Fixed the fieldcoeff issue and also removed the linear transforms that a...

[thelxinoe]

...re added to the warps. These are unnecessry

[coveralls]

Coverage remained the same when pulling feae48c on thelxinoe:master into 7410d05 on nipy:master.

[satra]

@thelxinoe - thanks for these changes - have you checked these against what feat does?

[thelxinoe]

Yeah, actually, there are more changes but I didn't want to make them
because maybe they are a little drastic. I have problems using this because
my T1 voxel size is too low <1mm. This is a problem because the apply warp
will suck up all the ram and then crash. In feat they don't do this. They
just combine the affine transformations and then apply them. This means
that it converts to the standard template which in this case is usually
2mm. This is much more reasonable and will run successfully. I don't really
want to say that it's a bug but it's maybe something to be aware of. Hence
I didn't make the changes.

Also, I removed the addition of the affine transformations because they are
already taken in to account in to the warpfield coefficients. Applying them
again causes major deformation of the registration. You can verify this
here:

http://fsl.fmrib.ox.ac.uk/fslcourse/lectures/practicals/reg/

On Fri, May 2, 2014 at 1:16 PM, Satrajit Ghosh notifications@github.comwrote:

In nipype/workflows/fmri/fsl/preprocess.py:

@@ -1188,8 +1189,7 @@ def create_reg_workflow(name='registration'):
warpmean = warp2anat.clone(name='warpmean')
register.connect(warp2anat, 'out_file', warpmean, 'in_file')
register.connect(inputnode, 'target_image', warpmean, 'ref_file')

• register.connect(anat2target_affine, 'out_matrix_file', warpmean, 'premat')

@thelxinoe https://github.com/thelxinoe - thanks for these changes -
have you checked these against what feat does?

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/840/files#r12226723
.