[pymvpa] Surface searchlight problems (alignment related)

Nick Oosterhof n.n.oosterhof at googlemail.com
Fri Jun 5 13:40:52 UTC 2015


just a quick note about this earlier problem that Mike encountered. It has been resolved; see below for details.

On 14 Feb 2015, at 01:11, Mike E. Klein <michael.e.klein at gmail.com> wrote:

> I’m still stuck, but perhaps getting closer to the solution [to use pymvpa2-prep-afni-surf] 
> Part I: attempt to align T1 to epi, then align surfaces to aligned T1

The solution included two components:

1) alignment does not work well if the input functional data is oblique.

    3dcopy func.nii func_orig.nii     # make a backup
    3drefit -deoblique func.nii 

Note that "3drefit -deoblique” will change the spatial locations of the voxels. This is fine in this case, as the alignment now uses the new location, and surfaces are aligned based on this new data. 
I’ve added this as in issue to PyMVPA to fix in the future [1].

2) for his particular data, the standard cost function did not result in good alignment.

Solution (after some trial and error): use the mutual information cost function.

In pymvpa2-prep-afni-surf this can be achieved by adding the following option:

--aea_opts "-cmass cmass+xyz -big_move -cost mi” 

[1] https://github.com/PyMVPA/PyMVPA/issues/327

More information about the Pkg-ExpPsy-PyMVPA mailing list