2 3 4 5 Clusters 0.2 0.0 0.2 0.4 0.6 0.8 1.0
Adjusted Rand Index
rsfMRI dMRI rsfMRI dMRI 0.0 0.1 0.2 0.3 0.4 0.5 Silhouette Index k 2 3 4 5
Niels Reuter
1,2, Sarah Genon
2, Shahrzad Kharabian
2, Felix Hoffstaedter
1,2, Tobias
Kalenscher
3, Kaustubh. R. Patil
1,2& Simon B. Eickhoff
1,21Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf, Düsseldorf, Germany;
2Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich, Jülich, Germany;
3Department of Comparative Psychology, Institute of Experimental Psychology, Heinrich-Heine University, Düsseldorf, Düsseldorf, Germany n.reuter@fz-juelich.de
CBPtools: A Python package for
regional connectivity-based parcellation
Here we have outlined and demonstrated the effectiveness of CBPtools to procure rsfMRI and
dMRI connectivity-derived parcellations.
By unifying methodological choices behind the rCBP procedure, we offer a fast and stable means
to effortlessly and efficiently conduct reproducible and data-driven parcellation analysis.
Computational demands highlighted by complex algorithms and large datasets are mitigated by
efficient parallel execution of the procedure using the Snakemake [6] software.
Acknowledgments: This study was supported by the European Union‘s Horizon 2020 Research and Innovation Programme under Grant Agreement No. 785907 (HBP SGA2) References:
[1] Eickhoff SB, Thirion B, Varoquaux G, Bzdok D (2015) Connectivity-based parcellation: Critique and implications. Hum Brain Mapp 36:4771–4792. doi: 10.1002/hbm.22933;
[2] Ruan J, Bludau S, Palomero-Gallagher N, et al (2018) Cytoarchitecture, probability maps, and functions of the human supplementary and pre-supplementary motor areas. Brain Struct Funct 223:4169–4186. doi: 10.1007/s00429-018-1738-6; [3] Glasser MF, Sotiropoulos SN, Wilson JA, et al (2013) The minimal preprocessing pipelines for the Human Connectome Project. Neuroimage 80:105–124. doi: 10.1016/j.neuroimage.2013.04.127;
[4] Van Essen DC, Smith SM, Barch DM, et al (2013) The WU-Minn Human Connectome Project: an overview. Neuroimage 80:62–79. doi: 10.1016/j.neuroimage.2013.05.041 [5] Generated with Mango (Multi-image analysis GUI; http://ric.uthscsa.edu/mango/)
[6] Köster J, Rahmann S (2012) Snakemake--a scalable bioinformatics workflow engine. Bioinformatics 28:2520–2522. doi: 10.1093/bioinformatics/bts480
Regional connectivity-based parcellation (rCBP) is a widely used
procedure for revealing the underlying brain organization of a region-of-interest (ROI) by mapping its whole-brain connectivity profiles [1].
No standardized guidelines currently exist for applying rCBP No standardized software currently exist for applying rCBP Most studies utilizing it are based on in-house software.
Here, we outline and demonstrate CBPtools, an open-source plug-and-play yet customizable rCBP software package made in Python (3.5+). By introducing CBPtools we provide researchers the means to conduct reproducible, data-driven rCBP analyses on multiple neuroimaging modalities and large amounts of subject data.
Group-clustering labels are mapped on
seed image
2-cluster solution shown to best fit input
data (Fig. 7)
ARI between cytoarchitectonically defined
subdivision and both dMRI and rsfMRI > .7
1. Connectivity-Based Parcellation Workflow
2. Connectivity Computation
3. Clustering
6. NIfTI Images
Get CBPtools @ https://github.com/inm7/cbptools
7. Conclusion
k-means clustering used to cluster
connectivity matrices per subject and per k
Randomly chooses k seed voxels as
cluster centers (Fig. 3, circles)
Assigns each seed voxel to closest
centroid using Euclidean distance of its connectivity profile
Move centroid to the mean of all
assigned voxels and repeat
Subject clustering results are used to construct a representative group clustering.
Hierarchical clustering
creates reference clustering of all subject clusterings
Subject clusterings are
relabeled to match
reference clustering
Mode of all relabeled subject
clusterings taken as group clustering
5. Grouping
A F D B C G EFig. 3 k-means clustering with k = 3. Voxels
(squares) group around the 3 centroids (circles)
X
Connectivity matrices are calculated for each subject using linear
correlations for resting-state functional magnetic resonance imaging (MRI) and probabilistic tractography for diffusion MRI data
A B C F G D E Target Voxels S e e d V o xe ls
1
2
3
1
2
3
2
1
3
3
2
1
1
3
2
2
3
1
3
1
2
Fig. 5 Relabeling through permutations.Here is a recursion tree for
permutations of the labels “1, 2, 3“
Fig. 2 Computation of a seed-voxel by target-voxel connectivity matrix
Fig. 1 CBPtools workflow for applying rCBP to diffusion MRI (dMRI) or resting-state functional MRI (rsfMRI) data
S im ila ri ty A B C D E F G
Fig. 4 Hierarchical clustering with a 3-cluster cut-off.
Voxels (squares) grouped at each step
Fig. 9 3D representation [5] of volumetric 2-cluster combined preSMA and
SMA region-of-interest solution
4. Validity & Similarity
Fig. 8 Adjusted rand index scores between subject clusterings and group-clusterings
for rsfMRI and dMRI at k = 2, 3, 4, and 5
Fig. 7 Silhouette index for rsfMRI and dMRI at k =
2, 3, 4, and 5
Example data: 300 unrelated healthy subjects from the
Human connectome Project [3, 4]
Example region: combined preSMA-SMA mask from the
Juelich Cytoarchitectonic Atlas [2]
Cluster quality metrics computed for each subject- and
group-clustering at all values of clustering granularity k
Mask Preprocessing
Mask Preprocessing Mask PreprocessingMask Preprocessing
Input
Subsample
Remove every second voxel
Subsample
Remove every second voxel
Seed extraction
Remove seed voxels from target space
Seed extraction
Remove seed voxels from target space
Upsample
Decrease voxel size of mask
Upsample
Decrease voxel size of mask
Connectivity
Connectivity ConnectivityConnectivity probtrackx2 Probabilistic tractography to derive connectivity probtrackx2 Probabilistic tractography to derive connectivity Apply Masks
Mask input time-series with seed- and target-masks
Apply Masks
Mask input time-series with seed- and target-masks
Compute Connectivity
Linear correlations between seed- and target-masked time-series
Compute Connectivity
Linear correlations between seed- and target-masked time-series Grouping Grouping Clustering Clustering Connectivity Connectivity Transform
Fisher Z and/or PCA transform connectivity
Transform
Fisher Z and/or PCA transform connectivity
Transform
Fisher Z and/or PCA transform connectivity
Transform
Fisher Z and/or PCA transform connectivity Similarity Similarity Validity Validity dMRI BedpostX output Report Report rsfMRI Resting-state time-series Seed Mask
ROI NIfTI image
Target Mask
Feature NIfTI image
Median Filter
Apply a median filter to the ROI image
Median Filter
Apply a median filter to the
ROI image MasksMasks
Downsample
Increase voxel size of mask
Downsample
Increase voxel size of mask
Median Filter
Apply a median filter to the ROI image
Median Filter
Apply a median filter to the ROI image
Target Mask Seed Mask
Seed Mask Target Mask
Signal Cleaning
Smoothing, confound regression, band- pass filtering
Signal Cleaning
Smoothing, confound regression, band- pass filtering A B C D E F G