Noise reduction
The following guide describes the noise reduction procedure in task-related data. In general, there are many ways to achieve the same goal: to increase the signal to noise ratio. Thus, the following list presents solely those which have already been successfully performed in our lab:
- 1) Regressing out the noise signal using MELODIC and fsl_regfilt
Regressing out the independent components attributed to artificial signal based on visual inspection of the ICA report page. See the FSL Wiki. - 2) Adding a confounding matrix using fsl_motion_outliers for severe motion
Creating a confounding matrix for timepoints that exceed the severe motion levels. See the FSL Wiki. - 3) Adding nuisance signal regressors
In general, the nuisance signal regressors can be added either manually, or automatically using the following tool:
loop <list/range> -f=<folder pattern>:<options>
list/range: list or range of numbers representing the studies to be analysed (delimited by commas without spaces)
folder pattern: name of the *.feat directory or its matching pattern (as in ls command)
options: are delimited by commas and include following:
mp - add motion parameters to the model (add as a first option when using multiple regressors)
wm - use wm_timecourse_demeaned as regressor
csf - use csf_timecourse_demeaned as regressor
swm - use swm_timecourse_demeaned as regressor
scsf - use scsf_timecourse_demeaned as regressor
gs - use gs_timecourse_demeaned as regressor
r=<search>@<replace> - search for the specified string (or regular expression) in the original desing.fsf and replace it with the new one, useful when the original analysis uses obsolete path entries
mpr or mps - update the path to T1 image, useful if the original analysis was a multiple analysis
f - force the analysis and ignore the safey checks (e.g. for obsolete path entries or when the path to T1 does not match the path to functional)
A) Counfounding motion parameters:
a) Motion parameters can be added manually in the Feat GUI. See the FSL Wiki.
b) Motion parameters can be enabled in the fmriStats.sh script using the --addmotionparam option. See the Tools page.
c) Motion parameters can be added after the initial analysis is already done. First, make sure that the original analysis had the Motion correction (MCFLIRT) done. Then, from the study root directory (e.g. /md1/NM-RMV) run the loop command (see above).
In this case the command would look similar to following:
loop 1-10 -f=RFingers*[0-9].feat:mp
or, more complex:
loop 2,5,7-9 -f=*Foot*[0-9S].feat:r=md5@md1,r=4.1@5.0,mp
B.1) Adding a nuissance signal of the white matter (using the standard space template)
The following guide describes the procedure to extract the white matter signal using a standard template mask:
a) Run your standard FEAT analysis first. Make sure you register your data to the standard space and to the high resolution T1. See the Basic manual for more details.
b) In your study root directory (e.g. /md1/NM-RMV) run the following command:
loop <list/range> -wm=<folder pattern>:<full path to mask>
list/range: list or range of numbers representing the studies to be analysed (delimited by commas without spaces)
folder pattern: name of the *.feat directory or its matching pattern (as in ls command)
full path to mask: full path to the white matter mask, the mask should be a probabilistic mask in standard space; the standard mask is stored in /md1/masks/wm.nii.gz
This will create the wm_timecourse_demeaned with the mean signal extracted from the white matter mask. The file is saved in the original *.feat folder and in the newly created directory Results/seeds.
c) Now, use the loop command (see above) and add the wm option. It can be directly combined with other regressors as well (note, that the mp option has to precede other regressors since it re-runs the Pre-Stats as well). The command would look similar to the following:
loop 1-10 -f=RFingers*[0-9].feat:wm
or, more complex:
loop 2,5,7-9 -f=*Foot*[0-9S].feat:r=md5@md1,r=4.1@5.0,wm
B.2) Adding a nuissance signal of the white matter (using the segmentation)
The following steps describe, how to create the white matter signal regressor using the segmenated T1-weighted image:
a) Run your standard FEAT analysis first. Make sure you register your data to the standard space and to the high resolution T1. See the Basic manual for more details.
b) In your study root directory (e.g. /md1/NM-RMV) run the following command:
loop <list/range> -s=<folder pattern>
list/range: list or range of numbers representing the studies to be analysed (delimited by commas without spaces)
folder pattern: name of the *.feat directory or its matching pattern (as in ls command)
This will segmentate the highres.nii.gz within the speciefied folder using the FSL's Fast tool. The output is also transformed into the native BOLD data space and saved in the seg directory within the specified input folder. There is another seg folder created in the parent Results directory, where the original segmentated T1 is saved, thus saving time in future analyses.
c) In your study root directory (e.g. /md1/NM-RMV) run the following command:
loop <list/range> -swm=<folder pattern>
list/range: list or range of numbers representing the studies to be analysed (delimited by commas without spaces)
folder pattern: name of the *.feat directory or its matching pattern (as in ls command)
This will create the swm_timecourse_demeaned with the mean signal extracted from the white matter mask. The file is saved in the original *.feat folder and in the newly created directory Results/seeds. This command requires the previous step to be performed first!
d) Now, use the loop command and add the swm option. It can be directly combined with other regressors as well (note, that the mp option has to precede other regressors since it re-runs the Pre-Stats as well). The command would look similar to the following:
loop 1-10 -f=RFingers*[0-9].feat:swm
or, more complex:
loop 2,5,7-9 -f=*Foot*[0-9S].feat:r=md5@md1,r=4.1@5.0,swm
C) Adding a nuissance signal of the CSF
You can basically follow the guide from point B.1 or B.2, replacing all occurences of wm or swm with csf or scsf, respectively. The former will use the standard mask stored in /md1/masks/csf.nii.gz or your own probabilistic csf mask in the standard space. The latter uses the segmentated T1-weighted image as a mask source.
D) Adding a global nuisance signal
You can basically follow the guide from point B.1., replacing wm with gs. The mask used is the same mask, as created by Feat script to mask all the non-brain voxels.
Back to Process.
upol.cz
