Converting fiducials read via mne.io.read_fiducials() to BIDS anatomical landmarks

[hoechenberger]

MNE stores fiducials in the mri (surface RAS) coordinate system; however, BIDS requires MRI voxel coordinates. @alexrockhill has created the helpful function get_anat_landmarks() that generates landmarks in the correct coordinate system based on digitization points in an Info and a transformation matrix.

However, I now have a file created via the coregistration GUI (which internally calls mne.io.write_fiducials()), and there is no easy way to store these mri landmarks in my BIDS MRI's AnatomicalLandmarkCoordinates field.

I suppose we could modify get_anat_landmarks() to accept existing fiducials (or a path to a fiducials file), skip the first bits of processing it currently does with the Info and transfomation matrix, and convert the fiducials into the correct coordinate system.

WDYT?

cc @agramfort

[agramfort]

can you show me how the code would look like?

…

Message ID: ***@***.***>

[alexrockhill]

I'm probably missing something but why can't you pass the landmarks directly to write_anat?

[hoechenberger]

I'm probably missing something but why can't you pass the landmarks directly to write_anat?

Because they're not in the correct coordinate system. MNE stores them in mri (which is actually MRI surface RAS, unless I'm mistaken), but write_anat() and the new update_anat_landmarks() expect MRI voxel coordinates.

[hoechenberger]

can you show me how the code would look like?

Yes, in fact, I can try to make a proof-of-concept PR and we can see if it becomes a horrible mess or not ;)
But I won't have time for this until Friday

[agramfort]

no PR needed. You can just paste here some "mock up code snippet" to show how it could be

…

Message ID: ***@***.***>

[alexrockhill]

One suggestion to look into would be whether it would be easier to apply the trans directly to the landmarks like it is done for montage.

[hoechenberger]

One suggestion to look into would be whether it would be easier to apply the trans directly to the landmarks like it is done for montage.

In my case, I don't have a trans, and I may not even have an Info (actually I do have an Info, but there could be situations where I don't)

Based on the landmarks in mri alone + the FreeSurfer reconstructions available, it should be possible to go to voxel space, rigth?

[alexrockhill]

The trans would be from the T1.mgz which you could just load with nibabel and get T1.header.get_vox2ras_tkr and invert it.

You could add a convenience function in _freesurfer.py too like estimate_head_mri_t to get the mri to voxel transform with the subject and subjects_dir.