Thank you all very much for your quick and detailed response. It is very helpful!
I plan to run Melodic with all subjects:
1.The control sample size is not big enough (only 14), which may not be sufficient to provide a robust reference (spatial independent components).
2. I was not able to find much structural differences between groups (age matched) from the literature (I can always run a FSLVBM to confirm).
3. Previously I tried Melodic/Dual regression with a smaller sample size (mixed of controls, early patients and advanced patients, unbalanced group). Melodic identified 30+ components, but only 1 component (default mode network) showed significant difference between groups. This suggested the groups are not that different from each other.
I have one more question regarding our data set. For the advanced patients, we acquired two set of data (medication on vs medication off). I should leave the data (e.g. medication on) out during Melodic, then add the data (medication on) back on during dual regression, right?
Melodic: 14 controls + 15 early patients + 15 advanced patients off medication
Melodic: 14 controls + 15 early patients + 15 advanced patients off medication + 15 advanced patients on medication
Thank you very much for your help. I appreciate your input very much.