 |
 |
Hi Paulo, Can you zip up a feat directory from a single subject that fails, and send to us? You can upload securely at: https://oxfile.ox.ac.uk/oxfile/work/extBox?id=690991047F8A89D9F1 Cheers, Taylor Hanayik Taylor Hanayik PhD Analysis Research Software Engineer FMRIB, John Radcliffe Hospital University of Oxford [log in to unmask] > On 28 Mar 2019, at 22:08, Paulo Bazán <[log in to unmask]> wrote: > > Hi everyone, > > I am having the same error message (error: pinv(): svd failed) in some subjects. > > In my case, I did the preprocessing (motion correction; slice timing correction; registration; smoothing; linear detrend, temporal filtering and removal of confound regressors, including scrubbing step) in CONN and then tried to use the preprocessed image in Feat. > > The error only occur if I use pre-whitening option, which I believe is important. > > To be able to use this image in Feat, I need to mask it and add a constant value in order for Feat to be able to separate the background from the image (important for Feat's mask creation step). So to prep the image for Feat, I created a mask considering that the background of the image was zero, and the brain regions would have non-zero values. I calculated the mask as the square of the image (fslmaths preprocessed_image -sqr -bin mask). This provided a 4D mask. Then I noticed that, do to scrubbing, some voxels in the brain the "scrubbed" volumes would be 0 in the mask (probably because they were very small values, and -sqr -bin round them to 0). Therefore, I decided to correct the mask and used the maximum value of the volumes after bin (fslmaths preprocessed_image -sqr -bin -Tmax mask). This provided a 3D mask, which look ok. > > Then I add 100 (also tested 10000 with similar results) and multiplied the preprocessed image by the mask (fslmaths preprocessed_image -add 100 -mul mask). > > Curiously, when using the 4D mask, the analysis proceeded normally (without error message). However, the cope images do not look normal, having several outlier voxels (probably related to the voxels that were zeroed by the 4D in some volumes when using the 4D mask). Also, if I do not mask the preprocessed image and do not add the constant, the analysis also finishes without error, but again the cope image does not look ok, having a lot of 0 values. > > The tests were performed with FSL 6.0 both in CentOS 7 and in macOS Sierra and also with FSL 5.0.6 in OS X El Captain (same errors). > > Hope this information can help understand the issue. Please let me know if I should send you some data related to these tests for you to take a look. > > Best regards, > Paulo Bazán > > ######################################################################## > > To unsubscribe from the FSL list, click the following link: > https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1 ######################################################################## To unsubscribe from the FSL list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1