Assigning independent components to canonical brain networks

September 14, 2023

A lot of the work my group does these days focuses on independent component analysis (ICA). ICA is a blind-source separation algorithm that is a popular way to analyze fMRI data. With ICA, you get a set of spatial independent component (IC) maps and a “mixing matrix” that contains the temporal activity associated with each IC. From that mixing matrix, you can compute functional connectivity (FC) matrices, either static or dynamic. When it comes to displaying FC matrices, we typically want to group the ICs by brain network, which results in a nice block-diagonal structure that aids in visual interpretation. We may also want to summarize FC or the spatial ICs by network.

Therefore, we often need to “assign” ICs to a canonical network or area of the brain. While there are many ways to approach this, my group has been experimenting with this for a while, and we finally have a procedure that works pretty well. In this post, I will describe our approach and show some examples.

If you happen to be working with the Human Connectome Project group ICA at the 25- or 100-component resolution, this recent preprint describes the IC assignment method and reports the final IC network labels. Please cite that paper if you use those assignments or adopt our methodology. All of the figures shown below are generated using our our ciftiTools R package, which is freely available on Github and CRAN. Links available at statmindlab.com/software. If you use ciftiTools, please cite Damon Pham’s paper describing and illustrating the software.

Here we are attempting to match our ICs to the well-known 17 Yeo cortical networks and the Freesurfer subcortical parcels. However, the procedure described below should work well for other sets of networks/regions.

**… this post was originally written on my old blog. Read the full post here. **