Hi everybody,
let me explain my situation:
I'm doing a resting state functional connectivity analysis between 2 groups (autism and control) with 2 sessions of resting-state
I did a classic preprocessing, I ran melodic group-ICA, then I chose around 10 ICs of interest
(using the RSNs of PNAS , Smith et al. 2009). I did a dual-regression, and finally I ran randomise 2 sample unpaired T-test with my 'dr_stage2_ic#(of interest)' using TFCE option, where each time point of my 'dr_stage2_ic' is a mean of this IC of the 2 sessions per subject (as it's described in the userguide of Randomise).
Then, I checked my results which were not convincing for me. In fact I had just 1 or 2 ICs of interest showing something à p<0.05 (corrected_tfce)(contrast between 0.95/1 in fslview), containing 1 or 2 small clusters (beetween 2 and 5 voxels) showing the autism>control (or control>autism).
So i tried to improve the results.
First I remade randomise but this time without the meaning between the 2 sessions (i.e I consider now the 2nd session as new control and autistic people)
With that I had better results. More ICs of interest shew something, but still with small clusters.
In a second hand, I remade all the processing by improving the registration (I judged too bad), rejecting 1 subject because of a bad registration and rejecting another one because of a huge head motion. With that I think my results are more accurate, but I lost some clusters in the contrast images.
So now, those are my questions :
- Is it acceptable to do what I did, considering the 2nd session as new people in my samples ?
- How can I improve my result ? Do you think I must use different parameters of -T option in randomise (or use -T2) ?
And maybe, as I didn't resample my fMRI after registration (in registration part in MELODIC GUI), the -v option would be interesting ?
- Finally, I have for example 2 MELODIC-ICs really correlated to 1 RSN, Is it a good option to consider them as subcomponents, merge them in one component and remake randomise on it ?
Sorry for the length of my text, but I prefered to be precise.
Thank you in advance,
Antoine B.
|