Calculating cross-talk functions

Requests
1

Dear all,

I'd like to calculate the cross-talk functions to assess the accuracy of my inverse solution, in my case dSPM. I should be able to calculate this from the leadfield matrix and the inverse matrix. In other words, multiply the imagingkernel from my inverse solution with the gain from the headmodel.

Would that be correct?

I noticed that my results do not contain an ImagingKernel. This is the code that I use for calculation of the sources.

    sFiles = bst_process('CallProcess', 'process_inverse_2018', sFiles, [], ...
        'output',  3, ...  % Full results
        'inverse', struct(...
        'Comment',        'dSPM: EEG', ...
        'InverseMethod',  'minnorm', ...
        'InverseMeasure', 'dspm2018', ...
        'SourceOrient',   {{'fixed'}}, ...
        'Loose',          0.2, ...
        'UseDepth',       1, ...
        'WeightExp',      0.5, ...
        'WeightLimit',    10, ...
        'NoiseMethod',    'reg', ...
        'NoiseReg',       0.1, ...clc
        'SnrMethod',      'fixed', ...
        'SnrRms',         1e-06, ...
        'SnrFixed',       3, ...
        'ComputeKernel',  1, ...
        'replacefile',    1, ...
        'DataTypes',      {{'EEG'}}));

Why does it not provide me with an imagingkernel?

Thanks in advance,

Jord

2

I'd like to calculate the cross-talk functions to assess the accuracy of my inverse solution, in my case dSPM. I should be able to calculate this from the leadfield matrix and the inverse matrix. In other words, multiply the imagingkernel from my inverse solution with the gain from the headmodel.
Would that be correct?

@Sylvain @John_Mosher?

Why does it not provide me with an imagingkernel?

Because you explicitly ask for a full source file (first option: 'output', 3, ... % Full results)
Generate again this script but select the option "Kernel only: shared" or "Kernel only: one per file" (values 1 or 2)

3

@Francois Thank you for your quick reply!

I have been able to calculate CTFs for all my sources using this method. Next, I averaged the cross talk functions over the 60 ROIs (selection from Desikan-Killiany atlas) that I calculated from these sources. Does this matrix correctly represent cross-talk between all the parcels in my analysis?

Another post on this topics mentions whitened data. However, I don't think this is applicable to my analysis. Could you confirm that?
https://neuroimage.usc.edu/forums/t/resolution-matrix/755

Jord

4

I calculated cross-talk functions for my different parcels using the following code. Could someone verify whether this is the correct way to do that?

    CTF = ImagingKernel*Gain_constrained; %Multiply inverse solution with the forward solution to obtain the cross talk functions for all vertices. 
    
    % Calculate the mean cross talk for each parcel
    for j = 1:size(DK_atlas,2)
        for i = 1:size(DK_atlas,2)
            CTF_parcel(j,i,z) = mean(mean(CTF(DK_atlas(j).Vertices,DK_atlas(i).Vertices)));
        end
    end

I visualized the cross talk functions and they look plausible to me, but it would be really helpful if someone could verify the validity of this approach.

The image below shows cross talk with the left caudal middle frontal parcel.

HC001_LDLPFC_CTFmap
HC001_LDLPFC_CTFmap1904×987 108 KB

Best,

Jord

5

@Sylvain @John_Mosher @MartinC @Marc.Lalancette @juangpc @tmedani @pantazis?

6

Hi @jordvink and @Francois ,
I would be very interested, too, to know if it's the correct way, please.

Emmanuelle,

7

Hello there:

This the correct approach for evaluating the point-spread function of the linear source estimator for each source. Cross-talk is different in that it evaluate how the amplitude of a source is affected by others. They are both estimated from the resolution matrix R, which is indeed the product of the linear source estimation matrix (aka ImagingKernel) with the lead field matrix (aka Gain matrix).

If you have access to the following book chapter, it summarizes very well these notions:
https://link.springer.com/referenceworkentry/10.1007%2F978-3-319-62657-4_85-1

8

Thank you. I don't have access to this chapter but I found a prepint:
https://www.biorxiv.org/content/10.1101/672956v1.abstract

1 Like