[caret-users] Freesurfer to Caret scripts: Trouble with talairach.xfm (MNI transform too big)

Donna Dierker donna at brainvis.wustl.edu
Thu Nov 12 16:12:44 CST 2009

So far, I believe the most likely explanation for the problem I saw with 
the one case I'm testing is an isolated sub-optimal registration from 
orig to MNI305 (or 711-2C in MNI305 space).  It's not a disaster, but it 
looks 5-10% too big.  (The attached B12mritotal_on_mni305.cor.jpg shows 
good fit between the 711-2B_as_mni_average_305_mask segmentation volume, 
mni305.cor anatomical underlay, and mean PALS_B12 MRITOTAL surface.  The 
B12mritotal_on_orig.mni.cor.jpg capture shows the same segmentation 
volume/mask and surface, but the orig.mni anatomical volume, which was 
generated by applying talairach.xfm to the orig.LPI volume.)  Looking 
through the atlas volumes, etc., I have no reason to believe this will 
be an issue on other cases.  I will run more cases on my end, to see if 
this is a systematic problem.  I very much doubt it, because it would 
have surfaced elsewhere.

I abandoned the notion of combining mri_convert steps for two reasons:

* When I tried to combine a transform with a reslice, occipital cortex 
got shaved.
* Concatenating the orig->MNI and MNI->711-2B matrices left me with no 
orig.mni volume to troubleshoot bad talairach.xfm problems.

The extra resampling doesn't worry me, because the resulting 
orig.711-2B.nii volume is more than adequate for the purposes of 
auto-landmarks and vetting surface alignment.

I pulled the mri_convert steps out of the import subroutine, so that 
they don't get run twice (once for left, once for right).  Help requests 
this afternoon have prevented me from properly testing this change, and 
I'll be in an all-day retreat tomorrow.  Back at it Monday.


On 11/10/2009 09:28 AM, Donna Dierker wrote:
> Because there are several users following the progress of these 
> scripts, I wanted to alert them to an issue I came across yesterday.
> I was underwhelmed with the quality of one of the hemispheres 
> auto-landmarks, and I suspected it might be due to the 711-2B fit 
> being sub-optimal.
> While it's too soon to hang the auto-landmark problem on that cause, I 
> did confirm that the orig.711-2B.nii volume being produced is bigger 
> than the typical 711-2B bounding box.  Backing upstream, I confirmed 
> the MNI transform exceeds the typical MNI bounding box.  The 
> B12flirt_on_flirt.jpg illustrates how the mean PALS_B12 surface looks 
> overlaid on the MeanBuckner12_FLIRT volume.  The fit looks good.  The 
> B12flirt_on_orig.mni.jpg shows the same mean PALS_B12 surface overlaid 
> on my subject's orig volume, as transformed to MNI using the 
> talairach.xfm.  The volume bleeds out much further in all directions, 
> but is especially noticeable near the occipital pole.
> It could be a bad talairach.xfm, or it could be my processing sequence 
> (e.g., the reslicing to LPI somehow?).
> I don't want to brush this off, so I'll need a bit of time to study it.
> Donna

-------------- next part --------------
A non-text attachment was scrubbed...
Name: B12mritotal_on_mni305.cor.jpg
Type: image/jpeg
Size: 95975 bytes
Desc: not available
URL: <http://brainvis.wustl.edu/pipermail/caret-users/attachments/20091112/4723ff09/attachment-0004.jpg>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: B12mritotal_on_orig.mni.cor.jpg
Type: image/jpeg
Size: 130575 bytes
Desc: not available
URL: <http://brainvis.wustl.edu/pipermail/caret-users/attachments/20091112/4723ff09/attachment-0005.jpg>

More information about the caret-users mailing list