Amplitude envelope correlation (AEC)

sugata_hisato · September 25, 2019, 10:21am

Hi

I have a question about the amplitude envelope correlation (AEC).
Is Pearson coefficient used for AEC?

Thanks

Francois · September 30, 2019, 10:50am

AEC is computed in bst_connectivity.m:

brainstorm-tools/brainstorm3/blob/master/toolbox/connectivity/bst_connectivity.m#L467


      
                  if (nTime ~= size(sInputB.Data,2))

                      bst_report('Error', OPTIONS.ProcessName, {FilesA{iFile}, FilesB{iFile}}, 'Files A and B must have the same number of time samples.');

                      CleanExit; return;

                  end

                  % Remove average

                  if ~isempty(sAverageB)

                      sInputB.Data = sInputB.Data - sAverageB.Data;

                  end

              % Else: Use the same info as FileA

              else

                  sInputB = sInputA;

              end

          end

          % Get the sampling frequency

          sfreq = 1 ./ (sInputA.Time(2) - sInputA.Time(1));

          % Round the sampling frequency at 1e6

          sfreq = round(sfreq * 1e6) * 1e-6;

          nA = size(sInputA.Data,1);

          nB = size(sInputB.Data,1);

          % Number of sources if B is kernel-based

          if ~isempty(sInputB.ImagingKernel) && ismember(OPTIONS.Method, methodsKernelBased)

github.com

brainstorm-tools/brainstorm3/blob/master/toolbox/connectivity/bst_connectivity.m#L841


      
                      error('Unknown connectivity method.');

              end

              nWin = 0;

              % Add the number of averaged samples & files to the report (only once per output file)

              Message = sprintf('Estimating across %d time samples', nWinLenSamples); % samples are not independent due to bandpass filter

              if ~strcmpi(OPTIONS.OutputMode, 'input') && nFiles > 1

                  Message = [Message sprintf(' per file, across %d files', nFiles)];

              end

              if ~isempty(Time)

                  Message = [Message ', for each output time point'];

              end

              bst_report('Info', OPTIONS.ProcessName, unique({FilesA{iFile}, FilesB{iFile}}), Message);

          end

          

          % Process one band at a time to minimize memory requirements.

          for iBand = 1:nFreqBands

              switch OPTIONS.tfMeasure

                  case 'hilbert'

                      % Band-pass filter in one frequency band

                      DataBand = process_bandpass('Compute', sInputA.Data, sfreq, BandBounds(iBand,1), BandBounds(iBand,2), 'bst-hfilter-2019', OPTIONS.isMirror);

                      % Analytic signals (original + i * Hilbert transform)

In a simplified way, the measure used for the correlation is something like:
sum((A - mean(A))/sqrt(sum(A - mean(A))^2) * (B - mean(B))/sqrt(sum(B - mean(B))^2))

Which is (I think) a Pearson's correlation coefficient: