That would also be my recommendation in this case.
If you have enough subjects with MRIs and time and curiosity, you could also compare your results for constrained vs unconstrained on those with individual MRIs only. But there's nothing "wrong" with using unconstrained even when the accurate anatomy is available.
Thank you both for your information. I really appreciate your time. We have about half of each: MRI/no MRI, so I would be interested to see the differences if I have the time.
I’m trying to decide how to proceed. I’ve already analyzed ~30 subjects with constrained sources & MRI, but now as I’ve reached subjects without MRIs, I’ve learned this is problematic. Should I (1) re-run coherence for the MRI participants using unconstrained sources, losing a lot of time? (2) continue with constrained sources for the non-MRI participants? (noting it as a limitation), or (3) switch to loose/unconstrained sources for the non-MRI participants and flatten the orientation of unconstrained, accepting it as a mismatch (as i believe @Marc.Lalancette may discuss here?: https://neuroimage.usc.edu/brainstorm/Tutorials/PCA#Unconstrained_source_flattening_with_PCA)?
Could you guys please advise me on this? I'd appreciate it. Thank you very much.
I'd say doing it differently is only acceptable as long as you have practically equal numbers in both groups you're comparing, e.g. 15-15 with MRI, and 10-10 without. If you have more with MRI in one group than the other, it would already be an issue, which would be compounded if you analyze them differently. If option 1 is not acceptable to you - though it should be computer time mostly if you use a pipeline or script to re-run it - then I don't think I could really recommend 2 or 3 over the other without knowing much more in detail what you're trying to do. In the end it's a question of how best to extract the signals you're interested in, which is often something you're discovering as you work with your data... At that point it's as much a scientific question for you to investigate as a technical one where we can help. So my recommendation: try doing both on a few subject to see how your data changes or not.
Thank you all for your replies. I began running some of the patients without MRI with unconstrained sources but when I've done this, MATLAB doesn't seem to be able to handle the computation on my computer and either crashes or is very close to doing so (and becomes unresponsive). My computer has 32GB RAM. I am analyzing only 8 scouts: M1, S1, SMA, PMC bilaterally, creating a matrix of 2171x2171 with constrained. As expected, the matrix increases to 6513x6513 with unconstrained sources.
Could revising my PCA settings help the computer manage the computation, or something else? Below are my PCA settings - could you please provide your opinion on them?
'pcaedit', struct(...
'Method', 'pcaa', ...
'Baseline', [-5,5], ... % length of epoch
'DataTimeWindow', [-5, 5], ...
'RemoveDcOffset', 'file'), ...
'removeevoked', 0, ...
Thank you in advance for the help. I appreciate it!
Adam
Hello,
Can you let us know what process this is part of, connectivity maybe? Maybe show us the full process call so it's clearer. There may be some ways to run it in separate steps (e.g. scout PCA first) that could help with the memory issue.
Marc
@bakerdon, please post a new questions in a new topic. (The question about computing connectivity).
The new question will be easier to find and useful to other users