 |
 |
Hi I'm not quite sure if I understand you correctly. You derived two versions of group template maps? Both from the control population only and one with and one without denoising? You then test these two versions of templates for group differences in the control (not denoised) vs patients contrast? First, this approach is invalid no matter what version you're using for your templates. It's because then you derive the templates from group A and then these templates are biased towards A so using them in the A vs B comparison will not give you an unbiased estimate of group differences. In the case where you test the un-denoised ctrl against the patients then the templates from the undenoised ctrl is likely to give you highest group difference as it't the most biased comparison... which I think is what you are seeing. One possibility is to derive the templates from a secondary ctrl population instead. hth Christian On 21 Nov 2012, at 16:57, Alberto Inuggi <[log in to unmask]> wrote: > hi all > > I cleaned (with fsl_regfilt) my controls population and my rsn number raised from 5 to 10. > then I wanted to run the dual regression over both my patients and controls using the two controls templates (noised & denoised) > > I observed that I get much more significant differences (patients vs controls) over the DMN when using the noised template. > > do I have to assume that it's a nonsense to sort noised data over a denoised template? as I try to "split" poor data over rich one. > > but actually in the dr script the first regression ( fsl_glm) is just spatial and my two DMN (noised & denoised) have an almost identical spatial pattern, > only the time/frequency plots were much better in the denoised version (and that's why i expected better results). > > can you give me some advices. try cleaning the patients group, or just use noised data (templates and subjects data) > > we are now not creating the template using both controls and patients, as patients have a strong frontal atrophy that would probably underestimate the frontal networks in the groupS template. is it wise to calculate the template from controls only in such cases? > > thaks in advance > Alberto