[python read_inverse_operator()] Can not find parent MRI location

Mailing List Archive (read-only)
1

With help and a new mne_volume_source_space binary from Junpeng Zhang,
run_meg_volume_tutorial.sh finishes and produces the desired files, namely *
sample_audvis-meg-vol-7-meg-inv.fif*.

Now plot_compute_mne_inverse_volume.py<http://www.martinos.org/mne/auto_examples/inverse/plot_compute_mne_inverse_volume.html>
fails
with "Can not find parent MRI location." This error is raised by
read_inverse_operator() reading sample_audvis-meg-vol-7-meg-inv.fif .

I haven't had much success diving into the code, but from what I gather the
fiff file(?) as explored by fiff/tree.py and fiff/tag.py does not include
the tag for the parent MRI (FIFFB_MNE_PARENT_MRI_FILE=353).

Is there a way to confirm this?
Can a fiff file be easily modified to include it or is there a way to get
python to look in a default location
(environ['SUBJECT_DIR']/environ['SUBJECT']/MRI/
)?

Thanks!
Will

2

hi Will,

I am not sure of what goes wrong on your side.

Can try the python script on the data available at:

ftp://surfer.nmr.mgh.harvard.edu/pub/data/MNE-sample-data-processed.tar.gz

it contains the processed dataset as obtained with the shell scripts
available on line and
the nightly build of MNE.

you can explore the content of a fif file with something like:

mne_show_fiff --in sample_audvis-meg-vol-7-meg-inv.fif

if the python script works with these files I suspect you have an
outdated version of MNE commands.

hope this helps,
Alex

3

I had the same problem using 2.7.3-3268 and mne-python v 0.3-1 with my
data, although it worked fine with the sample dataset.

After switching to 2.7.4-3357, the sample dataset still works, but with my
own data now I get an error about the MRI RAS transformation matrix:
File
"/na/homes/jhouck/.local/lib/python2.6/site-packages/mne-0.3-py2.6.egg/mne/source_estimate.py",
line 799, in save_stc_as_volume
    affine = np.dot(src[0]['mri_ras_t']['trans'], affine)
KeyError: 'mri_ras_t'

I checked with mne_collect_transforms and this transform was present in all
of the MRI sets for my subject, as well as in talairach.xfm, but not in
T1.mgz. I applied it to that file using mri_add_xform_to_header but still
see the same error. Does that transfomation matrix need to exist in T1.mgz
before the source space and inverse solution are made?

Thanks,

Jon

4

hi Jon,

that's weird. I don't really have a clue.

to inspect the content of the generated forward operator you can do
something like this:

mne_show_fiff --in sample_audvis-meg-vol-7-fwd.fif --verbose

to only search for the coordinate transformation:

mne_show_fiff --in sample_audvis-meg-vol-7-fwd.fif --verbose --tag 222

you can then check the difference between the sample subject and yours
and see if the transform is present or not in the fif file.

Alex

5

Hi Alex,

The sample data has transforms for both MRI (surface RAS) -> head and MRI
(surface RAS) -> RAS (non-zero origin); my file had only MRI (surface RAS)
-> head. Re-running the MRI processing steps using 2.7.4 (from
mne_setup_source_space onward) added this tranformation, and a
slightly-modified version of the sample code gives me reasonable output in
volume source space. Thanks!

Jon

6

great news !

thanks for also sharing the success with MNE and Python :slight_smile:

Alex