Dear experts,
I have been using ICA-AROMA to clean some resting-state fMRI data that I have acquired with a multiband acquisition (TR = 7540 ms; 616 volumes). As an initial test, I ran the pipeline on the same dataset twice, under identical conditions, to examine variability in the results.
Across 12 regions-of-interest, I find that the correlation in the time series between time 1 and time 2 can be quite low, ranging between .4 and .9.
This problem seems specific to ICA-AROMA, as I have tried other pipelines (e.g., regression of head motion parameters and wm/csf signals) and the correlations between time 1 and time 2 are perfect.
Also, I’m not sure if this is a problem specific to the multiband acquisition, as if I repeat the analysis using non-multiband data, the correlations between time 1 and time 2 under the ICA-AROMA pipeline are all >.9.
Looking at the multiband results, I find that melodic identifies around 350 components, of which a large number (230) are classified as noise.
Can anyone provide any clue as to what is happening here? Are there known issues in applying ICA-AROMA to multiband data that I’m not aware of?
Thanks.
Kristina
|