Population average atlas for RecobundlesX

Multi-atlas bundle segmentation

This data is made to be used with the following script:
https://github.com/scilus/scilpy/blob/master/scripts/scil_recognize_multi_bundles.py

Or the following Nextflow pipeline:
https://github.com/scilus/rbx_flow

This script is in fact a multi-atlas, multi-parameters version of Garyfallidis et al. (2018) with labels fusions. We name this algorithm RecobundlesX. This version facilitates parameter exploration and stabilizes/robustifies the results from Recobundles.

Etienne St-Onge, Kurt Schilling, Francois Rheault, "BundleSeg: A versatile, reliable and reproducible approach to whitte matter bundle segmentation.", arXiv, 2308.10958 (2023)

Rheault, François. "Analyse et reconstruction de faisceaux de la matière blanche." Computer Science (Université de Sherbrooke) (2020), https://savoirs.usherbrooke.ca/handle/11143/17255

Usage
Here is an example (for more details use `scil_recognize_multi_bundles.py -h`) :

`antsRegistrationSyNQuick.sh -d 3 -f ${T1} -m mni_masked.nii.gz -t a -n 4`
`scil_recognize_multi_bundles.py ${TRACTOGRAM} config_fss_1.json atlas/*/ output0GenericAffine.mat --out_dir ${OUTPUT_DIR}/ --log_level DEBUG --minimal_vote 0.4 --processes 8 --seed 0 --inverse -f`

To facilitate interpretation, all endpoints were uniformized head/tail. To see, which side of a bundle is head or tail, you can load the atlas bundle into the software MI-Brain

Notes on bundles
- AC and PC were added mostly in case the atlas is used for lesion-mapping or figures. Likely, segmentation won't produce good results. This is mostly due to difficult tracking for these bundles.
- The CC are split for each lobe. However, for technical consideration, the frontal portion was split in two to facilitate clustering and segmentation. For the same reason, the portion fanning to the pre/post central gyri were separated.
- The streamlines present in the CC are homotopic, Recobundles will allow for variation and thus lead to 'some' heterotopy. However, it is expected that the results will be mostly homotopic.
- CG has 3 possible endpoint locations. However, the full extent of the tail is difficult to track and is often missing.
- FPT and POPT should terminate in the pons. However, to fully capture candidate streamlines and improve segmentation quality even streamlines reaching down the brainstem are selected. 
- PYT should reach down the brainstem. For similar reasons to the FPT/POPT, streamlines ending in the pons are selected. Otherwise, fanning is affected and bundles is too skinny. 
- OR_ML will most likely have difficulty capturing the full ML. However, this is often due to difficult tracking.
- The cerebellum is often cut due to acquisition FOV. In such a case, all projection bundles will be more difficult to recognize and most cerebellum bundles will be missing (ICP, MCP, SCP).

See Mosaic of bundles here.

Acronym
AC - Anterior commisure
AF - Arcuate fasciculus
CC_Fr_1 - Corpus callosum, Frontal lobe (most anterior part)
CC_Fr_2 - Corpus callosum, Frontal lobe (most posterior part)
CC_Oc - Corpus callosum, Occipital lobe
CC_Pa - Corpus callosum, Parietal lobe
CC_Pr_Po - Corpus callosum, Pre/Post central gyri
CC_Te - Corpus callosum, Temporal lobe
CG - Cingulum
FAT - Frontal aslant tract
FPT - Fronto-pontine tract
FX - Fornix
ICP - Inferior cerebellar peduncle
IFOF - Inferior fronto-occipital fasciculus
ILF - Inferior longitudinal fasciculus
MCP - Middle cerebellar peduncle
MdLF - Middle longitudinal fascicle
OR_ML - Optic radiation and Meyer's loop
PC - Posterior commisure
POPT - parieto-occipito pontine tract
PYT - Pyramidal tract
SCP - Superior cerebellar peduncle
SLF - Superior longitudinal fasciculus
UF - Uncinate fasciculus