Dear FSL experts,
We are trying to use fsl_regfilt to clean up some resting state data. We understand that the ideal way is to run a single session ICA per subject and use fsl_regfilt per subject, and then feed the cleaned data into a group melodic.
However, we were wondering if you could also clean-up the data by the following method:
1) run a group melodic ICA (temporal concatenation) and determine 'bad' components visually
2) the resulting melodic_mix file appears to have the subject & component timeseries arranged as: columns = components, rows = each TR of every subject sequentially. Could we split this file per subject (so that each resulting file has columns = components, rows = each TR for a single subject)
3) use this file in individual subject clean-ups with fsl_regfilt.
The reason we think this is not ok is that this assumes that every subject has noise in the same spatial areas.
However, we also wonder if it IS ok because a voxel located in the CSF (for example) will always produce noise rather than neurological signal? (so, restricting the choice of 'bad components' to sinus and CSF components).
Sincerely,
Jarek Rokicki, Ph.D.
Integrative Brain Imaging Center (IBIC)
National Center of Neurology and Psychiatry (NCNP)
4-1-1 Ogawahigashi-chyo, Kodaira,
Tokyo 187-8551 Japan
|