Error: The ROI time-course matrix is not full rank

[pauldhami]

Dear OBHA Team,

I have an EEG resting state dataset of approximately 200 participants.

I am using the HMM-MAR toolbox, and I am trying to perform orthogonalization on my EEG signal using the method of Colclough et al. (2015). However, I am running into this error regarding my data being rank deficient.

`Error using ROInets.closest_orthogonal_matrix (line 70)
The ROI time-course matrix is not full rank.
    This prevents you from using an all-to-all orthogonalisation method.
    Your data have rank 45, and you are looking at 68 ROIs.
    You could try reducing the number of ROIs, or using an alternative orthogonalisation method.


Error in leakcorr (line 31)
        data = ROInets.closest_orthogonal_matrix(data);

Error in loadfile (line 43)
    X = leakcorr(X,T,options.leakagecorr);

Error in checkoptions (line 253)
    X = loadfile(data{1},T{1},options);

Error in hmmmar (line 86)
[options,data] = checkoptions(options,data,T,0);`

I am using Brainstorm to perform the actual LCMV beamforming, then extract the ROIs using the D-K atlas. Running rank on each participant's dataset leads me to find that none of the files have full rank. Is this common? Or is something wrong on my source reconstruction side of things?

I was hoping for some suggestions as to figure exactly where the problem is with my data.

Thank you in advance,
Paul

[neurofractal]

Hi Paul,

As the error states, the rank of your data is lower than the number of ROIs.

Also see Colclough et al., (2015)

"This orthogonalisation process to find corrected time-coursesP by definition removes all correlations between ROIs at zero phase lag. Any remaining correlations between the band-limited power envelopes of these orthogonalised time-courses are thought to represent true biological dependencies (at the expense of true zero-phase-lag connectivity) Brookes etal. (2012a), Hipp etal. (2012), Luckhoo etal. (2012). Our approach is limited by the rank of the data: we cannot correct more ROIs than we have dimensions in Z as there is no longer a unique solution to (5)."

You could try a lower-dimenstional atlas, like the one used in Colclough, G. L., Woolrich, M. W., Tewarie, P. K., Brookes, M. J., Quinn, A. J., & Smith, S. M. (2016). How reliable are MEG resting-state connectivity metrics?. Neuroimage, 138, 284-293, or simply ignore this step (but potentially have issues with correlated sources).

Hope this helps.

[pauldhami]

Thank you very much for your insight! It does indeed help.

If I may ask, each of my participants have different rank values (with none being full). Is this typical of source reconstruction, or is there something wrong on my end?

Thank you again,
Paul

[neurofractal]

No worries. My guess is you rejected different numbers of independent components between participants?

…

On Wed, 25 Aug 2021, 19:10 pauldhami, ***@***.***> wrote: Thank you very much for your insight! It does indeed help. If I may ask, each of my participants have different rank values (with none being full). Is this typical of source reconstruction, or is there something wrong on my end? Thank you again, Paul — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#6 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AENNJZPATSRS6TQ5JY5JBKDT6UW2ZANCNFSM5CZIT4RQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email> .

[pauldhami]

Yes, that is correct! Assuming I would like to use this orthogonalization procedure on my EEG data, does that mean I would have to use an atlas that has a number of parcels that is equal to the lowest rank value of my subject sample data?