Hi Cinly,
There's no easy answer here - it depends so much on the images.
I've just been working with some images here to get good registrations
and it has been the bright fat (and bone marrow) signal that has caused
us a lot of trouble. If you have this then it is harder to get BET or registration
to work, as the marrow often pretty much touches the edge of the brain
and in these images it had an intensity quite similar to brain tissue.
So I'm afraid I'm going to have to go with "try it and see" as the only
true answer at this stage.
All the best,
Mark
P.S. bet is just a wrapper script for bet2.
On 8 Sep 2011, at 13:30, Cinly Ooi wrote:
> Dear Mark,
>
> Thank you for your reply. I am lucky to not need the transformation, but your transformation information is useful. I will put it in the FSL notes that we have over here. Sorry if I created more work for you all.
>
> Just for the sake of completeness: Someone told me to raid 'fslvbm_1_bet -N' for its neck removal subroutine. That neck removal subroutine involve BET, standard_space_roi then BET -f 0.4. I tried it (with BET2 in place of BET to get around the 'Cannot find input.nii.hdr') and it works. At present is not that well compared to fslroi->BET because I lose frontal cortex, but that is the norm for us here with standard space transformation and therefore our problem. I also has yet to scrutinize the intermediate images to fine-tune the routine.
>
> My question is, generally speaking, which routine would you recommend? fslroi->BET or BET->standard_space_roi -> BET -f 0.4 ?
>
> Many thanks in advance and Best Regards
> Cinly
>
> On 8 September 2011 10:07, Mark Jenkinson <[log in to unmask]> wrote:
> Dear Cinly,
>
> If you are including the (0,0,0) voxel in your ROI then there should
> be no change to the sform or qform. If you are starting a min
> voxel that is non-zero then it should modify the qform and sform
> to account for this. However, I just noticed in my quick test that
> it is not doing that, so this is a small bug that we need to fix.
> Currently it leaves the qform and sform unchanged, which preserves
> the left-right orientation, but will mean that the mm coordinates
> associated with voxel locations will change. We will get onto that
> and have a patch out soon for this.
>
> In the meantime you can work out the appropriate sform or qform
> and modify them with fslorient. The formula that should have
> been applied in the code is:
>
> new_qform = old_qform * shiftmat
>
> where shiftmat = [ 1 0 0 xmin ; 0 1 0 ymin ; 0 0 1 zmin ; 0 0 0 1]
> (matlab notation - which I hope it clear).
> This has the effect of taking the new voxel at the (0,0,0) location
> and effectively shifting its voxel coordinates to be (xmin,ymin,zmin)
> as it used to be prior to the fslroi call, and hence then the old
> qform can be used. The formula is the same for the sform.
>
> Not sure what you were expecting to see with FLIRT. As you've
> started at (0,0,0) for your ROI I wouldn't expect any change
> as the formula above keeps the qform and sform the same.
>
> All the best,
> Mark
>
>
> On 7 Sep 2011, at 22:09, Cinly Ooi wrote:
>
> > Dear FSL,
> >
> > I am trying to use the BET2-FAST-FLIRT pipeline on MPRAGE (3D structural) data. As part of the process, I need to remove the neck using fslroi before BET2.
> >
> > My question is how does using fslroi to crop the image affect the sform and qform code? The specific instruction I changed using fslroi is y-axis, ymin=0, ysize=175 (original is 240).
> >
> > Since FSL website make a big effort to document Left/right flip on fslorient and other tools [Thank you for doing this. I deeply appreciated this], I am pretty sure the fact that you did not mention anything about L-R flip and qform/sform code means there is no effect at all.
> >
> > I pushed the data through FLIRT and cannot see any difference on FLIRT output.
> >
> > Many thanks in advance and Best Regards,
> > Cinly
> >
>
>
>
> --
> Best Regards,
> Cinly
>
> *****
> Don't bother with footer please. I don't read them and will not be bounded by them.
> It cannot be enforced legally anyway. If it can, then remember this: This footer always triumph yours.
>
|