CHIMERA: An open source framework for combining multiple parcellations

Creating multi-source parcellations of the human brain is a fundamental task at several steps of the MRI analysis research workflow. Chimera facilitates this otherwise difficult operation with an intuitive and flexible interface for humans and machines, thereby assisting in the construction of sophisticated and more reliable processing pipelines. This repository contains the source code and atlases needed by Chimera.

Parcellations fusion

Chimera defines nine different supra-regions (cortex, basal ganglia, thalamus, amygdala, hippocampus, hypothalamus, cerebellum, brainstem and white-matter). Basal ganglia includes only the regions that are not labeled as supra-regions. Subdivisions in each supra-region will be populated with the parcellation information of a single source. The available parcellation sources per supra-region, as well as one corresponding parcellation name, and a one-character unique identifier are configured in a JSON (JavaScript Object Notation) file.
Chimera code: A sequence of nine one-character identifiers (one per each supra-region) unambiguosly denotes a single instance of combined parcellation (Figure. 1B). Given the sequence of nine identifier characters, Chimera selects the atlas and/or applies the corresponding methodology to obtain the parcellation for each supra-region. These supra-region-specific parcellations are finally integrated to obtain the combined volumetric parcellation for each input subject, as well as its corresponding tab-separated values table of labels, region names, and rendering colors for visualization. Chimera uses FreeSurfer to map cortical templates from fsaverage to individual space. It also applies different methods to obtain the hippocampal subfields and brainstem parcellations as well as the thalamic, amygdala and hypothalamic nuclei segmentations. FIRST and ANTs are also used for segmenting subcortical structures and thalamic nuclei respectively.

Requirements

Required python packages:

• pandas, pybids, numpy, nibabel, time, os, pathlib, argparse, sys, csv, subprocess

Required image processing packages:

• FreeSurfer (version>7.2.0), FSL, ANTs

---

Options:

Brief description of input options:

Option	Description
--regions, -r	List available parcellations for each supra-region.
--bidsdir, -b	BIDs dataset folder. Different BIDs directories could be entered separating them by a comma.
--parcodes, -p	Sequence of nine one-character identifiers (one per each supra-region).
--derivdir, -d	Derivatives folder. Different directories could be entered separating them by a comma.
--freesurferdir, -fr	FreeSurfer subjects dir. If the folder does not exist it will be created.
--scale, -s	Scale identification. This option should be supplied for multi-resolution cortical parcellations (e.g. Lausanne or Schaeffer).
--seg, -e	Segmentation identifier.
--nthreads, -n	Number of processes to run in parallel (default= Number of cores - 4).
--growwm, -g	Grow of GM labels inside the white matter (mm).
--subjids, -ids	Subject IDs. Multiple subject ids can be specified separating them by a comma.
--mergectx,, -mctx	Join cortical white matter and cortical gray matter regions.
--force, -f	Overwrite the results.
--verbose, -v	Verbose (0, 1 or 2).
--help, -h	Help.

---

Usage

General command line to use Chimera:

    $ chimera -b <BIDs directory> -d <Derivatives directory> -p <Chimera code>

This command will run Chimera for all the subjects in the BIDs directory.

Simple examples

1. Running Chimera for 3 different parcellation codes (LFMFIIFIF,SFMFIIFIF,CFMFIIFIF). This will obtain the combined parcellations for all the T1-weighted images inside the BIDs dataset.

    $ chimera -b <BIDs directory> -d <Derivatives directory> -p LFMFIIFIF,SFMFIIFIF,CFMFIIFI

2. Running Chimera for T1-weighted images included in a txt file:

    $ chimera -b <BIDs directory> -d <Derivatives directory> -p LFMFIIFIF -ids <t1s.txt>

Example of t1s.txt file | sub-00001_ses-0001_run-2 | sub-00001_ses-0003_run-1 | sub-00001_ses-post_acq-mprage

3. Cortical volumes will grow 0 and 2 mm respectively inside the white matter for the selected cortical parcellations.

    $ chimera -b <BIDs directory> -d <Derivatives directory> -p LFMFIIFIF -g 0,2

Main files in the repository

1. chimera.py__: Main python library for performing Chimera parcellations.
2. supraregions_dictionary.json__: JSON file especifying the available parcellation sources per supra-region.
3. annot_atlases and gcs_atlases: Folder containing cortical atlases in .annot and .gcs file formats.

Parcellations and methodologies for each supra-region

1. Cortical parcellation

Code	Citation	Code	Citation
D	Desikan et al, 2006	X	Destrieux et al, 2009
T	Klein and Tourville, 2012	B	Fan et al, 2016
R	Broadmann, 1909	C	Campbell, 1905
K	Kleist, 1934	L	Symmetric version of Cammoun et al, 2012
H	Glasser et al, 2016	S	Schaefer et al, 2018
M	Smith et al, 1907	V	von Economo and Koskinas, 1925
Y	Yeo et al, 2011	F	Flechsig, 1920

2. Basal ganglia parcellation

Code	Citation	Code	Citation
F	Fischl et al, 2002	R	Patenaude et al, 2011

3. Thalamus parcellation

Code	Citation	Code	Citation
F	Fischl et al, 2002	I	Iglesias et al, 2018
M	Najdenovska and Alemán-Gómez et al, 2018	R	Patenaude et al, 2011

4. Amygdala parcellation

Code	Citation	Code	Citation
F	Fischl et al, 2002	I	Saygin et al, 2017
R	Patenaude et al, 2011

5. Hippocampus parcellation

Code	Citation	Code	Citation
F	Fischl et al, 2002	I	Iglesias et al, 2015
H	Iglesias et al, 2015	I	Patenaude et al, 2011

6. Hypothalamus parcellation

Code	Citation	Code	Citation
F	Based on in-house protocol	I	Billot et al, 2020

7. Cerebellum parcellation

Code	Citation
F	Fischl et al, 2002

8. Brainstem parcellation

Code	Citation	Code	Citation
F	Fischl et al, 2002	I	Iglesias et al, 2015
R	Patenaude et al, 2011

9. Gyral white matter parcellation

Code	Citation
F	Cortical (Depends on the cortical parcellation)

Results

Chimera parcellations were generated using the following codes: LFMIIIFIF, HFIIIIFIF, BFIIHIFIF (162, 492 and 314 regions respectively). Figure 2A shows the corresponding results of the fused parcellations for a single subject. By filtering each individual's tractogram with the corresponding Chimera parcellations, we generated connectivity matrices (Figure 2B).

License