when seeding from this location you are sampling from the full range of possible orientations at that point, and therefore are more likely to latch on to some spurious orientation for commencement of the streamline.
The problem is potentially exacerbated by a detail of iFOD1 / iFOD2 that I noted some time ago but have seemingly not listed anywhere. During streamline propagation, FOD amplitudes have two sources of influence. Firstly, the amplitude must be above some cutoff threshold in order to even be considered. Secondly, rejection sampling is used so that orientations where the FOD amplitude is large are more likely to be followed; orientations where the FOD amplitude is smaller (but supra-threshold) can still be followed, just with reduced probability. For streamline seeding however, currently only the first of these applies. This has the effect that for a given spatial location and orientation in the image where the FOD amplitude is small, that orientation is far more likely to be tracked if that vertex is a seed than it is if that vertex is just another point along the streamline. Whether this is right or wrong is I think open to interpretation.
It was this kind of observation in conjunction with the known biases of seeding homogeneously throughout WM that drove me to implement the dynamic seeding approach; however that is only applicable to whole-brain tractography, whereas from your description I’m inferring that you’re attempting to use targeted tractography…
Cheers
Rob