NITRC masked ICA (mICA) Toolbox Forum: open-discussion

RE: convergence error_mICA(masked Independent Component Analysis)

sankalp tikoo — Thu, 09 Dec 2021 9:59:32 GMT

Thanks Jorge, I did try fsljiscmail but I haven't received any reply yet. Thanks for your timely help though. If I happen to troubleshoot my error, I shall write back to you on this thread.

RE: convergence error_mICA(masked Independent Component Analysis)

Jorge Manuel Sánchez — Thu, 09 Dec 2021 9:53:48 GMT

Unfortunately, I have not come across this error before. It seemst to me that there is some kind of error in your data, but it is only a supposition.
You could try the FSL mailing list (https://www.jiscmail.ac.uk/cgi-bin/wa-jisc.exe?A0=FSL), maybe they can provide some help.

RE: convergence error_mICA(masked Independent Component Analysis)

sankalp tikoo — Thu, 09 Dec 2021 9:42:25 GMT

Dear Jorge,

When I run a simple melodic command using the -v option I get the following error:

Removing mean image ... done
Reducing data matrix to a 179 dimensional subspace

Excluding voxels with constant value ... done
Normalising by voxel-wise variance ... done
Data size : 179 x 2145
Starting PCA ... done
Start whitening using 15 dimensions ...
inv(): matrix seems singular

Have you overcome across this error? Moreover single subj ICA runs smoothly (mICA does output ICA components at single subj level). The no convergence error and the inv(): Matrix seems singular error is thrown when I am trying to run group ICA with -a concat option, both via mICA or via melodic command line.

RE: convergence error_mICA(masked Independent Component Analysis)

Jorge Manuel Sánchez — Thu, 09 Dec 2021 9:22:52 GMT

Dear Sankalkp,

I could not find anything in the logs. It seems to be a problem with melodic. You could try to run the melodic command from the top of the logs using the verbose option (-v). You might get some useful information.

RE: convergence error_mICA(masked Independent Component Analysis)

sankalp tikoo — Thu, 09 Dec 2021 8:49:59 GMT

Please find attached the requested files. Yes I did check for non-motion artefacts also. The functional data seems fine to me and so does the registration to the UNC 2 year old template.

RE: convergence error_mICA(masked Independent Component Analysis)

Jorge Manuel Sánchez — Thu, 09 Dec 2021 8:38:48 GMT

Could you please post the following files?[list][*]mica_log.txt[*]mica_config.cfg[*]log.txt[/list]
Did you look for non-motion artefacts?

RE: convergence error_mICA(masked Independent Component Analysis)

sankalp tikoo — Thu, 09 Dec 2021 8:17:53 GMT

Dear Jorge,

Thank you for your prompt response. Any kind of movement artifacts have already been regressed out.
Yes I have tried using other dimensionalities such as 10,15 or 20. When I use dimensionality 15 I get the following error:

Reading data file /Users/tikoos/Desktop/test_2yr/GrpICA/masked_input/input_29_masked_s2.0 ... done
Removing mean image ... done
Reducing data matrix to a 179 dimensional subspace

Excluding voxels with constant value ... done
Normalising by voxel-wise variance ... done
Data size : 179 x 2145
Starting PCA ... done
Start whitening using 15 dimensions ...

After this the masked ICA stops and no resting state components are generated. The index.html file doesn't output anything. Moreover the eigenvalues_percent file has nan values.

Looking forward to your reply.
Thanks.

RE: convergence error_mICA(masked Independent Component Analysis)

Jorge Manuel Sánchez — Thu, 09 Dec 2021 7:48:30 GMT

Dear Sankalp,

Have you tried another dimensionality? Sometimes, the ICA algorithm does not converge.
Another thing that I would suggest is to look for artefacts in your data.

Best,

Jorge

convergence error_mICA(masked Independent Component Analysis)

sankalp tikoo — Wed, 08 Dec 2021 20:30:50 GMT

Dear mICA developers,

I am using mICA to run group ICA within the cerebellum of 2 year old children. My data is already pre-processed for motion correction and coregistered onto a 2 year old UNC template. I have a cerebellar mask in 4mm space to match my preprocessed func data (4mm Vox size). While running mICA I get the following error:

Step no. 489 change : -inf
Step no. 490 change : -inf
Step no. 491 change : -inf
Step no. 492 change : -inf
Step no. 493 change : -inf
Step no. 494 change : -inf
Step no. 495 change : -inf
Step no. 496 change : -inf
Step no. 497 change : -inf
Step no. 498 change : -inf
Step no. 499 change : -inf

No convergence -- giving up

It would be great if someone can help me troubleshoot this error.
Thank you for your prompt response.

RE: cannot install mICA

Jorge Manuel Sánchez — Wed, 12 May 2021 6:04:09 GMT

Dear Jingyi,

What operating system are you running?
Is the mICA file executable? You could check with "ls -l". The output should look like this:

-rwxr-xr-x 1 user group 2156 Jun 5 2020 mICA

Concerning Python, the toolbox is not yet programmed to work with python3. It still requires python2. However, that should not be a problem for the GUI. Python is only required for the reproducibility analysis.

Best regards,

Jorge