These scripts and the EEG Data are part of Xie, W., Toll, R., & Nelson, C.A. (2022). EEG Functional Connectivity Analysis in the Source Space. Developmental Cognitive Neuroscience. https://doi.org/10.1016/j.dcn.2022.101119

Background

In the current tutorial and manuscript we aimed to demonstrate recently developed pipelines to conduct EEG FC analysis in the source space focusing on phase-phase synchrony (PPS) (Xie et al., 2019a, b, Dev Sci and BMC Med) and amplitude-amplitude correlation (AAC) (Toll et al., 2020, AMJ; Zhang et al., 2021, Nat Biomed Eng) respectively. The two pipelines were developed by two different research labs -- the Nelson Lab at Harvard and the Etkin Lab at Stanford. You may also find the information about the first and second authors of the current tutorial at WX at PKU ([email protected]) and RT at UTSW Med Center ([email protected]). For the sake of brevity, the two pipelines were referred to as “pl_pps” and “pl_aac” in the following sections.

To use pl_pps, please do the followings

1. Download the SourceSpaceFC folder

   You should see the following files in your folder:

   
   • Go to the "Data" folder.
     • Example data have already been uploaded to the "Age12mos" and "Age36mos" folders. You should be able to see them once you open the folders.
     • To download all the data used in the manuscript, open the "Data download.md" file and then download the data following the instructions. Please save the EEG data for different ages to the "Age12mos" and "Age36mos" folders respectively.
   • Go to the "Sourcemodels" folder. Download the source models and filters following the instructions in the "Download sourcemodels and efilters.md" document and have the files saved in your local "Sourcemodels" folder.
   • Go to the "Programs" folder. Open or run "RunMe_example.m" in Matlab as an example for how to use the scripts.
     • Please download and install EEGLAB and Fieldtrip toolboxes and add them to Matlab search path before running the programs. The recent EEGLAB and Fieldtrip versions (e.g., after 2018) should work. We have the programs tested with the following Fieldtrip and EEGLAB versions:
     • Both fieldtrip-20210409 and fieldtrip-20180415
     • eeglab2021.1
     • To change the parameters and methods used for source space FC analysis, please see the following scripts as an example. Please type "help SourceSpaceFCanalysis_PPS" in your Matlab command window for more information.

       help SourceSpaceFCanalysis_PPS
   % SourceSpaceFCanalysis_PPS performs source-space functional connectivity analysis.
   % This program was written based on the example data in Fieldtrip preprocessing format.
   % Use as
   %   SourceSpaceFCanalysis_PPS(cfg, participantnumber)
   % Inputs:
   %   cfg: The configuration (structure) that contains various parameters:
   %   cfg.foi: frequency of interest. The following boundaries are arbitrarily defined based on the literature;
   %            They can be changed to other values. Please read Xie et al.(under revision)_DCN for more information;
   %            =  'theta' (run the analysis for the theta band)
   %                   .12mos:[3 6], which means 3 to 6Hz
   %                   .36mos:[3 7], which means 3 to 7Hz
   %            =  'alpha' (defualt)
   %                   .12mos:[5 10]
   %                   .36mos:[6 11]
   %            =  'beta'
   %                   .[11 22]
   %            =  'beta'
   %                   .[22 45]
   %   cfg.fcmethod: functional connectivity method. Our program calls ft_connectivityanalysis;
   %            Thus, the majority of the fc options in Fieldtrip should work. 
   %            =   'wpli'(default)
   %                   .weighted phase lag index 
   %            =   'imag'
   %                   .imaginary part of the coherency 
   %   cfg.atlastype: brain atlas used for segmentation.
   %            Only the LPBA40 atlas was provided as an example for this tutorial.
   %            Please find more atlases from John E. Richards' website: https://jerlab.sc.edu/projects/neurodevelopmental-mri-database/
   %            =   'LPBA'(default)
   %                   .lpba40 brain atlas 
   %   cfg.replacearg: whether to replace the existing output file.
   %            =   0 (default)
   %                   .Do not replace the existing file. Return if the file already exists
   %            =   1 
   %                   .Replace the existing file
   %   cfg.parcmethod: How to parcellate the voxels into ROIs.
   %            =   'centroid' (default)
   %                   .Use the voxels near the centroid for each ROI
   %            =   'average' 
   %                   .Calculate the average across all voxels for each ROI
   %            =   'PCA' 
   %                   .Do PCA with all voxels in the ROI and use the first component;
   %   cfg.methodtype: Source localization inverse solution.
   %            =   'eloreta' (default)
   %                   .Use eLORETA
   %            =   'mne' 
   %                   .Use minimum norm estimation
   %   cfg.gridresolution: Source localization "spatial resolution".
   %            This is also an arbitrary value; however, you would need to re-generate the source models if you use other values;
   %            Please follow this link to create new models https://www.fieldtriptoolbox.org/tutorial/sourcemodel/
   %            =   '6mm' (defualt)
   %   cfg.plotarg: whether do plots for certain steps while the program is running.
   %            =   0 (default)
   %                   .No plot. Thanks.
   %            =   1 
   %                   .I like plots.
   %   participantnumber: an arbitrary participant number. Used for loading and saving the data.

2. Optional: Run the source-space FC analysis for all participants and frequency bands

   type edit RunAnalysis4All in the command window This program will run the analysis for all frequency bands and participants in different age groups.

   RunAnalysis4All

To use pl_aac, please do the followings

1. Download the "SourceSpaceFC" folder. You should be able to find the following folders in the "pl_aac" folder.

   
   • The kernels and .mat files are saved in the 'ADMIN' folder, and the preprocessed data are saved in the FINAL' and 'OPEC' folders.

2. Change directory (cd) to the "pl_aac" folder. For example:

   pl_aac_path = '/Users/wanzexie/Documents/GitHub/SourceSpaceFCAnalysis_DCN/SourceSpaceFC/pl_aac';
   cd(pl_aac_path);

3. The Matlab code should be executed as follows: a. Please make sure Brainstorm is installed and added to the search path because pl_aac adopts a few functions in Brainstorm.

   b. Run the babyPublishOrthog.m by typing its name babyPublishOrthog in the command window.

   • Type "help babyPublishOrthog" in the command window for more information. Please see below.

   • The first section of this program includes instructions on how to create the source models/kernels in Brainstorm using child brain templates (O'Reilly et al., 2021, NeuroImage; Richards & Xie, 2015, Advances in Child Dev and Beh). We also made pictures showing the steps to create source/head models in Brainstorm. They can be downloaded from the Sourcemodels/pl_aac_sourcemodels/ folder.

     

• Note: The later connectivity steps between pl_aac and pl_pps are not dependent on the earlier source differences.

• We externally have the babyPublishOrthog.m program tested with the 4 example datasets on a MacBook Pro laptop that has the following system configuration. It took about 8 hours. So please be patient if you are not running the program on a workstation. An alternative is to first try the program with one frequency band: on lines 54 and 131 change "bands = {'THETA','ALPHA','BETA','GAMMA'};" to "bands = {THETA};".

  

• Note: We also provide a program SourceSpaceFCAnalysis_AAC.m that will calculate orthogonalized power correlation (i.e., AAC) with the head models and functions used in pl_pps (e.g., FEM model and Fieldtrip function). The OrthogonalPowCorr.m (called by SourceSpaceFCAnalysis_AAC) was written for this purpose.

• Note: The Signal Processing Toolbox (SPT) of Matlab is recommended to be installed to ensure all the programs would run smoothly. The functions in Fieldtrip and Brainstorm might use functions in the SPT, and our customized scripts also adopt SPT functions, e.g., the "bandpass.m" used to filter data in "SourceSpaceFCAnalysis_AAC" is a SPT function.

Preprocessing programs

In this tutorial we are also sharing our preprocessing programs (primarily automatic). Since the purpose of the paper is not about data preprocessing, the preprocessing programs we shared might need minor adjustments before they can run smoothly.

1. The preprocessing programs used for the pl_pps pipeline were saved in the pl_pps_preproc folder: BaselineDataProcessing_pps.m and DefineArtificialICAs_AdjustAndSASICA.m. These are the programs we modified based on those used for Xie et al.(2019), BMC Medicine. The parameters in these two programs can be tweaked, and please email WX ([email protected]) if you have questions about pediatric EEG data preprocessing.

   • A few functions from FASTER, ADJUST, and SASICA are called by the programs.

2. The proprocessing program used for the pl_aac pipeline was named as babyPublishPreProc. This program was written based on the preprocessing procedures in Toll et al. (2020), AJP. You may also reach out to RT for questions about these procedures.

Plotting

The brain FC figures depicting the results from pl_pps were made with the Surf Ice software.

1. Download and install Surf Ice using the link above.
2. Load a brain image, e.g., "MNI152_2009" by clicking on File --> Open --> Surf Ice --> sample --> mni152_2009.mz3
3. Load a node file, e.g., File --> Open --> BrainNet --> LPBA40.node
4. Load an edge file, e.g., File --> Open --> BrainNet --> LPBA40.edge

The brain FC figures for pl_aac were manually created in Brainstorm.

1. A document that describes the plotting procedure in Brainstorm has been uploaded to the updated "pl_aac" folder on July-24-2022. The document was written by Armen Bagdasarov (Duke University).

Intermediate outputs

In this section, we demonstrate some of the intermediate outputs from SourceSpaceFCAnalysis_PPS.m with one participant's data.

1. Load the EEG_FT data and run ft_timelockanalysis to change the format of the data from "preprocessing" to "timelock".

line 111ff 
%load the EEG data 
eegfilename = ['Experiment 1 ...']; 
load ([eegfilepath eegfilename],'EEG_FT'); 
% change the data format from preprocessing to timelock format
cfg = [];
cfg.keeptrials = 'yes';
cfg.covariance = 'yes';
EEG_FT = ft_timelockanalysis(cfg,EEG_FT);

EEG_FT has the following structure.

2. Cortical source reconstruction --> "sourcedata" that has the following structure

from line 205ff
% The start of the source reconstructure
tic;
disp('Source analysis starts...');
textprogressbar('initialize');
textprogressbar(['num of dipoles finished  ',length(ndipole)]);
for j = 1:length(ndipole)
    % do the source reconstruction dipole by dipole
    % moments represents the source activation in one dipole, and it has 3 orientations/dimensions;
    moments = efilter{ndipole(j)} * catdata_filt;
    % Use Singular Value Decomposition to find out the virtual dimension that explains the most variance of the dipole activation; 
    [u, s, v] = svd(moments, 'econ');  
    m = u(:,1)'* moments; 
    % m represents the "strongest" dimension or orientation of the dipole activation, so the 3-D dipole activation is reduced to 1-D;
    sourcedata.trial(j,:) = m;    
    if (mod(j,1000) == 1)
        textprogressbar(j);
    end
end
analysistime = toc;
textprogressbar('finished');
disp(['The Source analysis took ' num2str(analysistime) 's']);

3. Parcellation of the dipoles across the brain into ROIs --> "roitrialdata" that has the following structure

line 227 to line 325

rpt_chan_time stands for the "number of trials/epoch" x "48 LPBA ROIs" x "time samples (250Hz x 2 seconds)"

4. Frequency analysis of the source-space ROI data --> "EEG_Freq" that has the following structure

line 347ff
%% Frequency analysis
    % A few options for cfg.method in Fieldtrip: {'coh', 'csd', 'wpli', 'wpli_debiased', 'plv','imag'}
        cfg            = [];
        cfg.output     = 'fourier';
        cfg.method     = 'mtmfft';
        cfg.foilim     = bpfreq; % 6 to 11 Hz in this example
        cfg.keeptrials = 'yes';
        cfg.taper      = 'hanning';
        EEG_freq       = ft_freqanalysis(cfg, EEG_pm);

5. Functional connectivity analysis --> "source_conn" that has the following structure

line 357ff 
%% FC analysis
        cfg = [];
        F;
        if strcmp(fcmethod,'imag');
            cfg.method = 'coh';
            cfg.complex = 'absimag';
        end
        source_conn = ft_connectivityanalysis(cfg, EEG_freq);

Other programs

• textprogressbar.m This is a prgram written by Paul Proteus (e-mail: proteus.paul (at) yahoo (dot) com) to show a progress bar during cortical source reconstruction. It is not a necessary piece for the source-space FC programs.

• GenerateTxtFilesForStats.m This program would load the output FC matrix for all participants in one age group (e.g., 12 or 36 mos) and calcuate the average FC between ROIs in the four major lobes, i.e., frontal, temporal, parietal and occipital lobes.

  • To use this program, do the following:

    % The following 'cfg' settings are to calculate the average FC in the theta band for the 12 mos group, using the FC outputs from using a combination of the following parameters: 'AAC' for FC calculation, 'average' for ROI parcellation, and 'eLORETA' for source reconstruction.
    cfg = [];
    cfg.foi              = 'theta'; % 'alpaha', 'beta', 'gamma'    
    cfg.fcmethod         = 'AAC';  % or 'wPLI'      
    cfg.replacearg       = 1; % 1: to replace previously generated file; 0: no  replacement         
    cfg.parcmethod       = 'average';  % 'centroid', 'PCA'   
    cfg.methodtype       = 'eloreta';  % 'MNE'         
    cfg.age              = 12;    
    [fcmatrix_all] = GenerateTxtFilesForStats(cfg);

  • The output of this program is a text file with the participant numbers, age, frequency band, and the FC for different pairs of lobes (e.g., FF refers to the average FC between ROIs within the frontal lobe; FO refers to the average FC between ROIs in the frontal and those in occipital lobes; and lpba is the average FC between all ROIs, i.e., whole-brain FC).

  • This program will also plot the average adjacent matrix across participants.