FIX Constant column in motion regressor causes singular matrix

[nathanntg]

The constant column in the motion regressor results in the FSL GLM reporting a SingularMatrix exception. I can not find reference to a constant (intercept) term in the fsl_glm documentation. It could potentially be a result of the --demean argument when calling the fsl_glm command. Do you have any insights into what may be causing this?

For now, I am opening the pull request to start discussion. If this seems like a reasonable fix, I will go through the other steps listed in the "contributing" document.

[satra]

@nathanntg - thanks - this is indeed a bug when used with fsl_glm (especially with the demean flag set to True)

[satra]

and while you are at it, would you mind replacing the BSpline interpolation in applytransforms with Linear

[nathanntg]

No problem. I'll also update rsfmri_vol_surface_preprocessing_nipy.py with the corresponding change as well. I will let you know once those changes are complete. Thanks!

[nathanntg]

@satra - I made similar changes to the other resting state example script, and changed the transform interpolation to linear for both scripts. I tested the revised script up on my data set and it is working correctly. Thanks!

[satra]

thanks @nathanntg

[satra]

@nathanntg - i'm trying to regenerate the error on some old data and couldn't get it to do so on fsl 5.0.6/7/8 (these are centos distributions). i still think this is a good fix, but it would be good to know what system you are getting the message on.

[nathanntg]

Hi @satra - I am using CentOS 6.6 with fsl_glm build 507. I reran the script this morning on one of my data sets and the error message returned is:

Exception: Subnodes of node: filtermotion failed:
Subnode 0 failed
Error:
Command:
fsl_glm -i spacetime_realign/corr_rest_01.nii.gz -d makemotionbasedfilter/filter_regressor00.txt -o corr_rest_01_glm.nii.gz --demean --out_f=F_mcart.nii.gz --out_pf=pF_mcart.nii.gz --out_res=corr_rest_01_filtermotart.nii.gz
Standard output:

Standard error:
terminate called after throwing an instance of 'NEWMAT::SingularException'

Return code: -6
Interface GLM failed to run.

Let me know if you want me to try any other configurations or if I should try updating FSL.

[satra]

@nathanntg : would it be possible to put those two files (corr_rest_01.nii.gz and filter_regressor00.txt) somewhere (e.g., dropbox, bittorrent sync, etc.,.)

[nathanntg]

@satra Sorry for the delay, I don't think that I can share the corr_rest_01.nii.gz file (I need to check with my lab director). Is there a good open source data set that I can use to test and see if it produces the same issue?

[satra]

plenty of open data sets - search for fcp-indi, nki, abide, etc.,.

[nathanntg]

Hi @satra . Unfortunately, I tried a couple of the files from those open data sets and it did not produce the same output. I am not sure what specifically is accounting for the difference. When I looked at the design matrix, nothing jumped out at me. But the "singular matrix" exception was consistent across subjects in our work. Any other ideas I should try to potentially uncover what is going on?