Talk: Using Functional Connectivity to Predict Working Memory Performance in the Human Connectome P…
Speaker: Corey Richier, University of Illinois at Urbana Champaign (grid.35403.31)
Title: Using Functional Connectivity to Predict Working Memory Performance in the Human Connectome Project
Emcee: Ehsan Rezayat
Backend host: Forouzan Farahani
Details: https://neuromatch.io/abstract?submis...
Twitter: / cjrichier
Presented during Neuromatch Conference 3.0, Oct 26-30, 2020.
Summary: Introduction: A central goal of neuroscience is predicting behavior based on features of the brain. Resting state functional connectivity (rsFC) and task evoked functional connectivity are both commonly utilized measures of in vivo brain function. Utilizing data from the HCP, our project sought to explore their potential as a predictor for various other aspects of neural function and behavior. We attempted to predict working memory task performance based on patterns of task evoked activity and rsFC. We theorized that task based activity would be more predictive than rsFC.
Methods: In a subset (n = 338) of the Human Connectome Project, we used the Glasser parcellation to establish functional connectivity patterns during resting state as features to model. As the first preprocessing step, we generated a functional connectivity matrix for all parcels across all participants. Separate matrices were created for each task and resting state separately. We then utilized connectome parametric mapping (CPM, Shen et al, 2017) to model the prediction of working memory performance.
Results: Resting-state was found as a better predictor (R2= 13%) of behavioral performance respect to task-evoked working memory (R2= 0.5%) activity. However, the overall predictive power of either model was not high. Nodes belonging to the default mode and frontal parietal network were found to better predict working memory performance.
Discussion: Interestingly, resting state was a better predictor of WM performance than the WM task activity, but overall performance was not high. Nodes belonging to the frontal-parietal network were more engaged in the working memory task than nodes in the default mode network. The narrow differences between task evoked and rsFC may partially explain the difference in model performance. Another potential contributing factor is the comparatively smaller amount of task-evoked activity to the rsFC. These factors have considerations for building predictive models using fMRI data.