ICA changes over the last few months?

1

Hi Francois,

Whilst I have successfully ran EEG ICA's in the past in brainstorm (including infomax - this is important for later), over the last month to two, I have noticed there have been changes with how the analysis runs. I am not totally sure where things are affected specifically (it might even user error given I have relatively only really started to use brainstorm itself), inclusive of considering reading forum threads.

Originally, the ICA didnt take very long to run, but ended up taking 8+ hours for one analysis. I was under the impression this was related to sharing channel files (ICA running), but it seemed I was using the ideal protocol. After installing the latest update (6th Jan, 2019), I noticed that the ICA ran as quickly as it used to, back when i first started using brainstorm. This makes me think that maybe the brainstorm updates (recent and prior ones over the last few months) may have changed the way ICA algorithm is called in the code? As another point, right now, I can only remove ICA factors using JADE rather than infomax. Because there is some literature evidence that indicates infomax is better than JADE, I'd like to identify how i can go about using infomax. At the moment, in MATLAB I get the following warning when it comes to using infomax:

Warning: Matrix is close to singular or badly scaled. Results may be
inaccurate. RCOND = 3.219110e-18.

In runica (line 1487)
In process_ssp2>Run (line 714)
In process_ssp2 (line 29)
In process_ica>Run (line 139)
In process_ica (line 28)
In bst_process>Run (line 229)
In bst_process (line 36)
In panel_process_select>ShowPanel (line 2895)
In panel_process_select (line 30)
In panel_record>CallProcessOnRaw (line 2442)
In panel_record>@(h,ev)CallProcessOnRaw('process_ica') (line 194)

With your knowledge of the updates and how the code is integrated (I haven't really delved into the code myself largely because I only have a moderate - at best - programming background),

Any thoughts? Is it just me?

Thanks,

Lee.

3

Hi Lee,

The last updates are the following:

I don’t think any of these modifications would produce the differences you mention. There were probably changes in the way you call the function (different options) or in the the data itself.
To check if the difference comes from changes in the software, the best would be to run the ICA on the same data with two different versions of Brainstorm.
I can pull out any version of Brainstorm if you feel like testing this further: just le me know the date of the version of Brainstorm you want to test and I’ll send it to you.

4

Hi Francois,

You are right… it works on a different computer without problem. Not sure where the issues are coming from my work computer then, but that’s my problem.

Thanks for your quick reply and help, and sorry to have wasted your time. Overall, as a new user to brainstorm, I am overall quite impressed with it.

All the best,

Lee.