fMRI minimal preprocessing pipeline

This section contains an online copy of the book chapter:

Nieto-Castanon, A. (2020). FMRI minimal preprocessing pipeline. In Handbook of functional connectivity Magnetic Resonance Imaging methods in CONN (pp. 3–16). Hilbert Press. doi:10.56441/hilbertpress.2207.6599

alternative sources: pdf (ResearchGate); e-book (Google Books); paperback (Amazon)

Please use the reference above when citing contents of this section

Default preprocessing pipeline

CONN's default preprocessing pipeline (labeled "default preprocessing pipeline for volume-based analyses (direct normalization to MNI-space)" in CONN's gui, or 'default_mni' in CONN's batch commands), performs the following preprocessing steps: functional realignment and unwarp; slice-timing correction; outlier identification; direct segmentation and normalization; and functional smoothing

Functional realignment and unwarp

Functional data is realigned using SPM12 realign & unwarp procedure (Andersson et al. 2001), where all scans are coregistered and resampled to a reference image (first scan of the first session) using b-spline interpolation. This procedure also addresses potential susceptibility distortion-by-motion interactions by estimating the derivatives of the deformation field with respect to head movement, and resampling the functional data to match the deformation field of the reference image. If a double-echo sequence is available, the field inhomogeneity inside the scanner (fieldmap) is also estimated and used for Susceptibility Distortion Correction (SDC) as part of the unwarp step, where the functional data is resampled along the phase-encoded direction in order to correct the absolute deformation state of the reference image caused by field inhomogeneities within the scanner

Slice-Timing Correction

Temporal misalignment between different slices of the functional data, introduced by the sequential nature of the fMRI acquisition protocol,  is corrected using SPM12 slice-timing correction (STC) procedure (Henson et al. 1999), where the functional data is time-shifted and resampled using sinc-interpolation to match the time in the middle of each TA (acquisition time)

Outlier identification

Potential outlier scans are identified from the observed global BOLD signal and the amount of subject-motion in the scanner. Acquisitions with framewise displacement above 0.9mm or global BOLD signal changes above 5 s.d. are flagged as potential outliers (note: an alternative "conservative" setting in CONN uses 0.5mm and 3 s.d. thresholds, while an alternative "liberal" setting uses 2mm and 9 s.d. thresholds). Framewise displacement is computed at each timepoint by considering a 140x180x115mm bounding box around the brain and estimating the largest displacement among six control points placed at the center of this bounding-box faces. Global BOLD signal change is computed at each timepoint as the change in average BOLD signal within SPM's global-mean mask scaled to standard deviation units

Direct segmentation and normalization

Functional and anatomical data are normalized into standard MNI space and segmented into grey matter, white matter, and CSF tissue classes using SPM12 unified segmentation and normalization procedure (Ashburner and Friston, 2005). This procedure iteratively performs tissue classification, estimating the posterior tissue probability maps (TPMs) from the intensity values of the reference functional/anatomical image, and registration, estimating the non-linear spatial transformation best approximating the posterior and prior TPMs, until convergence.  Direct normalization applies this unified segmentation and normalization procedure separately to the functional data, using the mean BOLD signal as reference image, and to the structural data, using the raw T1-weighted volume as reference image. Both functional and anatomical data are resampled to a default 180x216x180mm bounding box, with 2mm isotropic voxels for functional data and 1mm for anatomical data, using 4th order spline interpolation

Functional smoothing

Last, functional data is smoothed using spatial convolution with a Gaussian kernel of 8mm full width half maximum (FWHM), in order to increase BOLD signal-to-noise ratio and reduce the influence of residual variability in functional and gyral anatomy across subjects

Alternatives/additions to the default preprocessing pipeline

While the above preprocessing pipeline is designed to offer robust performance in most common scenarios, researchers are encouraged to explore different alternatives tailored to the specificities of their scanning protocol or analysis plans. Some popular alternatives either directly implemented or supported by CONN are:

1. default preprocessing pipeline for volume-based analyses (indirect normalization to MNI-space) (also known as 'default_mnifield' in CONN's batch commands) is an alternative preprocessing pipeline designed for cases when high-quality fieldmaps are available for susceptibility distortion correction. In those cases  it can be argued (Calhoun et al. 2017) that an indirect normalization procedure results in better performance compared to a direct normalization procedure. In a direct normalization procedure the functional and anatomical data are normalized separately/independently, each using its own non-linear transformation to project the data from its original space to a common MNI-space. In contrast, in an indirect normalization procedure both sets use the same non-linear transformation, which is estimated using only the structural data (often with higher spatial resolution, better tissue contrast, and more anatomical detail). In particular, this pipeline uses the same sequence as CONN's default pipeline, but enforcing the usage of fieldmaps for susceptibility distortion correction during the functional realignment & unwarp step, as well as using an Indirect segmentation and normalization step instead of the original Direct segmentation and normalization step (see "indirect segmentatioon and normalization" section below for details about this step)

2. default preprocessing pipeline for surface-based analyses (also known as 'default_ssfield' in CONN's batch commands) is an alternative preprocessing pipeline designed for surface-based rather than volume-based data analyses. It requires the structural data to have been already processed using FreeSurfer. The pipeline uses the same sequence as CONN's default pipeline, but: 1) enforcing the usage of fieldmaps for susceptibility distortion correction during the functional realignment & unwarp step; 2) using a functional/anatomical coregistration step instead of the original Direct segmentation and normalization step (see "functional/anatomical coregistration" section below for details about this step); 3) sampling the functional data only within the cortical surface (see “resample to fsaverage space” section below for details); and 4) using a functional surface-based smoothing step instead of the original functional smoothing step (see "surface-level smoothing" section below for additional details about this step)

3. fMRIPrep preprocessing pipeline. It is also possible to use data in CONN that has been already either fully or partially preprocessed using other software packages like SPM or FSL. One popular and robust preprocessing pipeline combining multiple tools is fMRIPrep. When using data preprocessed by fMRIPrep, CONN can automatically import the resulting preprocessed MNI-space functional and anatomical volumes, as well as Grey/White/CSF tissue masks,  a full list of potential confound regressor variables, and an associated CONN-formatted first-level realignment, scrubbing, and QC_timeseries covariates that can be directly used by CONN's default denoising procedure. After importing fMRIPrep data it is recommended to run only an additional functional_smoothing preprocessing step before directly proceeding to Denoising the resulting functional data (see fMRIPrep documentation for additional details about fMRIPrep preprocessing pipeline)

Indirect segmentation and normalization

Functional data is first co-registered using an affine transformation to the structural data using SPM12 inter-modality coregistration procedure with a normalized mutual information cost function (Collignon et al. 1995, Studholme et al. 1998). Then the anatomical data is normalized into standard MNI space and segmented into grey matter, white matter, and CSF tissue classes using SPM12 unified segmentation and normalization procedure (Ashburner and Friston, 2005). This procedure iteratively performs tissue classification, estimating the posterior tissue probability maps (TPMs) from the intensity values of the reference anatomical image, and registration, estimating the non-linear spatial transformation best approximating the posterior and prior TPMs, until convergence.  Indirect normalization applies this unified segmentation and normalization procedure to the structural data using the raw T1-weighted volume as reference image, and then applies the same estimated non-linear transformation to the functional data. Both functional and anatomical data are resampled to a default 180x216x180mm bounding box, with 2mm isotropic voxels for functional data and 1mm for anatomical data, using 4th order spline interpolation

Functional/anatomical coregistration

Functional data is co-registered to the structural data using using SPM12 inter-modality coregistration procedure with a normalized mutual information cost function (Collignon et al. 1995, Studholme et al. 1998).  This procedure estimates an optimal affine transformation between the reference functional image (mean BOLD signal) and the reference structural image (T1-weighted volume) that maximizes the mutual information between the two, storing this information in the functional image voxel-to-world mapping header information without resampling the data

Resample to fsaverage space

Functional data is resampled at the location of the subject-specific FreeSurfer-estimated cortical ribbon. This procedure computes the BOLD signal at the location of each vertex in fsaverage icosahedral level-8 tessellation (163,842 vertices and 327,680 faces), projected to the subject-specific white matter surface, and averaged across 10 intervals along the normal between the white matter and pial surfaces

Functional surface-level smoothing

Functional data in fsaverage space is smoothed using iterative diffusion smoothing with 40 iterations, approximately 8mm FWHM kernel (Hagler et al. 2006), in order to increase BOLD signal-to-noise ratio and reduce the influence of residual variability in functional and gyral anatomy across subjects

References

Andersson JLR, Hutton C, Ashburner J, Turner R, Friston K (2001). Modelling geometric deformations in EPI time series. NeuroImage 13:90-919

Ashburner, J. and Friston, K. (1997). Multimodal image coregistration and partitioning—a unified framework. Neuroimage, 6(3), 209-217.

Ashburner, J. and Friston, K.J.. (2005). Unified segmentation. NeuroImage, 26:839–851

Calhoun, V. D., Wager, T. D., Krishnan, A., Rosch, K. S., Seymour, K. E., Nebel, M. B., ... & Kiehl, K. (2017). The impact of T1 versus EPI spatial normalization templates for fMRI data analyses. Human brain mapping, 38(11), 5331-5342.

Collignon, A., Maes, F., Delaere, D., Vandermeulen, D., Suetens, P., & Marchal, G. (1995, June). Automated multi-modality image registration based on information theory. In Information processing in medical imaging (Vol. 3, No. 6, pp. 263-274)

Friston KJ, Williams SR, Howard R, Frackowiak RSJ and Turner R (1995). Movement-related effect in fMRI time-series. Magn Reson Med 35:346-355

Hagler Jr, D. J., Saygin, A. P., & Sereno, M. I. (2006). Smoothing and cluster thresholding for cortical surface-based group analysis of fMRI data. Neuroimage, 33(4), 1093-1103.

Henson, R. N. A., Buechel, C., Josephs, O., & Friston, K. J. (1999). The slice-timing problem in event-related fMRI. NeuroImage, 9, 125

Studholme, Hill & Hawkes (1998). A normalized entropy measure of 3-D medical image alignment. Proc. Medical Imaging 1998, vol. 3338, San Diego, CA, pp. 132-143

How to run CONN preprocessing pipeline

CONN's default preprocessing pipeline can be run using any of the following options:

Option 1: using CONN's gui

After entering in CONN the raw functional data (in Setup.functional) and raw structural data (in Setup.structural) for one or multiple subjects, click on the 'Preprocessing' button, select the first option/pipeline there, labeled 'default preprocessing pipeline for volume-based analyses (direct normalization to MNI-space)' and click 'Ok' and 'Start' (optionally change the 'local processing' option available in that window to 'distributed processing' if you want to parallelize this pipeline across multiple processors or nodes in an HPC cluster)

note: alternatively, if fieldmap sequence files are available,  enter these (e.g. Magnitude and PhaseDiff images) in Setup.functional labeled as a 'fmap' secondary dataset, and then select the second or third pipelines here (for indirect or direct normalization procedures, respectively)

Option 2: using CONN's batch commands

Similarly, if you have already entered your raw functional/anatomical files in CONN (either using the GUI or batch commands), you may run the default preprocessing pipeline using Matlab command syntax:

conn_batch( 'Setup.preprocessing.steps', 'default_mni' )

optionally adding to this command any desired alternative field name/value pairs (see doc conn_batch for additional details), for example:

conn_batch( 'filename',                       '/data/Cambridge/conn_Cambridge.mat', ... 

            'Setup.preprocessing.steps',      'default_mni', ...

            'Setup.preprocessing.sliceorder', 'interleaved (Siemens)', ...

            'Setup.preprocessing.fwhm',       12 )

Option 3: using CONN's modular functions

If you prefer to run this preprocessing pipeline separately from the rest of CONN's functionality (e.g. to preprocess some data and then continue analyzing the preprocessed data on a different software package), you may also run CONN's default preprocessing pipeline using the following Matlab command syntax:

conn_module( 'preprocessing', ...

             'structurals',   {'/data/anat.nii'}, ...

             'functionals',   {'/data/func.nii'}, ...

             'steps',         'default_mni' )

optionally adding to this command any desired alternative field name/value pairs (see doc conn_module and doc conn_batch for additional details), for example:

conn_module( 'preprocessing', ...

             'structurals',   {'/data/anat.nii'}, ...

             'functionals',   {'/data/func.nii'}, ...

             'steps',         'default_mni', ...

             'RT',            2, ...

             'sliceorder',    'interleaved (Siemens)', ...

             'fwhm',          12 )