Dear all,
I have been working on a group-level ICA analysis using MELODIC, and have
been trying to figure out the appropriate procedures to do so (if any). I
have read through the posts on this subject in the archive, but still have
some lingering thoughts.
1. I have found that when I perform MELODIC on smoothed normalized
partial-brain (10 slice) data (1000 volumes). I get far too many components
to be of practical use (500-600 components). If, however, I do not smooth or
normalize before I use ICA, I receive far fewer components (100-200). I
have inspected the smoothed normalized volumes and they look as they should
given the specified parameters. What might be accounting for this
difference in the decomposition of the dataset at different stages of
processing? Does this mean that if I'm interested in a group-level
analysis, I should wait to normalize the components for between subject
comparison only after ICA?
2. For my study, I select components of interest at the individual level
based on an external metric. I then take these selected components from
each subject (approx. 10-20 from each), concatenate them into a new 4D
dataset with now an arbitrary time dimension, and perform ICA on this
dataset. The idea is that this will generate components which represent
spatial consistency in the selected individual components across subjects.
Is this an appropriate use of MELODIC? And, if so, how can I interpret the
statistics that it generates?
Thanks so much for your help and advice!
Tim
|