Tutorial 2, Part 3 - Site Effects in ABIDE¶
In [1]:
%Add path
run('setup.m');
datadir = fullfile(basedir,'data');
filename = 'ABIDE_controlData_7Sites';
load(fullfile(datadir,filename));
load(fullfile(datadir,'cc200Map2Communities'))
max_recursion_depth(1024)
Exercise - Does global signal induced nuisance covariance matrix show site effects?¶
- Estimate nuisance covariance matrix using mean ROI signal as nuisance covariate
- Apply site effect statistic on nuisance matrices
In [2]:
disp('Fields in dataset')
fieldnames(data{1})
disp('signals contains subjects x brain regions x time-series')
disp(size(data{1}.signals))
namefield = 'dataName';
results = {};
for dataset=[1:7]
disp(data{dataset}.(namefield))
nsubjects = size(data{dataset}.signals,1);
denoised_corr = [];
nuisance_corr = [];
nsr = [];
for subjectno=1:nsubjects
X = squeeze(data{dataset}.signals(subjectno,:,:))';
Y = mean(X,2);
[Sigma output] = covariance.sample_conditional_correlation(X, ...
struct('verbose',false), ...
'nuisance',Y);
denoised_corr(subjectno,:,:) = Sigma;
nuisance_corr(subjectno,:,:) = output.nCov;
nsr(subjectno) = output.NSR;
if(mod(subjectno,10)==0)
disp(['Subject ...' num2str(subjectno)]);
disp('Nuisance to Signal Ratio')
disp(real(output.NSR))
end
end
results{dataset}.denoised_corr = denoised_corr;
results{dataset}.nuisance_corr = nuisance_corr;
results{dataset}.nsr = nsr;
end
In [3]:
concat_corr = @(corr1,corr2)(real(cat(1,corr1,corr2)));
sitecluster = [];
dataset = [1:5];
currtotal = 0;
denoised_corr = [];
nuisance_corr = [];
nsr = [];
for ii=1:length(dataset)
switch dataset(ii)
case 1
clusterval = 1;
case 2
clusterval = 1;
case 3
clusterval = 3;
case 4
clusterval = 4;
case 5
clusterval = 5;
case 6
clusterval = 6;
case 7
clusterval = 7;
end
sitecluster(currtotal+1:currtotal+size(data{dataset(ii)}.signals,1)) = clusterval;
currtotal = currtotal+size(data{dataset(ii)}.signals,1);
denoised_corr = concat_corr(denoised_corr,results{dataset(ii)}.('denoised_corr'));
nuisance_corr = concat_corr(nuisance_corr,results{dataset(ii)}.('nuisance_corr'));
nsr = concat_corr(nsr,results{dataset(ii)}.('nsr')');
end
reshape_corr = @(corr)(reshape(corr,[size(corr,1) size(corr,2)*size(corr,3)]));
denoised_corr = reshape_corr(denoised_corr);
nuisance_corr = reshape_corr(nuisance_corr);
figure;
subplot(1,2,1); imagesc(corr(denoised_corr')); colorbar; axis equal image;
subplot(1,2,2); imagesc(corr(nuisance_corr')); colorbar; axis equal image;
% Apply site detection statistic
[hom_com sep_com mw_com] = compareWithinAndBetweenGroupsSim(corr((denoised_corr+nuisance_corr)'), sitecluster)
communityAgrmntScore=mw_com.stats.zval;
[hom_com sep_com mw_com] = compareWithinAndBetweenGroupsSim(corr(denoised_corr'), sitecluster)
communityAgrmntScore=mw_com.stats.zval;
% Apply site detection statistic
[hom_com sep_com mw_com] = compareWithinAndBetweenGroupsSim(corr(nuisance_corr'), sitecluster)
communityAgrmntScore=mw_com.stats.zval;
Out[3]:
In [4]:
% Plot NSR by Site
figure;
plot(sitecluster,nsr,'o');
xlim([-.5 length(data)+1]);
xlabel('Site');
ylabel('Nuisance to Signal Ratio')
Out[4]:
Reduction in site effects using conditional correlation¶
Nuisance covariates used are principal components from shared within-site variation
In [7]:
results = analyze_abide_siteeffects('within_site')
Out[7]:
Exercise: In some sites the nuisance variation in global signal sometimes contains desired BOLD signal. Find which one?¶
Modify line 33 to change the set of sites used in the analysis and then re-run results = analyze_abide_siteeffects('within_site')
You will find that one of the sites is very different from the other sites. Why?