A brain-based universal measure of attention: predicting task-general and task-specific attention performance and their underlying neural mechanisms from task and resting state fMRI

Abstract

Attention is central for many aspects of cognitive performance, but there is no singular measure of a person's overall attentional functioning across tasks. To develop a universal measure that integrates multiple components of attention, we collected data from more than 90 participants performing three different attention-demanding tasks during fMRI. We constructed a suite of whole-brain models that can predict a profile of multiple attentional components - sustained attention, divided attention and tracking, and working memory capacity - from a single fMRI scan type within novel individuals. Multiple brain regions across the frontoparietal, salience, and subcortical networks drive accurate predictions, supporting a universal (general) attention factor across tasks, which can be distinguished from task-specific attention factors and their neural mechanisms. Furthermore, connectome-to-connectome transformation modeling enhanced predictions of an individual's attention-task connectomes and behavioral performance from their rest connectomes. These models were integrated to produce a new universal attention measure that generalizes best across multiple, independent datasets, and which should have broad utility for both research and clinical applications.

Functional Connectivity during Encoding Predicts Individual Differences in Long-Term Memory

“…We successfully built predictive models of individual differences in recognition and recollection memory performance based on whole-brain FC during an N-back task. On the contrary, we found that FC during rest had limited predictive power compared to task, consistent with other recent works on increased prediction performance from task data compared to resting data (e.g., Yoo et al, 2018Yoo et al, , 2020Yoo et al, , 2021Jiang et al, 2020;Tomasi & Volkow, 2020;Greene et al, 2018;Rosenberg et al, 2016). By comparing the recognition memory model and a model built to predict WM performance based on the same FC data, we found that there are some shared components between WM and recognition memory supported by evidence from overlapping predictive networks and cross-measure prediction.…”

“…The entire pipeline described above was repeated 1000 times with different group partitions, resulting in 1000 mean r values. Significance of the model performance was assessed with 1000 iterations of nonparametric permutation testing (Scheinost et al, 2019), and p values were calculated as (1 + number of null models whose r values ≥ mean of all iterations of empirical models)/ (1 + 1000) ( Yoo et al, 2021;Avery et al, 2019). Resulting p values were Bonferroni corrected for multiple comparisons (3, 2, 4, and 6 for the no-control, overlapping, partial regression, and orthogonalization models, respectively).…”