MRI processing

Tractoflow

MRI processing was done using TractoFlow, which is a fully automated and reproducible diffusion MRI processing pipeline used to reconstruct the whole brain white matter architecture.

How

TractoFlow first outputs diffusion metrics.

1. DTI metrics were computed from the b=1000 mm2/s shell using scil_compute_dti_metrics.py which is based on dipy.reconst.dti (tutorial). The tensor fit method: weighted least squares (WLS)
2. fODF were computed from b=2000 mm2/s (>50 direction -> HARDI) shell using scil_compute_ssst_fodf.py which is based on dipy.reconst.csdeconv(tutorial). The FRF were by default. fODF metrics were then computed using scil_compute_fodf_metrics.py see [Raffelt et al. NeuroImage 2012] and [Dell'Acqua et al. HBM 2013] for the definitions. Tractoflow then outputs tractograms
3. PFT tractogram are computed from fODF maps using scil_compute_pft.py based on dipy.tracking.local_tracking.ParticleFilteringTracking (tutorial). We use a probabilistic tracking algorithm, the stopping criterion is based on partial volume estimation (PVE) maps outputed by FSL-FAST and the continuous map criterion (CMC).
4. Local tracking tractogram are computed from fODF maps using scil_compute_local_tracking.py based on dipy.tracking.local_tracking.LocalTracking (tutorial). We use a probabilistic tracking algorithm, the stopping criterion is based on a FA map treshold mask
5. Free Water maps were computed from b=300 mm2/s and b=1000 mm2/s shells using scil_compute_freewater.py which is based on Accelerated microstructure imaging via convex optimization (AMICO). Instructions can be found here

Robustness and reproducibility

leveraging Nextflow and Singularity for robustness and reproducibility. takes raw Diffusion-weighted and T1-weighted images as input and outputs diffusion metrics and a whole brain tractogram.

To pay attention

• The singularity.conf file is not necessary for newer versions of singularity: nextflow -c singularity.conf run <COMMAND>

• When using tractoflow use --input <DATASET_ROOT_FOLDER> instead of --root <DATASET_ROOT_FOLDER>

ROI segmentation

Establishing correspondences across brains for the purposes of comparison and group analysis is almost universally done by registering images to one another either directly or via a template.

1. De-skulling aids volume registration methods.
2. Custom-made optimal average templates improve registration over direct pairwise registration.

RecobundleX

Automatic bundle segmentation was done using rbx_flow, which is a fully automated and reproducible pipeline based on the dipy tool Recobundle.

To pay attention

• If you download Ants it can be helpful to change the command make -j 4 to make -j 1 to not overflow RAM. Install with the following instructions on GitHub or on NeuroDebian (this did not work for me because the package is not available for my distribution -- Ubuntu 20.04 Focal Fossa)
• scil_recognize_multi_bundles.py use --outdir and not --output

Nextflow

RecobundleX can be run using a nextflow pipeline called rbx_flow that can be found here. It is important to note that it must be run with singularity DSL1 because Operator `into` has been deprecated -- it's not available in DSL2 syntax. To use DSL1 run command export NXF_DEFAULT_DSL=1

Our pipeline (bundle_segmentation original)

Nextflow pipeline:

1. Register Atlas to reference image: brainnetome atlas is sent to subject space. To compute transformation from atlas space to subject space we register 'FSL_HCP1065_FA_1mm' to the subject's native map. We then apply transformation (affine + warp) on Brainnetome atlas with generic_label interpolation.
2. Create masks of mPFC and NA: these are the ROIs for filtering the tractogram based on the atlas
3. Filter tractogram: remove fibers that do not join both ROIs and remaining outliers
4. Compute centroid: important step for tractometry