Hello Experts,
I am looking for a statistically acceptable way of joining 2 large data resting state fMRI sets (100 subjects at 10 minutes and 30 subjects at 8 minutes). I would prefer to not truncate 20% of the 100 subjects data to match the smaller 8 minute run. Would it be possible to add some kind of a null image to fill in the last 2 minutes of the 30 subjects with 8 minute runs - so that they are all the same length and accepted by Melodic for processing?
I've tried adding a run of NaN images - Melodic's ICA step was not able to converge. I did try duplicating the very last image of each subject to fill in the last 2 minutes. The thought behind this method is that without any variability in the data, there will be no signal change detected, so no "correlations" between areas would be detected from these last two minutes of data in these subjects - essentially null images. Not sure how this will be handled by randomize later, though.
The last method is to try to add white noise. Not sure how Melodic would handle this yet.
Ideally the last image duplication technique could be made to work acceptably because I already know Melodic will accept the image.
Thank you for your help,
Varina
|