Dear MNE community,
I would like to determine the correspondence between the sources of a
forward model (say fsaverage, ico 4 5124 sources) and the labels of a
parcellation (say aparc). I found some nice examples on MNE website in
which the activity of the sources are averaged over a given label ( *Generate
a functional label from source estimates
<https://martinos.org/mne/stable/auto_examples/inverse/plot_label_from_stc.html#sphx-glr-auto-examples-inverse-plot-label-from-stc-py>
*)
but I am actually looking for something simpler (for a given parcellation
find the corresponding indexes of the sources).
I believe it should be feasible.
Any ideas?
Thanks in advance
Thierry
def label2idx(src, label):
"""
finding the vertex numbers corresponding to vertices in parcel specified in label
:param src: source spaces
:param label: label object of the desired parcel
:return: parc_idx: indexes of the vertices in label,
parc_hemi_idx: indexes of the vertices in label, but in the corresponding hemisphere
"""
offset = src[0]['vertno'].shape[0]
this_hemi = 0 if (label.hemi == 'lh') else 1
idx = np.intersect1d(label.vertices, src[this_hemi]['vertno'])
parc_hemi_idx = np.searchsorted(src[this_hemi]['vertno'], idx)
parc_idx = parc_hemi_idx + offset * this_hemi
return parc_idx, parc_hemi_idx
I am still stuck with the correspondence of the source space with
parcellation.
I have a much better understanding on how functions&data are organized
after Mina Jamshidi's feedback on that but I am still missing something.
I thought that sources were organized the same way as the parcellation
(source 1 corresponds to vertices 1 in the parcellation etc...). So, for a
given source space (e.g. ico 4, 2562 sources per hemisphere) I would expect
to find the vertices 0...2561 in the parcellation. Although there is a good
correspondence between source, parcellation and lobes I cannot locate all
my sources in the parcellation.
For instance, if I run the following:
import numpy as np
import mne
label=mne.read_labels_from_annot('fsaverage',parc='aparc',hemi='lh')
vertices=np.zeros(0,int)
for lab in label: vertices=np.hstack((vertices,lab.vertices))
vertices=np.sort(vertices)
print(vertices[2561])
I expected to get 2561 instead of 2792 so there is something about my
belief that is wrong.
I am still stuck with the correspondence of the source space with
parcellation.
In general, a source space takes the high resolution mesh (~100k verts) and
subselects a small subset of them (~10k or less). The subset is stored in
`src[0]['vertno']` for the left hemi and `src[1]['vertno']` for the right.
Parcellations are generally defined (by FreeSurfer) on the entire high
resolution cortical surface, and thus should contain many more points than
our decimated source space.
So, for a given source space (e.g. ico 4, 2562 sources per hemisphere) I
would expect to find the vertices 0...2561 in the parcellation.
The parcellation can have holes. In the case of "aparc" the medial wall
vertices are missing:
Also keep in mind that the decimated source space vertices being numbered 0
to 2561 is a property unique to `fsaverage` (and how its spherical mesh is
arranged), in general for other subjects these can and will be integers
spanning the whole ~100k brain space.
Hi Eric,
thanks a lot for the clarification. The last days I went also though the
mailing list and found a "patch" to the problem of matching the sources
with the lobes.
For the convenience of other users with similar issues I post the direct
link to it:
Fill missing vertices in parcellation and save the labels
