SIFT2 with no ACT


#1

Dear all,

Would it make sense to use SIFT2 without ACT? I am running an experiment where I am not convinced that I would like to use the ACT constraints for a variety of reasons.

Would it still make sense to use SIFT2?

Claude


#2

It can be used without ACT, but some awareness of the ramifications is worthwhile.

  • If ACT is not used, and a three-tissue decomposition is not possible, and no explicit processing mask is provided to SIFT2, then those streamlines that project through the cortex will be assigned greater weights than those that terminate closer to the GM-WM interface. The non-zero FODs in the GM need to be “explained” by the tractogram: if their size is comparable to those in the WM, but only a subset of the streamlines in the WM go partly / all the way through the GM, then those streamlines going through the GM need to be up-weighted in order to reconstruct the fibre density reported to be present there.

  • Any subsequent processing of the streamlines data following SIFT2 (e.g. connectome construction) would typically be more difficult / contain errors if ACT were not used.
    This is not technically directly tied to SIFT2 itself, but is often where the bulk of experimental issues arise with such propositions.

For the former point you could potentially somewhat alleviate the issue by using the -proc_mask option (note this can actually receive a floating-point image, e.g. containing partial volumes, it doesn’t have to be a binary mask) to regulate which voxels contribute to the SIFT2 model (and how much) v.s. which will be ignored.

Rob


#3

Hi Rob,

So to alleviate the first problem would it be possible to use the ACT tissue types for SIFT2 but not for the tractography? I want a situation where I do not want constrain my tractography but still use SIFT2 in the best possible way.

Would the command be something like:
tcksift2 -act 5TT.mif in_tracks in_fod out_weights ?

Or would I still need -proc_mask ?

Further how does SIFT2 deal with deep grey matter?

Claude


#4

tcksift2 -act 5TT.mif

Yes, you could still do that even if ACT had not been used during tractography. The only purpose of the -act option in both tcksift and tcksift2 is to automatically derive an appropriate processing “mask” from the 5TT image, given it’s typically defined on a different image grid to the FODs. I have a private command 5ttregrid that does just that step, and providing the resulting image via the -proc_mask option would behave exactly the same.

Or would I still need -proc_mask?

Since both that and the -act option are used to define / derive a processing mask, they are mutually exclusive.

Further how does SIFT2 deal with deep grey matter?

It gets excluded from the optimisation. This is because I believe it is possible for there to be fibrous connections contained entirely within such deep GM structures (i.e. don’t traverse the WM at all): while such fibres would contribute to the DWI signal and hence the FOD amplitude, they are expressly forbidden from reconstruction within ACT; therefore driving a correspondence between streamlines density and FOD amplitude is no longer appropriate.