Schedule of Events | Symposia

Unveiling the cognitive relevance of functional connectivity through deconfounding

Poster Session D - Monday, March 31, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

Michael Cole¹ (mwcole@mwcole.net), Kirsten Peterson, Lakshman Chakravarthula, Ravi Mill, Ruben Sanchez-Romero; ¹Rutgers University

Functional connectivity is typically defined as the statistical similarity between neural time series, yet this defines the measures used rather than what brain properties are being measured. We propose that the proper theoretical target of functional connectivity is the routes of neural activity flow – the propagation pathways of activity between neural populations. Activity flows generate neurocognitive functionality, such that measured functional connections are relevant to cognition (and cognitive disorders) to the extent that they accurately describe activity flow routes. This perspective reveals a strong need to remove the confounding of functional connectivity estimates introduced by the field-standard bivariate Pearson correlation approach. Confounded connections can arise due to a third region (or artifact) creating a falsely inferred connection among two or more other regions (e.g., A<--B-->C creating a false A–C connection). We recently reported that regularized partial correlation substantially reduces confounding from motion artifacts and neural confounding in resting-state fMRI data. Here, we tested whether regularized partial correlation also reduces confounding from task stimuli, as is common when estimating task-state functional connectivity with bivariate correlation. We used neural mass modeling and empirical fMRI to test the effectiveness of regularized partial correlation (graphical lasso) for removing task stimuli confounds. Regularized partial correlation performed better than expected, removing 99.8% of the task stimuli confounds while reducing false negatives. Further, this approach was more effective than best practices for regressing out task timing. These results demonstrate the efficacy of regularized partial correlation for removing stimulus-driven confounds for task-state functional connectivity estimation.

Topic Area: METHODS: Neuroimaging