Hi Anderson,
Yes of course I have a hypothesis, I misunderstood what you wanted to say.
Indeed I have to ask why they did those 2 scans, but the aim of my project is not to compare the 2 runs.
About the merging of component, if I can prove that 2 components are in fact strongly correlated (for example by a cross-correlation of the graphs of the powerspectrum of timecourse of the 2 components of interest given by melodic output), and so that actually The group-ICA should have given only 1 component instead of two, Do you think I can merge them and say that is meaningful ?
Otherwise I'm not sure I understand what you mean by incorrect results if I treat scan as a subject.
In fact, I did 2 final analysis (randomise) after group-ICA and dual-regression on the 52 images.
first with the mean of the 2 scans per subject, my designs were like that :
design.mat :
group EV1 EV2
1 1 0 (1st control)
1 1 0 (2nd control)
1 1 0
...
1 1 0 (13th control)
2 0 1 (1st asd)
2 0 1 (2nd asd)
2 0 1
...
2 0 1 (13th asd)
in a second hand, I tried this design, using the 52 images like that :
group EV1 EV2
1 1 0 (1st control, 1st scan)
1 1 0 (1st control, 2nd scan)
1 1 0 (2nd control, 1st scan)
...
1 1 0 (13th control, 2nd scan)
2 0 1 (1st asd, 1st scan)
2 0 1 (2nd asd, 2nd scan)
2 0 1
...
2 0 1 (13th asd, 2nd scan)
(as I said, I use the 2nd scan of a subject as it was a another subject of the same group)
For the both analysis I used this contrast design :
asd>con -1 1
con>asd 1 -1
asd mean 0 1
con mean 1 0
So for you, my second analysis (with the 52 scans) is not correct ?
Could you explain me why ? I think I didn't totally understand how randomise works
Thanks in advance
Best regards,
Antoine
|