Applying source imaging kernel on Morlet/Hilbert

jafnan · April 6, 2022, 9:31pm

Hope you are doing well. I am Jawata Afnan, from Grova lab. I have an inquiry regarding this tutorial-https://neuroimage.usc.edu/brainstorm/Tutorials/TimeFrequency#Full_cortical_maps

Here it says we can apply Morlet TF decomposition on full cortical map, and it shows an option called 'Optimize the storage of the time frequency file' with Hilbert, explaining that these two processes TF(Inverse(Recordings)) = Inverse(TF(Recordings)) can be used interchangeably. And when this option is activated, it applies the "wavelet transformation to the sensor recordings, and then multiply the wavelet complex coefficients by the inverse operator (ImagingKernel)".
I have two questions here:
First: I do not see this option 'Optimize the storage of the time frequency file' in my Brainstorm, is it in a specific version of Brainstorm?
Second: in the second step, when it applies inverse operator on complex wavelet coefficients, how does it use the noise covariance?
Could you please tell me which piece of process is called here? as I cannot reproduce this option on my side, I do not know how this is working.

Thank you for your time,

Regards,
Jawata

Francois · April 11, 2022, 2:51pm

First of all, note that time-frequency decomposition of full brain maps is complicated to handle technically and very difficult to explore. Reducing the dimension of the data is always advised (e.g. studying only a limited number of scouts, or only one frequency band at a time...)

First: I do not see this option 'Optimize the storage of the time frequency file' in my Brainstorm, is it in a specific version of Brainstorm?

This option is only available when processing files stored in the compact form (inverse kernel + data). The input files must be LINKS (if you show their contents, they should show the field ImagingKernel set, and the field ImageGridAmp empty).
https://neuroimage.usc.edu/brainstorm/Tutorials/SourceEstimation#On_the_hard_drive

Second: in the second step, when it applies inverse operator on complex wavelet coefficients, how does it use the noise covariance?

The noise covariance was already use in the computation of the imaging kernel.

Could you please tell me which piece of process is called here?

github.com

brainstorm-tools/brainstorm3/blob/master/toolbox/timefreq/bst_timefreq.m#L629


      
          if ~isempty(iGoodChannels)

              iBadChannels = setdiff(1:size(F,1), iGoodChannels);

              if ~isempty(iBadChannels)

                  TF(iBadChannels, :, :, :) = 0;

              end

          end

          % Clean memory

          clear F;

          

          % ===== REBUILD FULL SOURCES =====

          % Kernel => Full results

          if strcmpi(DataType, 'results') && ~isempty(ImagingKernel) && ~OPTIONS.SaveKernel

              % Initialize full time-frequency matrix

              TF_full = zeros(size(ImagingKernel,1), size(TF,2), size(TF,3), size(TF,4));

              % Loop on the frequencies and tapers

              for itaper = 1:size(TF,4)

                  for ifreq = 1:size(TF,3)

                      TF_full(:,:,ifreq,itaper) = ImagingKernel * TF(:,:,ifreq,itaper);

                  end

              end

              % Replace previous values with new ones

jafnan · April 12, 2022, 9:38pm

thank you!