Failed to find suitable seed point after 100000 attempts

Dear Friends,

I am doing Anatomically Constrained Tractography. I use this command:

tckgen out_wm.mif my_track.tck -seed_image seed_mask.nii.gz -include brain_outside_seed_mask.nii.gz -act 5tt.mif -number 0 -fslgrad bvecs bvals -nthreads 4 -info -maxlength 600

I divided my Hippocampus-L, gotten from AAL, in four clusters of voxels and built the masks, one for each cluster.
I can load and see my masks, voxels, clusters and the whole Hippocampus on FSL View.
Both masks, seed_mask and brain_outside_seed_mask, are on DWI space, as my out_wm.mif.
The 5tt.mif anatomical image is on Anatomical space.

When I run this command, I got a result for all first three clusters/masks. I can visualize the tracks using MRView.
But when I run the command with the fourth mask, I receive this error message:

tckgen: [ERROR] Failed to find suitable seed point after 100000 attempts - aborting.

I did not figure out what is happening, because in principle there is no difference between the masks.

Do you have any ideas? I appreciate.


Joao V. Dornas

Hi there Joao,

Whenever ACT is used, every seed point selected must be verified against the 5TT image to make sure that the seed point is valid given the anatomical constraints. In the case of -seed_image, this means that the (trilinear-interpolated) white matter tissue fraction must exceed the GM tissue fraction. My guess is that for this particular mask, none of the voxels within the mask are satisfying this criterion, hence the error message. Furthermore, there’s a good chance that for the other three parcels, only a fraction of your input masks are actually contributing streamline seeds; much of their volumes probably also lie within voxels labelled as cortical grey matter.

For your particular experiment, I would suggest re-running 5ttgen with the new -sgm_amyg_hipp option. That will take the explicit segmentations of the amygdalae and hippocampi (from either FSL or freesurfer, depending on your 5ttgen algorithm selection), and label them as sub-cortical grey matter, therefore allowing streamlines to propagate into these structures rather than terminating them at the interface with white matter. As of MRtrix3 version 0.3.13, tckgen now also permits streamlines to be seeded within sub-cortical grey matter, whilst still applying all of the ACT priors: for any streamline seeded within sub-cortical GM, one of the unidirectional streamline propagations must enter white matter, but not both, in order to be accepted given the anatomical priors and written to the output file. So I think this setup is more likely to exhibit the tracking behavior you’re expecting to see.


Thanks Rob. I will see which solution best suits my needs. I am trying right now the whole hippocampus, instead of dividing it in clusters.