regression per stc label or for src

[virvw]

If your issue is a usage question, please consider asking on our Gitter channel or on our mailing list instead of opening an issue.

Describe the problem

Describe your solution

Describe an alternative

We (@virvw and @poojaprabhu9) wish to perform the mne.stats.linear_regression in source space.
Ideally, we wish to perform these after morphing all participant data on a fsaverage stc or src so as to build to group statistics. So the idea: one linear regression (across N experimental conditions) per participant in source space then perform a group statistics on the betas. We (with @bgauthie Cereb Cortex 2018) had done something similar but only did this in sensor space and then projected the betas into source space. This time, we wish to perform this directly in source space in order to target specific brain structures (in particular hippocampus).

I can see two a priori options:
(1) perform the regression on all vertices on the full brain (stc) and extract significant vertices p values of betas, and correct for multiple comparisons using spatiotemporal clustering).

(2) perform the regression per label (e.g. HCPMMP-1 labels segmented stc).

However, (1) and (2) do not allow to look at hippocampus. Hence:

(3) perform the regression in full volume (incl. cortex + hippocampus volumes)

Once linear regression has been conducted for all participants, we average their beta value time courses (at the level of vertices or labels) and perform a group average stat on the betas.

Currently, we don't seem to be able to do (1), (2) or (3). Have you already thought about this? If not, how can we help (eventually adding a new def in the main function)

Additional context

Add any other context or screenshots about the feature request here.