4th seminar
While many efforts to improve brain-behaviour associations (BBAs) tend to focus on the quality of the brain data, optimizing the yield from the behavioural component tends to receive less attention. Computational modelling approaches hold promise for improving BBAs by modelling what may be more elementary cognitive sub-processes. In this presentation I will show how use of a specific model for the motor response inhibition Stop-Signal task (a racing-diffusion ex-Gaussian model) improves the strength of associations with brain function. Using a Bayesian parameter estimation procedure, 11 parameter distributions related to choice responding and response inhibition were derived for ABCD study participants. We categorized these estimates into “go process” and “stop process” parameters and used regularized regression to predict these estimates using region-of-interest task-fMRI data. Task-fMRI regressors were Stop-Signal Task contrasts (i.e., correct go vs. fixation, correct stop vs. fixation, incorrect go vs. fixation, incorrect stop vs. fixation). We found that the computational model captured all variance explained by standard behavioural measures of task performance suggesting that it is successful in decomposing stop task performance into more elementary processes. Moreover, large brain-behaviour associations were observed with, for example, brain data explaining more than 25% of the out-of-sample inter-individual variance in an Evidence Accumulation parameter (reflecting information processing speed). These analyses suggest that decomposing task performance into sub-process components increases brain-behaviour associations for the Stop Signal Task and suggests that many components with meaningful brain-behaviour associations are related to more general choice processes rather than specific to the “stop” process.