ICA Components with correlation level

VictorRodriguez · May 24, 2021, 3:59pm

Hi!

I am working with ICA to remove eye blinks, as I'm working with EEG data.

When using SSP, it is displayed the percentage of the amount of artifact that was captured by the component during the decomposition in the components. Also, when usign ICA you correlate the components with "noise" channels (in my case EOG) and sort the components based on that correlation values.

My question is if there is any way to see that correlation level (as you see the % when using SSP). It would be very useful to decide which components to remove!

Thank you in advance.
Victor

Francois · May 28, 2021, 12:53pm

Unfortunately, this measure of correlation is not kept in the current data structures and we have no way to display it after computation.

If you need this information, it is computed in process_ssp2.m:

github.com

brainstorm-tools/brainstorm3/blob/master/toolbox/process/functions/process_ssp2.m#L765


%             proj.Components = Wall;
%             proj.CompMask   = zeros(size(Wall,2), 1);   % No component selected by default
%             proj.Status     = 1;
%             proj.SingVal    = 'ICA';

        otherwise
            bst_report('Error', sProcess, sInputsA, ['Invalid method: "' Method '".']);
            return;
    end
    
    if ~isempty(icaSort)
        y = W * F;
        C = bst_corrn(Fref, y);
        [corrs, iSort] = sort(max(abs(C),[],1), 'descend');
        proj.Components = proj.Components(:,iSort);
    end
    
    % Modality used in the end
    AllMod = unique({ChannelMat.Channel(iChannels).Type});
    strMod = '';
    for iMod = 1:length(AllMod)