Time Frequency (complex) and projection on default anatomy

davidetabarelli · October 28, 2019, 5:23pm

Dear all,

I would like to compute a time frequency map at the sensors level (Wavelet transforms) and then keep the complex values, instead of reducing to power.

My goal (to spare space and computing time) is then to transform the sensor-space TF map using directly the imaging kernel A. Since I use a current density estimation (MNE) and the Fourier transform is linear this (I think) is possible.

The next step is to select only a few scouts from the source space projected complex TF map. The problem is that the atlas I'm currently using is defined only on the default anatomy. The I have to project (in some way) the source TF map to the default anatomy. I think this is possible by using the interpolation matrix B that is computed by Brainstorm every time you project something on the default anatomy.

Question: is this procedure correct? Where I can find, or how can I compute, the interpolation matrix?

Thank you.

D.

Francois · October 29, 2019, 8:53am

My goal (to spare space and computing time) is then to transform the sensor-space TF map using directly the imaging kernel A. Since I use a current density estimation (MNE) and the Fourier transform is linear this (I think) is possible.

Yes, this works. This solution is offered in the process options when computing time-frequency maps on sources ("Optimize storage: Yes, save morlet(sensors) and inverse kernel").
https://neuroimage.usc.edu/brainstorm/Tutorials/TimeFrequency#Full_cortical_maps

The next step is to select only a few scouts from the source space projected complex TF map. The problem is that the atlas I'm currently using is defined only on the default anatomy.

Where did you get your subject anatomy from? If it was processed with FreeSurfer or CAT, you should have the same atlases available in your subject and in the MNI template. If it was not processed with FreeSurfer or CAT, you don't have any reliable way to get a correspondence between your subject anatomy and your template.
https://neuroimage.usc.edu/brainstorm/Tutorials/CoregisterSubjects

I have to project (in some way) the source TF map to the default anatomy

You could also project the scouts you need on the individual anatomy. This is a lot lighter solution.
If you don't need full source time-frequency maps on the MNI template for group analysis, don't compute them at all.

Where I can find, or how can I compute, the interpolation matrix?

The interpolation matrix is computed in function tess_interp_tess2tess.m. This is not easy to use manually... See how this is done from bst_project_sources.m and bst_project_scouts.m

davidetabarelli · October 29, 2019, 9:09am

Thank you very much for your quick response Francois.

The scouts are defined on the default anatomy and comes from FreeSurfer/HCP. The default anatomy I'm using is a fsaverage on which I imported the HCP-MM atlas from Glasser et al., 2016 (https://www.nature.com/articles/nature18933). Thus everything (registered spheres etc..) are available. However, I don't have the atlas in the individual surfaces.

As for the interpolation matrix, for another project where I was using Fieldtrip, I computed it using the same approach as in the Matti's MNE manual (morphing) and used it to switch from individual to group space by matrix multiplication. Is it computed differently in Brainstorm? How I'm supposed to project my results for a group analysis?

Thanks !

D.

Francois · October 29, 2019, 9:30am

The scouts are defined on the default anatomy and comes from FreeSurfer/HCP. The default anatomy I'm using is a fsaverage on which I imported the HCP-MM atlas from Glasser et al., 2016 (A multi-modal parcellation of human cerebral cortex | Nature). Thus everything (registered spheres etc..) are available. However, I don't have the atlas in the individual surfaces.

Everything should work then!
Project your scouts of interest from the template to the individual subjects, and do a ROI-based analysis. Then you don't need to any additional interpolation for the group analysis.

I computed it using the same approach as in the Matti's MNE manual (morphing) and used it to switch from individual to group space by matrix multiplication. Is it computed differently in Brainstorm?

This is described in this tutorial:
https://neuroimage.usc.edu/brainstorm/Tutorials/CoregisterSubjects#Using_FreeSurfer_registered_spheres

If you need additional details, refer to the code of tess_interp_tess2tess.m:

github.com

brainstorm-tools/brainstorm3/blob/master/toolbox/anatomy/tess_interp_tess2tess.m#L318


      
          iScoutDest = find(strcmpi(sScoutSrc.Label, {sScoutStructDest.Label}));

          if isempty(iScoutDest)

              disp(['PROJECT> Warning: Structure not found in destination surface: ' sScoutSrc.Label ]);

              continue;

          end

          sScoutDest = sScoutStructDest(iScoutDest);

          % Is it a cortex region

          isCortexL = ismember(sScoutSrc.Label, {'lh', '01_Lhemi L', 'Cortex L'});

          isCortexR = ismember(sScoutSrc.Label, {'rh', '01_Rhemi R', 'Cortex R'});

          

          % ===== USE FREESURFER SPHERES =====

          % Interpolate using the sphere and the Shepard's algorithm

          if isCortexL && isFreeSurfer

              Wmat(sScoutDest.Vertices, sScoutSrc.Vertices) = bst_shepards(vertSphLdest, vertSphLsrc, nbNeighbors, 0);

          elseif isCortexR && isFreeSurfer

              Wmat(sScoutDest.Vertices, sScoutSrc.Vertices) = bst_shepards(vertSphRdest, vertSphRsrc, nbNeighbors, 0);

              

          % ===== USE BRAINSUITE SQUARES =====

          % Interpolate using the Brainsuite squares and the Shepard's algorithm

          elseif isCortexL && isBrainSuite

              % Interpolation: Subject => BrainSuiteAtlas1

GiorgioM · November 7, 2019, 11:32am

Hi Francois,

I work with Davide and I'd like to have a feedback on example pictures of some of the projected scouts after I followed your workflow suggestion:

for some scouts and some subjects the projection transformation seems to distort the scouts, but maybe the distortion is acceptable:

3D_19851030GBGG_L_STSda.tif (161.6 KB)
3D_19901008CNBU_L_A5.tif (194.0 KB)
3D_19901008CNBU_PBelt.tif (193.3 KB)
3D_19901209FBTM_L_PI.tif (186.7 KB)
3D_19901209FBTM_STSda.tif (186.3 KB)

Is this normal? Can be due to the small parcels of the atlas? Is there any way to improve the result of the projection?

Thanks

Giorgio

Francois · November 7, 2019, 12:17pm

This is difficult to judge, as you don't include what the original scouts looks like.

You can't expect this solution to be perfect, especially on low-resolution surfaces. If you want to get a better idea of how one scout projects from the template to the subject, do it using the high-resolution surfaces.
And keep in mind that the goal of this projection is to distort your ROI between different brains that have different shapes.

If your goal is to do some functional analysis using these ROIs, they are maybe a bit small an inhomogeneous spatially.

GiorgioM · November 7, 2019, 12:49pm

Thanks Francois for the fast and precise answer.

My mistake, I include the original scouts now:

3D_Group_analysis_L_A5.tif (183.6 KB)
3D_Group_analysis_L_PBelt.tif (182.3 KB)
3D_Group_analysis_L_STSda.tif (182.2 KB)

I understood that the goal is to distort the scouts to adapt to a new brain shape, I just didn't expect that some vertices would be projected also "outside" of the main scout boundary.
Thanks for suggesting the high resolutions surfaces to get a better idea of the capability of the projection, I didn't think about that.

Yes the goal is also to perform functional analysis using these ROIs, so I appreciate you gave your opinion on the quality of these.

Cheers
Giorgio

Francois · November 12, 2019, 12:29pm

These projected scouts look really good to me.
They are however very small and may not be able to capture the brain activity of interest. Make sure you really observe what your are expecting to get in these small regions for a few subjects before you do all your group analysis with them. Or smooth the cortical maps first.