Cognitive Process Models Reveal Meaningful Brain-Behavior Associations in the Adolescent Brain Cognitive Development Study Stop-signal Task

Abstract

Characterizing associations between individual differences in brain activity and behavior remains a primary challenge in functional neuroimaging research. A growing literature supports the use of formal computational models to represent the mechanistic processes underlying behavior during cognitive tasks. This study applies one such model to the Adolescent Brain Cognitive Development Study stop-signal task (SST) and quantifies relationships between mechanistic model parameters and task-related brain activation using a machine-learning-based predictive modeling approach. With a large sample of task performance and task-related neuroimaging data from 9- to 11-year-olds (n = 6469), we found that SST formal model parameters showed relatively strong relationships with fMRI task-related activation (average variance explained as high as R2 = 26.86 ± 1.69%) compared with empirically derived performance measures (largest R2 = 20.89 ± 1.41%). Our approach suggests that neuroimaging data are most closely associated with evidence accumulation for the go choice process and with attentional lapses that prevent the initiation of the stop process ("trigger failure"). Increased salience network (i.e., insula and anterior cingulate) activity on correct go trials was associated with worse evidence accumulation, and greater visual cortex activity on error trials was associated with fewer attentional lapses. In addition, through relationships with phenotypic measures of inhibition, impulsivity, and cognition, we provide evidence supporting the formal model's construct validity. We demonstrate the utility of computational cognitive modeling for revealing stronger, and more meaningful, associations between brain function and behavior.