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.
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_parcellation.py: Main python library for performing Chimera parcellations.
parcTypes.json: JSON file especifying the available parcellation sources per supra-region.
ANNOT_atlases and GCS_atlases: Folder containing cortical atlases in .annot and .gcs file formats.
mni_icbm152_t1_tal_nlin_asym_09c: Folder containing the reference atlas used by the MIAL atlas-based thalamic parcellation method. The atlas is referenced in standard MNI (Montreal Neurological Institute) space with a high resolution T1 weighted image (ICBM 2009c Nonlinear Asymmetric ).
thalamic_nuclei_MIALatlas: Folder containing the spatial probabilistic maps of 14 thalamic nuclei.
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).