 |
 |
Hi Val, Please, see below: On 11 May 2016 at 17:24, Valeriya Azorina < [log in to unmask]> wrote: > Hi Anderson, > > Thanks a lot! Just have a few follow up questions about that: > > 1. Should all nuisance regressors be in one file? > Yes. > Does it work with a .txt file or does is need to be in .mat? > I'm not sure. Give it a try with .txt, and if it fails, add the first lines so as to transform it into a .mat (create a test example with Glm_gui to see how one looks like internally). > 2. What do you think about the following approach? In particular, could > you comment on the regressing global mean outliers part if you happen to > know anything about it, please. > > To regress out the WM and CSF signals, I add the time-series in the > filename part and I think determining 1 for each EV (CSF and WM) in > contrast tab doesn't change the result. However I'm not sure if I'm > correct. *According to articles, I think global signal regressing out > make some anti-correlation in analysis, so I preferred to not doing that. *I > use res4d.nii.gz in the stat folder for other analysis such as bandpass > filtering (this nii file is residual of regression step). > Yes, there is a large discussion on this matter. I don't see a problem: regressing out the signal indeed reveals negative residual correlations between regions. Arguably, it's positive correlations due to global fluctuation that would not be interesting. That said, there are good arguments for not including it, and perhaps a suggestion is to use FIX/ICA so as to remove signals that are clearly of no interest. > > 3. After this step I was going to go directly into extracting the > timeseries of my ROIs with fslmeants. Are there any steps I am forgetting > or am I good to go? > FIX is a tool you might want to have a close look. Likewise, ICA-AROMA. All the best, Anderson > > Thanks a lot for your answer, it's a great help! > Val > > On 11 May 2016 at 11:21, Anderson M. Winkler <[log in to unmask]> > wrote: > >> Hi Val, hi Bob, >> >> To regress out a set of nuisance variables use the command fsl_glm. >> Something as this: >> >> *fsl_glm -i filtered_func_data.nii.gz -d design_with_nuisance.mat >> --out_res=filtered_func_data_new.nii.gz* >> >> >> The filtered_func_data_new.nii.gz are the residuals of the model that has >> these variables, hence will be free of these nuisance effects. >> >> If there are hundreds of timepoints and only a few nuisance variables >> regressed in this way, it's fine. However, if there are only a few >> observations and too many variables, this procedure will cause false >> positives later, because the residuals are no longer independent, and the >> true degrees of freedom of the subsequent model will be overestimated. >> >> All the best, >> >> Anderson >> >> >> On 10 May 2016 at 15:31, Robert Kraft <[log in to unmask]> wrote: >> >>> Val, >>> >>> Thanks for posting this question. I am also very interested in knowing >>> how to do this. At the moment, I am using SPM via the CONN toolbox to >>> perform this task. I would much rather use FSL. I just didn’t know how. >>> I am looking forward to the response. >>> >>> Bob >>> >>> >>> > On May 10, 2016, at 10:11 AM, Val Arizona <[log in to unmask]> wrote: >>> > >>> > How can I regress out nuisance regressors (WM, CSF signal, motion and >>> global mean outliers) without having a model or main EVs because my data is >>> a 6min resting state scan? I want to regress out the unwanted signal and >>> then have a new clean filtered_func_data.nii.gz to use for time series >>> extraction of ROIs. As I understood from the literature regressing out >>> nuisance as a preliminary step can be done, but HOW it is done especially >>> with FSL I couldnt figure out. >>> > I have extracted timeseries from CSF and WM and I have ART (Artifact >>> Detection Tool) output of motion and global mean outliers as separate txt >>> files. I have combined them into one txt file with all nuisance regressors >>> (first two columns CSF and WM and ART output after that) which I tried >>> uploading under confound EVs, but I didnt know what to do about the model >>> set up and all the other options. >>> > A step by step guidance through setting up the regression would be >>> much appreciated. >>> >>> >> >