Releases: brain-modelling-group/fmripop
Releases · brain-modelling-group/fmripop
fmripop - postprocessing of fmriprep data compatible with nilearn 0.6.2
The main changes included in this release are:
- All boolean input argument have a consistent behavior: by default they are set to
False
. To enable the functionality controlled by a given flag, just pass it as an argument when you inovoke the script.
--detrend
- The documentation of each input argument has been improved.
- Updated to work with nilearn 0.6.2 -- this version of nilearn fixed the bug that prevented having heterogeneous smoothing (ie, having a different FWHM value along each axis). With this change, it means this script does not longer work with versions of nilearn < 0.6.2
- Visual debugging functionality has been added. The functionality plots time-series of the default mode network taken from the input data, and the output data.
- The argument previously called
--add_mean_img_back
is now called--add_orig_mean_img
. This new name better reflects what the function actually does. The previous name was confusing. Many processing functions such as filtering, detrending and confound removal have the effect of removing the nonzero mean value of each voxel, where the mean is calculated across time frames. - Updated readme with the typical use cases for resting-state and task frmi data. These cases are also at the top of the script.
- Includes a .yml file to create a working environment with conda.
Save output parameters in json file
The main changes included in this release are:
-
- The file with the parameters passed to
post_fmriprep.py
is now a .json file rather than .txt.
- The file with the parameters passed to
-
- The .json file is saved in the same directory as the fmriprepr .tsv file is stored.
-
- The argument previously called
--scrubbing
is now called--calculate-scrubbing-mask
. This new name better reflects what the function actually does. If we pass this argument, the internally we determine which volumes should be scrubbed, in addition to some scrubbing stats, BUT we do not remove the bad volumes from the data. This change is reflected in the documentation at the top of the code.
- The argument previously called
-
- Correct handling of the argument
--confound_list
.
- Correct handling of the argument
If the user wants to remove confounds which are not the ones specified as default (["csf", "white_matter", "trans_x", "trans_y", "trans_z", "rot_x", "rot_y", "rot_z"]
), then the correct way to pass the new list of confounds is:
--confound_list "confound1,confound2,confound3" --nconf 3
note that we need to specify the number of confounds via different argument (--nconf
). The reason to do this is to make sure that the user really really wants to pass a different number of confounds.
This case is now documented at the top of post_fmriprep.py
as CASE 5.
Public release
Sharing our function with the world