Hi Daniel and all,
i found this thread very helpful but could one of the experts please
confirm that Daniel's pipeline in the last post is correct?
Thank you so much.
Best,
Anders
2015-09-03 0:09 GMT+02:00 Yang, Daniel <[log in to unmask]>:
> Hi Christian and all,
>
> Thanks so much! I am putting together along with what you suggested. Can you
> please comment again the following?
>
> (1) nuisance regression (WM, CSF)
>
> 1-a). With the WM, we can do:
>
> fslmaths WM_mask_in_highres_space -ero WM_mask_in_highres_space_eroded
>
> flirt -in WM_mask_in_highres_space_eroded \
> -applyxfm \
> -init rest.feat/reg/highres2example_func.mat\
> -ref example_func.nii.gz \
> -out WM_in_func_space
>
> fslmeants -i denoised_func_data.nii.gz -m WM_mask_in_func_space --no_bin -o
> WM_in_func_bin_timeseries
>
> Note: (i) The WM_mask_highres_space can be obtained in several ways (e.g.,
> FSL/fast, SPM12/segment). I use SPM12/segment here and I obtain a weighted
> WM_mask_in_highres_space. This weighted information is actually good stuff
> because it allows more certainty for WM voxels (vs. non-WM voxels). To use
> the information, I use --no_bin in fslmeants. (ii) Also I want to erode the
> WM mask to begin with, in order to increase the certainty (e.g., reducing
> some boundaries with GM). (iii) Finally, I use denoised_func_data (rather
> than filtered_func_data) as input for fslmeants because I want to add
> further denoising to it.
>
> 1-b). We can use similar procedure as above to obtain
> CSF_in_func_bin_timeseries
>
> 1-c). To regress out these two nuisance regressors:
>
> paste WM_in_func_bin_timeseries CSF_in_func_bin_timeseries >
> nuisance_timeseries
>
> fslmaths denoised_func_data –Tmean tempMean
> fsl_glm –i denoised_func_data –d nuisance_timeseries –demean
> –res_out=residual
>
> Note: I use —demean to mean-center the nuisance timeseries so that the
> residual makes sense. I don’t use —des_norm or —dat_norm because I am
> looking at the residual (but not the GLM betas) here. I also want to add the
> mean back so that it is a firm data set (see step (3) below).
>
> (2) Linear detrending
>
> The standard procedure of linear detrending in FSL is done via high pass
> filtering (below).
>
> (3) highpass filtering
>
> Since my TR = 2.0 s and I want highpass filtering of 100 second, this means
> 50 TR (volume), and thus highpass sigma is 25.0 TR (volumes).
>
> fslmaths residual -bptf 25.0 -1 –add tempMean denoised_func_data_2
>
> Note: Finally, I will use the denoised_func_data_2 as input for doing
> participant-level statistics (postprocessing).
>
> Thanks again!!
> Daniel
>
> On 9/2/15, 2:27 PM, "FSL - FMRIB's Software Library on behalf of Christian
> F. Beckmann" <[log in to unmask] on behalf of [log in to unmask]>
> wrote:
>
> Hi
>
>
> (1) nuisance regression (WM, CSF)
> 1-a). With the WM, we can do:
> flirt -in WM_mask_in_highres_space \
> -applyxfm \
> -init rest.feat/reg/highres2example_func.mat\
> -ref example_func.nii.gz \
> -out WM_in_func_space
> fslmaths WM_in_func_space –bin WM_mask_in_func_space
> fslmeants -i filtered_func_data.nii.gz -m WM_mask_in_func_space -o
> WM_in_func_bin_timeseries
>
>
>
> 2 comments: first, there is no necessity to binarise the WM mask per se. If
> you start with the pve map for WM thenthe relative proportions reflect the
> confidence at each voxel of it being WM and you might want to incorporate
> that information by using a weighted mask so that voxels where you’re very
> confident that it’s a WM voxel weight more strongly than a voxel where
> segmentation is less clear.
>
> Secondly you may wont to erode the mask, in order to remove the influence of
> voxels at the GM/WM boundary. You can do this with fslmaths in -ero out
>
>
> Question: We should use filtered_func_data (but not
> ICA-AROMA-denoised_func_data) as input, correct?
>
>
> No, you would want to use the AROMA-denoised data and then add further
> denoising to it, not to the filtered_func_data
>
> 1-b). We can use similar procedure as above to obtain
> CSF_in_func_bin_timeseries
>
>
> Yes, but the same comments apply
>
> 1-c). To regress out these two nuisance regressors:
> paste WM_in_func_bin_timeseries CSF_in_func_bin_timeseries >
> nuisance_timeseries
> fsl_glm –i denoised_func_data –d nuisance_timeseries –dat_norm
> –res_out=residual
> Questions: The residual is the file with these nuisance regressed out,
> correct? We should use –dat_norm, correct? (Or –des_norm? What’s their
> difference?)
>
>
> You also want to switch on demeaning. The des/dat norm is irrelevant as
> you’re not using the betas but the residuals for further processing.
> Depending on what you want to do with the residuals later you also need to
> re-add the mean so it looks like a firm data set
>
>
> (2) Linear detrending
> The standard procedure of linear detrending in FSL is done via highpass
> filtering. How to perform linear detrending as suggested in the ICA-AROMA
> paper? Many thanks!!
>
>
> use fslmaths with the -bptf option, it will also remove the linear trend
>
> hth
> Christian
>
> (3) highpass filtering
> Since my TR = 2.0 s and I want highpass filtering of 100 second, this means
> 50 TR (volume), and thus highpass sigma is 25.0 TR (volumes).
> fslmaths input -bptf 25.0 -1 output
> Is that correct?
> Best,
> Daniel
> From: "[log in to unmask]" <[log in to unmask]> on behalf of Daniel Yang
> <[log in to unmask]>
> Reply-To: "[log in to unmask]" <[log in to unmask]>
> Date: Wednesday, September 2, 2015 at 1:13 AM
> To: "[log in to unmask]" <[log in to unmask]>
> Subject: [FSL] ICA-AROMA questions: the preprocessing steps after ICA-AROMA
> Dear all,
> In ICA-AROMA steps, after ICA-AROMA, it is suggested that we do:
> Nuisance regression: WM, CSF, & linear trend
> Highpass filtering
> Could we please provide details how these two can be properly done (e.g.,
> via bash scripts)?
> Many thanks!
> Daniel
>
>
|