This is to be expected, as some of the original components were probably masking the more subtle noise found on the second time through, or at least making it more difficult to separate. It is not common to run more than one iteration just because of the amount of time and effort involved, but there's no technical reason that I know of not to.

On 13 Jan 2014, at 21:36, Paul Dhami <[log in to unmask]> wrote:

Greetings FSL users/experts,

Sorry for perhaps a "silly" question I had a question in regards to using MELODIC ICA for denoising of my functional data. I ran it once, deemed what components were acceptable and which were not visually, and denoised my filtered.func.data using fsl_regfilt. However, I decided to run melodic once again on my denoised data just for my own interest. The number of estimated components decreased, but I still found a couple components to be "noise" label worthy.

I was just wondering if there was any use to run Melodic more than once for ICA denoising? From what I've gathered, I don't recall ever once seeing a mention of MELODIC being run more than once for ICA denoising. I'm assuming there's a reason NOT TO since like I mentioned, I never seen it mentioned, but was curious if there would be any benefits.

Thank you.

Paul