SPM8SourceEstimation - Meg Wiki

Upload page content

You can upload content for the page named below. If you change the page name, you can also upload content for another page. If the page name is empty, we derive the page name from the file name.

File to load page content from
Page name
Comment
Type the missing letters from: He's no ded, he's jus resing hs eys

location: SPM8SourceEstimation

Source Estimation in SPM8

This is a basic script for source estimation of individual subject data in SPM8. It uses as input the averaged data from SPM8 pre-processing, as well as individual MRI data.

% Root directory for EMEG data
bwd = '/MyDataDir/';
data_name = 'Mmacespm8_block1_raw_ssstf_raw.mat'; % File name for results after averaging (input for source estimation)

%% Define your subject data and MRIs
cnt = 0;

cnt = cnt + 1;
subjects{cnt} = {'meg10_1000', '101224'};   % letter strings for subject-specific paths
MRI_dir{cnt} = '/MRIdirectory/thisone.nii'; % directory with this participant's MRI (*.nii)

cnt = cnt + 1;
subjects{cnt} = {'meg10_1001', '101231'};   % letter strings for subject-specific paths
MRI_dir{cnt} = '/MRIdirectory/thisone.nii'; % directory with this participant's MRI (*.nii) 
% Note: you may have to copy your MRI first (e.g. from /imaging/local/structurals/cbu)

nr_subs = length(subjects);
fprintf(1, 'Going to process %d data sets\n', nr_subs);

%% Define Analysis Parameters 

% Names of conditions after averaging
inv_trls = {'cond1';'cond2';'control'};  

% Time (ms) and Freq (Hz) contrast window (0=all freqs)
con_win = {[0 400]; 0};

% Forward model options, for_typ{EEG/MEG}
for_typ{1} = 'EEG BEM'; % EEG
for_typ{2} = 'Single Shell'; % MEG (mags+grads)

%% Processing options
segment_only = 0;    % Flag whether only MRI segmentation (up to, not including, coregistration) shall be done (1: only segment; 0: do it all)
redoreg_flag = 0;    % Flag whether to redo MRI and registration model (eg if want to change fids)
redoformod_flag = 1; % Flag whether to redo forward modelling (eg if only one for_typ above)
display_flag = 1;    % display results on the fly
write3Dflag = 0;     % write SPM volumes of results

%% Define Inversion Options

% Which sensor types to use
inv_mods = {'MEG';'MEGPLANAR';'EEG'};   % Fusion of mulitple modalities

% Which inversion method to use
inv_typ = 'IID';    % Minimum Norm Least-Squares
%inv_typ = 'GS';   % (MSP)

% Cortical surface option
mesh_size = 2;   % [1-3] for coarse-medium-fine

% Whether to use headshape for coregistration (in datareg)
use_headshape = 1; 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% Source localisation
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

for ss = 1:nr_subs,
    
    % Create individual output directory if necessary
    cd(bwd)
    swd = fullfile(bwd,subjects{ss}{1},subjects{ss}{2});
    try
        eval(sprintf('!mkdir %s',swd))
    end
    cd(swd)

    % current subject's data directory
    subj_dir = fullfile(bwd, subjects{ss}{1}, subjects{ss}{2});
    fprintf(1, '\n Current subject directory: %s\n\n', subj_dir);

    % Load result of data averaging
    data_file = fullfile(subj_dir, data_name);
    D = spm_eeg_load( data_file );

    % Initialise... (if want to be safe!)
    D.inv = {struct('mesh', [], 'gainmat', [])};

    current_formod = 1;

    indF = [];
    val = 0;

    %% MRI processing and coregistration
    if redoreg_flag == 1 | ~isfield(D.inv{1},'mesh') | ~isfield(D.inv{1},'datareg')

        val = val + 1;
        D.val = val;
        D.inv{val}.date    = strvcat(date,datestr(now,15));
        D.inv{val}.comment = {sprintf('%s_%s',inv_typ,char(inv_mods)')};    % remember inversion type and sensor configuration    
        D.inv{val}.mesh    = [];
        D.inv{val}.datareg = [];
        D.inv{val}.forward = [];

        %% Check whether MRI present (care if more than one such *img!)        
        D.inv{val}.mesh.sMRI = spm_select('FPList',MRI_dir{ss},'^s.*\.nii$');
        if isempty(D.inv{val}.mesh.sMRI)
            error('MRI not found')
        end

        %% Check whether inverse-normalised surfaces present
        D.inv{val}.mesh.def  = spm_select('FPList',MRI_dir{ss},'^y_sM.*\.nii$');
        if isempty(D.inv{val}.mesh.def)
            warning('No inverse normalisation parameters found - segmenting may take a while!')
        end
        %% Normalise sMRI (if not done already), and create inverse-normalised surfaces
        D.inv{val}.mesh = spm_eeg_inv_mesh(D.inv{val}.mesh.sMRI, mesh_size);
        if display_flag, spm_eeg_inv_checkmeshes(D); end
        % D.save;
            
        % If coregistration not yet done...
        if segment_only,
            continue;   % Stop processing for this subject here, continue with next one
        end;
              
        % If coregistration done manually and saved...
        newmrifid           = [];
        newmrifid.fid.pnt   = D.inv{val}.mesh.fid.fid.pnt(1:3,:);  % I think these are the fids after "Save"...
        newmrifid.fid.label = {'Nasion';'LPA';'RPA'};
        newmrifid.pnt       = D.inv{val}.mesh.fid.pnt;        % Scalp mesh points from MRI above
    
        meegfid = D.fiducials;
        %% Remove nose points (assume those for which y>0 and z<0)
        meegfid.pnt(find(meegfid.pnt(:,2)>0 & meegfid.pnt(:,3)<0),:) = [];
    
        fprintf(1, 'Coregistering\n');
        D = spm_eeg_inv_datareg_ui(D, val, meegfid, newmrifid,use_headshape);
    
        fprintf(1, 'Done Coregistering\n');

        for ind = 1:length(D.inv{val}.datareg)
            fprintf(1, '%d ', ind);
            d = D.inv{val}.datareg(ind).fid_eeg.fid.pnt - D.inv{val}.datareg(ind).fid_mri.fid.pnt;
            err(ss,ind,val) = mean(sqrt(sum(d.^2,2)));
        end 
        fprintf(1, '\n');

        redoreg_flag = 0;

    else                
        D.inv{val}.mesh    = D.inv{1}.mesh;
        D.inv{val}.datareg = D.inv{1}.datareg;                
    end % if redoreg_flag...

    %% Computing forward model/leadfield
    if redoformod_flag == 1 | ~isfield(D.inv{1},'forward') | ~exist(D.inv{1}.gainmat)

        fprintf(1, 'Creating forward model\n');
        %% Create forward model (BEM) (could conditionalise this bit on modality inverted...)

        D.inv{val}.forward = struct([]);
        for ind = 1:length(D.inv{val}.datareg)
            D.inv{val}.forward(ind).voltype = for_typ{ind};
        end

        fprintf(1, 'Computing leadfield\n');
        D = spm_eeg_inv_forward(D);

        if display_flag
             for ind = 1:length(D.inv{val}.datareg)
                 spm_eeg_inv_checkforward(D, val, ind); pause(3);
             end
        end
        current_formod = val;
        redoformod_flag = 0;
        D.save;
    else
        D.inv{val}.forward = D.inv{current_formod}.forward;
        D.inv{val}.gainmat = D.inv{current_formod}.gainmat;
    end     % redoformod_flag = ...


    %% Invert & Contrast

    D.inv{val}.inverse = [];   % Clear to be safe!
    D.inv{val}.inverse.trials = inv_trls;
    D.inv{val}.inverse.type   = inv_typ;
    %D.inv{val}.inverse.lpf    = 0;
    %D.inv{val}.inverse.hpf    = 40;
    %D.inv{val}.inverse.woi    = [-100 300];
    %D.inv{val}.inverse.Han    = 1;

    D.inv{val}.inverse.modality = inv_mods;

    D = spm_eeg_invert(D);

    D.inv{val}.contrast.woi  = con_win{1};
    D.inv{val}.contrast.fboi = con_win{2};
    D.inv{val}.contrast.type = 'evoked';

    D = spm_eeg_inv_results(D);

    D.con=1; spm_eeg_inv_results_display(D);    % Display first condition

    % Write result to SPM volumes
    if write3Dflag
        D.inv{val}.contrast.smooth = 12;
        % D.inv{val}.contrast.rms = 1;
        % D.inv{val}.contrast.scalefactor = 1;
        D = spm_eeg_inv_Mesh2Voxels(D);
        SourceImgs{val}{ss} = strvcat(D.inv{val}.contrast.fname);
    end

    indF(ss,val) = D.inv{val}.inverse.F;

    D.save;

end;    % for ss = 1:nr_subs,...