Multi-Scale Brain Parcellator: a BIDS App for the Lausanne Connectome Parcellation

Connectome Mapper (CMP), part of the Connectome Mapping Toolkit (CMTK), is an open-source software which implements full anatomical, diffusion and resting state functional MRI processing pipelines to map connectivity matrices from raw Diffusion/T1/T2 and BOLD data (Daducci 2012). In parallel of its third release (connectome-mapper-3.readthedocs.io), we developed an independent tool: the Multi-Scale Brain Parcellator. This tool, which takes BIDS structured datasets as input, implements a 5-scale brain gray matter parcellation (Cammoun 2012) derived from the Desikan-Killiany atlas (Desikan 2004) and extended with new structures including a subdivision of the thalamus into 7 nuclei, the hippocampus into 12 subfields and the brainstem into 4 sub-structures. Such a brain parcellation can serve many add on applications such as volumetry and definition of regions of interest for tractography or functional connectivity analysis.

The pipeline workflow of the Multi-Scale Brain Parcellator is written in Python and uses Nipype (Gorgolewski 2011) to interface with FreeSurfer and CMTK. It is encapsulated in a BIDS App (Gorgolewski 2017), a framework based on light container technologies which promotes portability and reproducibility. Taking as input BIDS datasets (Gorgolewski 2016) that include an anatomical scan (T1w or MPRAGE), the Multi-Scale Brain Parcellator interfaces with FreeSurfer 6.0.1 to perform resampling to isotropic resolution, Desikan-Killiany brain parcellation (Desikan, 2004), brainstem parcellation (Van Leemput 2015), and hippocampal subfields segmentation (Iglesias 2015). Then, using the new version of CMTK, it performs cortical brain parcellation at 5 different scales (Cammoun 2012), probabilistic atlas-based segmentation of the thalamic nuclei (Najdenovska 2018), and combination of all segmented structures, to create the final parcellation at each scale. The container image of the BIDS App is built using Docker container technology (Merkel 2014) and available as sebastientourbier/multiscalebrainparcellator on the Docker Hub, but can be easily converted into a Singularity image for large scale processing on clusters.

The Multi-Scale Brain Parcellator is executed by running the Docker/Singularity image with the set of arguments specific to a BIDS App and the following set of additional options: –thalamic_nuclei to segment the thalamic nuclei, –hippocampal_subfields to segment the hippocampal subfields, and –brainstem_structures to parcellate brainstem. The processing workflow is illustrated by Fig. 1. Note that only the subject level processing pipeline is implemented (–participant). The BIDS App outputs FreeSurfer derivatives in <deriv-directory>/freesurfer/sub-<participant-label>/, and the different generated brain parcellations in <deriv-directory>/cmp/sub-<participant-label>/anat/ as sub-<participant-label>_space-orig_label-L2018_desc-<scale1:scale5>_atlas.nii.gz. Parcellisation labels for each scale are described in their respective <brain-parcellation-file>.graphml file.