 |
 |
Hi Anderson, thanks for the advice, I'll rerun and script the analyses to randomize everything. I'm not interested in results of all 20 components though, and wanted to identify only three components by cross-correlation with templates from the Smith et al. 2009 paper, thus I thought a correction of 0.05/4/3 would be adequate. I just fed all 20 components into the dual regression, because it's more convenient with the .filelist file and I thought this wouldn't make a difference, if I just looked at results from the three components of interest later on. Best, Nils Von: FSL - FMRIB's Software Library [mailto:[log in to unmask]] Im Auftrag von Anderson M. Winkler Gesendet: Freitag, 16. Mai 2014 13:25 An: [log in to unmask] Betreff: Re: [FSL] Hypersignificance in dual regression Hi Nils, Design looks fine, contrasts too. Make sure that you are really blind to group assignment when you run the preprocessing. The best way to accomplish this is to script the tasks and, ideally, not even know who are the subject and who are the controls, leaving to arrange the groups at the far end, just before computing the statistics. Also note that with 20 components and 4 contrasts, the significance level needs to be corrected. Use 0.05/4/20 = 0.000625, but to reach this, at least 1600 permutations are necessary. Try running again with, say, 5000, and threshold at 0.999375 then. All the best, Anderson On 16 May 2014 09:59, Kohn, Nils <[log in to unmask]<mailto:[log in to unmask]>> wrote: Hi FSL experts, I'm running a temporal concatenation ICA with subsequent dual regression. The components after ICA (20 components solution) look normal, but after dual regression, components show hypersignificant results (>.996, in tfcecorrp) in all contrasts (differential and group effect). I'm currently running the latest FSL version and have rerun the analysis several times. I have 48 inputs in total, with 2groups and 24 inputs per group. I attached the mat and con files I use and here's the dual regression command: dual_regression groupmelodic.ica/melodic_IC.nii.gz 1 tcICA_rest_final.mat tcICA_rest_final.con 500 DR_TC_48 `cat .filelist` I would be grateful for any advice on this! Thanks in advance! Best, Nils