Dear Kami,

In Matlab, SVD with one output gives only the singular values, which I think are not what you want. Try instead:

[U,~,~] = svd(temp3);

U = U(:,1);

See attached a little function that does the trick, perhaps more efficiently for large matrices (in case you have one).

All the best,

Anderson

2013/1/27 kambiz rakhshan <[log in to unmask]" rel=nofollow target=_blank ymailto="mailto:[log in to unmask]">[log in to unmask]>

Thanks a lot!

For some reasons, I am also planning to carry this procedure out in Matlab. Can you please let me know if the following piece of code is correct to compute indices, for instance, across the entire DM network:

Lets assume that temp is m*n 4D matrix in which m is number of subjects and n is the number of voxels:

% lets mask each subject-specific DM network

for
i=1:size(temp,1)

temp1=temp(i,:);

temp3(i,:)=temp1(mask);

end

s=svd(temp3);

If the above piece of code is correct, I do not see any variability in the scores at all (see attached figure). Can you please let me know what could be the problem? Note that I used Intensity normalization during preprocessing with no-scaling option during back-reconstruction.

Many thanks

Kami

--- On Sun, 1/27/13, Anderson M. Winkler <[log in to unmask]" rel=nofollow target=_blank ymailto="mailto:[log in to unmask]">[log in to unmask]> wrote:

From: Anderson M. Winkler <[log in to unmask]" rel=nofollow target=_blank ymailto="mailto:[log in to unmask]">[log in to unmask]>

Subject: Re: [FSL] Connectivity Indices for a network as a whole
To: [log in to unmask]" rel=nofollow target=_blank ymailto="mailto:[log in to unmask]">[log in to unmask]
Date: Sunday, January 27, 2013, 7:53 AM

Dear Kami,

About the eigenvectors: We used SVD in Matlab, but perhaps the easiest now is simply use fslmeants. After the dual regression, concatenate the spatial maps in a single 4D file, have the mask you want at hand, then call fslmeants with the option --eig. The output is saved to a textfile that you can use for further analyses.

About scaling, we used z-maps. If I remember correctly, the relationship between parameter maps and z-maps was near monotonic, with little impact on the analysis. About scaling after the SVD, what normalisation you choose doesn't matter for most analyses (unit norm, unit variance, scaled by the eigenvalues, etc).

Hope this helps!

All the best,

Anderson

2013/1/26 kambiz rakhshan <[log in to unmask]" rel=nofollow target=_blank>[log in to unmask]>

Dear Melodic experts,

I am trying to calculate indices of subject’s connectivity for subsequent quantitative
genetic analysis as suggested in Glahn et al., (2010) PNAS paper. The authors
used Melodic toolbox to extract this value as follow:

“Specifically, the individual (3D) subject-specific maps of the default mode network were collected into a single 4D data set, and the first principal Eigenvector representing the subject’s connectivity was calculated within a study-specific default-mode mask...."

My question is whether such an index of connectivity was measure as the eigenvalue of the covariance matrix, something like:

[subject_loadings, ~] = eigs(cov(ics2), 1); % IC2 is a subject-specific SM for a specific component

I would appreciate if someone clarify on this issue or at least briefly describe how to extract such indices from Subject-specific SM? Does it matter how the components have been scalled(e.g. Z-score vs. percent signal changes)?

Thanks a lot!

Kami