Resting State fMRI Preprocessing Pipeline (work in progress)

Resting State Preprocessing Pipeline

Preprocessing Steps suggested by Eduardo (in the order presented).

T1w Preprocessing

For rsfMRI, we need tissue segmentation of the T1w images and registration to the EPI space. This can be achieved using the Bpipe, then the CIVET pipeline. If we can join this, it would be easy to extract the GM, WM, CSF and Brain masks, convert them to NIFTI, register and resample to EPI space for nuisance variables extraction.

DICOM to NIFTI

Importantly, all preprocessing is usually done in NIFTI format.

Quality Control

Humans should always do a quick check of the data by hand (eye). You can use FSLVIEW, AFNI or MRVIEW (mrtrix). The FSLVIEW movie feature is quite useful for fMRI data.

Despike

AFNI tool for despiking data should be run first.

Motion Correction aka Realignment

This does not correct for motion artifacts in the BOLD signal of the voxels, it only realigns every volume to a reference volume. It does provide information for the motion correction afterwards, therefore it is important to save the outputs.

Slice-Timming Correction

This is usually done after the Motion Correction. It is important to know about the details of how the data was acquired, in this case the order and direction of the acquired fMRI sequence (i.e. bottom-up). The best practice for this would be to find the exact direction, order and timming by reading the header using AFNI, and then correct based on the header of the file. However, using FSL with prior information should be good enough.

Field Mapping (Distorsion Correction)

For this we use the fMRI sequence with the opposite acquisition direction to the original EPI sequence (ie. AP vs PA), and we use FSL-TOPUP. Here is a script example for it. It may be also possible to adapt 'dwipreproc' from MRTRIX to do this, as it read the necessary information from the header. Not everyone does Field Mapping but it is preferable.

Artifact and Structured Noise Removal

The method we use is CompCor and nuisance regression (24 motion parameters + CompCor (CSF & WM).

Volume Censoring aka Scrubbing

Usually 2 mm - 2.5 mm of threshold is adviced.

How many bad volumes can I accept per subject before rejecting it? It is a very personal question. Depends on the type of population, the brain area of study and volumes acquired. I suggest a maximum of 40% loss of volumes, but this varies.

Bandpass

Not necessary, but used. AFNI is preferable.

Spatial Smoothing

AFNI is preferable.

Normalization

rsfMRI to MNI space.

After this preprocessing, the data is ready for processing using Seed-based, ICA or Graph theory-based analysis.