fmriprep_denoising

Scripts for comparing different fmriprep denoising approaches on quality control functional connectivity (QC-FC) motion, QC-FC distance-dependence, and temporal degree of freedom loss.

The Python program denoise_fmriprep_output.py will execute any of several pre-set denoising pipelines (description forthcoming) which are mostly based on published, validated methods. For each functional file identified in its path, denoise_fmriprep_output.py will generate an RxT *tsv file containing the denoised time-series, where R is the number of "ROIs" (Regions Of Interest) in the atlas file and T is the number of TRs in the data. The atlas can be either a 3D *nii of discrete integers corresponding to ROIs, or a 4D *nii of probabalistic ROIs (e.g., ICA output).

Example usage:

In Linux, first clone the GitHub repository into your local directory. Alternatively, you can use the GitHub GUI to download and unzip the code.

git clone https://github.com/sjburwell/fmriprep_denoising.git

Next, move into the directory containing the cloned repository's code and create an Anaconda environment containing the necessary Python dependencies using the environment.yml file. The name of the created environment will be called linux fmriprep_denoising. Note that below, creating of the conda environment (i.e., conda env create...) will only need to be done once; after the environment is created, all that is needed is activation of the environment to use it (i.e., conda activate fmriprep_denoising).

cd fmriprep_denoising/
conda env create -f environment.yml
conda activate fmriprep_denoising

Next, run the denoising program. If you don't have AFNI and FSL neuroimaging toolboxes in your system path already, now is the time to load them. Below, they are added by the "load module" command, but things may be different on different machines. You can check whether these programs already exist in your path by using the "which fsl" or "which afni" command. Alternatively, you may need to install these programs from scratch, please see here for AFNI and here for FSL.

module load afni fsl

Finally, once AFNI and FSL are loaded, run the program...

python denoise_fmriprep_output.py \
  --prepdir=/labs/burwellstudy/data/fmri/fmriprep-es2/fmriprep \
  --atlas=./atlases/Conn17f_atlas.nii \
  --pipes=['24P+aCompCor+4GSR','03P+AROMANonAgg'] \
  --overwrite=True \
  --cachedir=./tmpdir \
  --funcpointer=/sub-88302*/*/*/*space-MNI152NLin2009cAsym_preproc*.nii*

In the above code, the arguements explained below:

• 'prepdir' refers to the directory in which fmriprep output is saved
• 'atlas' points to a Nifti file containing either a 3-dimensional "atlas" where voxels' integer values pertain to different regions of interest (e.g., 0=nonbrain, 1=hippocampus, 2=amygdala, etc.) or 4-dimensional series of "maps" where the 4th dimension contains statistical maps of activations (e.g., ICA weights)
• 'pipes' refers to the denoising pipelines to be requested (see below for more info)
• 'overwrite' is a boolean which determines whether previous output(s) will be overwritten (default: False)
• 'cachedir' is the directory in which output files will be written (default: './tmpdir')
• 'funcpointer' is a file-filtering string that can be used to match a selection of functional files for denoising (default: '/*/*/*/*space-MNI152NLin2009cAsym_preproc*.nii*' or ALL functional files in the fmriprep directory).

Pipeline options:

The below options may be passed as a list (i.e., ['00P','09P']) to the "pipes" argument in denoise_fmriprep_output.py:

• 00P: high-pass filter cosine functions, non-steady state outlier TR
• 01P: 00P+global signal
• 02P: 00P+white matter, csf
• 03P: 01P+02P
• 06P: 00P+motion parameters
• 09P: 03P+06P
• 24P: 06P+1st difference and quadratic expansion of the parameters
• 36P: 09P+1st difference and quadratic expansion of the parameters
• 09P+SpkRegFD20: 09P, plus spike regression of high motion (FramewiseDisplacement > 0.20) TRs
• 09P+SpkRegFD25: 09P, plus spike regression of high motion (FramewiseDisplacement > 0.25) TRs
• 09P+SpkRegFD30: 09P, plus spike regression of high motion (FramewiseDisplacement > 0.30) TRs
• 36P+SpkRegFD20: 36P, plus spike regression of high motion (FramewiseDisplacement > 0.20) TRs
• 36P+SpkRegFD25: 36P, plus spike regression of high motion (FramewiseDisplacement > 0.25) TRs
• 36P+SpkRegFD30: 36P, plus spike regression of high motion (FramewiseDisplacement > 0.30) TRs
• 00P+aCompCor: 00P, plus inclusion of the aCompCor columns
• 06P+aCompCor: 06P, plus inclusion of the aCompCor columns
• 24P+aCompCor: 24P, plus inclusion of the aCompCor columns
• 06P+aCompCor+1GSR: 06P, plus inclusion of the aCompCor columns and global signal
• 24P+aCompCor+4GSR: 24P, plus inclusion of the aCompCor columns, global signal its 1st difference and quadratic expansion
• 00P+AROMANonAgg: non-aggressive ICA-AROMA filtering, done using fsl_regfilt
• 01P+AROMANonAgg: non-aggressive ICA-AROMA filtering, done using fsl_regfilt, plus regression of global signal that was extracted after the initial ICA-AROMA filtering
• 02P+AROMANonAgg: non-aggressive ICA-AROMA filtering, done using fsl_regfilt, plus regression of white matter and csf signals that were extracted after the initial ICA-AROMA filtering
• 03P+AROMANonAgg: non-aggressive ICA-AROMA filtering, done using fsl_regfilt, plus regression of global signal, white matter, and csf signals that was extracted after the initial ICA-AROMA filtering
• 00P+AROMAAgg: aggressive ICA-AROMA filtering
• 01P+AROMAAgg: aggressive ICA-AROMA filtering, plus regression of global signal (i.e., 01P)
• 02P+AROMAAgg: aggressive ICA-AROMA filtering, plus regression of white matter and csf signals (i.e., 02P)
• 03P+AROMAAgg: aggressive ICA-AROMA filtering, plus regression of global signal, white matter, and csf signals (i.e., 03P)

For deeper explanation of AROMANonAgg vs AROMAAgg denoising, please see discussion on Neurostars forum.

Output:

In the cache directory, the program denoise_fmriprep_output.py generates subject directories containing tab-separated variable (TSV) files where each column reflects a single time-series for a requested region of interest (ROIs), and each row reflects the time-points of the scanning session (i.e., TRs). For each subject and each denoising approach, there should be one TSV file. Additionally, at the root of the cache directory, there are two TSV files containing meta-data from preprocessing.

References

• Ciric, R., ... Satterthwaite, T.D. (2017). Benchmarking of participant-level confound regression strategies for the control of motion artifact in studies of functional connectivity. NeuroImage, 154, 174-187.
• Parkes, L., Fulcher, B., Yucel, M., & Fornito, A. (2018). An evaluation of the efficacy, reliability, and sensitivity of motion correction strategies for resting-state functional MRI. NeuroImage, 171, 415-436.