 |
 |
Hi The randomisation is going to give exact stats no matter what so this is nothing to worry about. If you are worried about bias you can also split your controls into two, derive the maps from ctrl1 and test (with DR) ctrl2 and pat against these maps. hth Christian On 8 Oct 2011, at 23:18, Cornelius Werner wrote: > That's actually quite a good idea. Should've thought of that myself :-) > Cheers, > Cornelius > > On Sat, Oct 8, 2011 at 11:01 PM, Benjamin Kay <[log in to unmask]> wrote: >> You could to two separate tcICAs and compare the groups visually as an >> exploratory measure. Then, if you are satisfied that your ICs of interest are >> similar enough, you could go ahead with the combined tcICA and dual >> regression. Otherwise, short of graph theory, I'm not aware of a better >> statistical tool for your situation. I'm curious what others think about how >> robust dual regression is in your situation. >> >> On Saturday, October 08, 2011 16:45:44 you wrote: >>> Hi, thanks for your reply. Yeah, of course, this would work >>> (hopefully). The reasons for perhaps using two tcICAs are that a) I >>> only have half the number of patients compared to controls (20 vs >>> ~40), and there has been some dispute about whether this might throw >>> off balance across groups. Plus, b) the patients are quite severely >>> ill, so their RSNs might differ a lot (at least, that's what we >>> expect), and I am not exactly sure what a tcICA might make of that >>> when both groups are in the same analysis. What do you (and others) >>> think? >>> Cheers, >>> Cornelius >>> >>> On Sat, Oct 8, 2011 at 10:22 PM, Benjamin Kay <[log in to unmask]> wrote: >>>> As you explained, running dual regression on two different sets of ICs >>>> presents a challenge. Why not do tcICA on data from both groups, >>>> together, and then launch your dual regression from there? >>>> >>>> On Saturday, October 08, 2011 16:16:41 you wrote: >>>>> Hi, >>>>> >>>>> it has been suggested on the list to generate a set of RSNs from the >>>>> "healthy" population and to run it against the "patient" group when >>>>> large differences are expected. That's what I am trying to do because >>>>> I have reason to assume so. >>>>> I imagine this means running tcICA on both the control and the patient >>>>> group separately and then (when calling the dual_regression script) >>>>> using melodic_IC.nii.gz from the control.gica on the .filelist from >>>>> the patient.gica - is that right? But how can I include possible >>>>> additional/specific artefacts represented in the patient >>>>> melodic_IC.nii.gz that might not necessarily be present or identical >>>>> in the control group in the analysis? Fslmerge the ICAs of interest >>>>> from the control melodic_IC into the patient melodic_IC and use that >>>>> instead (and possibly throw out ICAs from the patient melodic_IC that >>>>> look similar beforehand)? Did anyone do this succesfully and can >>>>> advise? >>>>> Thanks a lot, >>>>> Cornelius >> > > > > -- > Dr. med. Cornelius J. Werner > Department of Neurology > RWTH Aachen University > Pauwelsstr. 30 > 52074 Aachen > Germany