dki-questions > Matrix is close to singular or badly scaled.
Showing 1-2 of 2 posts
Display:
Results per page:
Jun 12, 2017  03:06 AM | Zhang Yue - SCNU
Matrix is close to singular or badly scaled.
Dear experts:
    Sorry to interrupt you, I have met a problem in DKE_FT software.
    I have a DKI sample with 2 non-b0 values (1000, 2000) in 30 directions separately. Before running DKE_FT, I averaged the 6 b0 images and did eddy_correct in FSL. Then I got the two different rotated gradients vectors. In the DKE_parameters I do "fn_gradients = 'path/to/matrix1.txt; fn_gradients = 'path/to/matrix2.txt;". Finally, I moved the fa.nii DT/KT.mat to a new file for tracography.
    The problem did not appear when I did example data from your websites, but when I do my 4d nifti data, it said "matrix is close to singular or badly scaled" in optimizing the ODF1. However, to my surprise, other ODFs were optimized successfully.
    Thanks for your help in advance. If you need any more details, please apply me.

  



                                                                                                                                                                                                                       Yours                                                                                                                                                                                                                                Zyue
Jun 12, 2017  12:06 PM | Russell Glenn
RE: Matrix is close to singular or badly scaled.
Hi Zyue,

Thanks for your question. It sounds like you are doing everything correct to me. Does the software keep going / complete? I would visually inspect the ODFs in DSI Studio with the .fib file. 

I believe the error you are experiencing occurs because in the kurtosis dODF formula you have to invert the diffusion tensor (D^-1). If it is not well conditioned, matlab will throw up an error. This usually occurs in background or 'boundary' voxels where the diffusion tensor doesn't really mean anything and is just fitting noise.  The program should keep running through this. You can typically get around this error by applying a more restrictive brain mask or just ignore it.

Hope that helps. 

Sincerely,

Russell Glenn