Warning using Friston

Submitted by Ham on Thu, 11/06/2014 - 10:18

Hi dpabiers!

Dear Dr. Yan. 

Anyone know why when I perform Friston 24 I get this Warning; 

Warning: Matrix is close to singular or badly scaled. Results may be inaccurate. RCOND =  1.205703e-10.

There is attached the cov file from one subject. Some values are close to 0. 
 
My data info: 
 
- 236 TP
-2.5 TR
 
and preprocessing stps:
 
Slice timing
Realign.
Reoriented (t1 & func)
Coregistered
(Bet)
Segmentation
Nuisance covariate csf & wm, and Friston with (Compcor and spm apriori)
Normalize
Smooth
and FC.
 
 
Thank you in advance!
 
Best
AttachmentSize
Plain text icon CovariablesSub_001.txt157.9 KB

Forums:

YAN Chao-Gan

Mon, 11/24/2014 - 19:30

Hi,

Usually wannings can be ignored.

The rank of your covariate matrix is 27. Could you paste the nuisance setting here?

Best,

Chao-Gan

 

Ham

Tue, 11/25/2014 - 09:39

Thank you so much for your answer, 

Do you think the preprocessing steps are reasonable or I will have a hard time with a reviwer?

Note I use the component correction method. 

 

Best

YAN Chao-Gan

Mon, 12/01/2014 - 21:34

Hi,

I think this setting should be fine.

Best,

Chao-Gan

Ham

Wed, 12/03/2014 - 12:04

Dear Dr. Yan, 

Thank you so much!


Best, 

Sharka

Thu, 01/22/2015 - 14:54

 

Dear Chao-Gan

Using the same settings as above I get  images that seem to be processed badly (see above (2nd image)). I guess related to this is the message I get during preprocessing: 

Warning: Matrix is close to singular or badly scaled. Results may be inaccurate. RCOND =  8.499165e-10.

Additionally, looking at the CovariablesSub_***.txt file, the covariates  are close to zero. (I run into similar problems using nifti and analyze formated images, after normalization step using epi templates, as well as using the unified segmentation approach (and using SPM apriori / or segmented masks). 
 
Any help is very much appreciated. 
 
Best,
Diana
 
 
 
 

 

 

 

 

YAN Chao-Gan

Thu, 01/22/2015 - 17:27

Hi Diana,

The image is normal after regressing out covariates. For each voxel, the temporal mean will be zero after nuisance regression, thus you will no longer see the difference in mean intensity among gray matter, white matter and CSF. 
Also, you will see a clear boundary between voxels insider brain from those outside brain before regression. However, after nuisance regression, all the voxels were near zero, thus you will no longer see the boundary.
 
Most of the resting-state fMRI analyses (e.g., FC, ALFF, ReHo) were interested in the variation other than the mean, the temporal dynamics were still there within the "distorted" images.
 
CovariablesSub_***.txt file, the covariates  are close to zero is because head motion usually is small. As long as they have variations, it should be OK.
 
Best,
 
Chao-Gan