FIX: DTITK nonlinear workflow origin reslicing #2561
Conversation
…resent in some images where the volume gets truncated upon reslicing)
|
Note: in the reslice option you can specify the origin, which is how we had it before. This doesn't solve the problem, though... the workflow must have the origin moved to zero outside of the reslicing node or some images are missing slices when they are resliced. |
|
I would prefer to hold off until 1.0.4, at this point.
…On Sun, Apr 29, 2018, 09:49 Kesshi Jordan ***@***.***> wrote:
@effigies <https://github.com/effigies> and @satra
<https://github.com/satra> is it possible to get this in to the 1.0.3 or
is it too late?
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#2561 (comment)>, or mute
the thread
<https://github.com/notifications/unsubscribe-auth/AAFF8u8wFtws6bjoPWBszI0ij9MsKH9oks5ttcTygaJpZM4TrqMq>
.
|
|
As a non-DTI-TK user, this looks fine to me. My main question would be whether you want to restore the origin at the end of the pipeline, for viewing. (I'm assuming that (0, 0, 0) is in voxel space, not RAS.) |
|
For viewing purposes, I think the important thing is that it matches the fixed image. Now that you mention it, I should probably reslice the moving image/move origin to match the original fixed image (the fixed image has to be resliced to a dimension of power 2 to use the diffeomorphic registration). That way the registered volume would match the fixed image input, instead of having two resliced images (fixed & moving) as output that do not match the input fixed image. Before reslicing/move origin After reslicing/move origin |
Cool, that makes sense to me. |
|
Hi @kesshijordan, sorry to bother you, but is there any chance of finishing this one up this week or next? |
|
No worries, @effigies... thanks for the reminder. I'll put it on my list for this week. |
|
Hi @kesshijordan just a bump. We're going to try to freeze by Friday, if you want to try to get this in by then. |
|
Hi, @effigies. I was trying to add a workflow for applying the diffeomorphic transform (since the origin has to be moved and image resliced, applying the saved transform is not a matter of simply using the saved .df.nii file), but I'm having trouble figuring out why it isn't working. I have the other points we discussed fixed here and took out the apply_transform workflow; I'll put it back in when I manage to debug it. For now I just have a note that you need to reslice/move origin on any new images if you want to apply the workflow transform to another volume. |
Codecov Report
@@ Coverage Diff @@
## master #2561 +/- ##
==========================================
- Coverage 67.13% 64.08% -3.06%
==========================================
Files 333 334 +1
Lines 42539 42642 +103
Branches 5265 5274 +9
==========================================
- Hits 28559 27325 -1234
- Misses 13278 14284 +1006
- Partials 702 1033 +331
Continue to review full report at Codecov.
|
I was testing the workflow on some more difficult registrations (e.g. child-to-adult from different datasets) and found that the reslicing can cause problems when images have origins far from 0,0,0. This is just a few lines addition to the workflow to move all images to origin zero prior to any reslicing.
Fixes #2514 .
Changes proposed in this pull request