Deconfounding the Effects of Resting State Activity on Task Activation Detection in fMRI

Abstract

AbstractInferring brain activation from functional magnetic resonance imaging (fMRI) data is greatly complicated by the presence of strong noise. Recent studies suggest that part of the noise in task fMRI data actually pertains to ongoing resting state (RS) brain activity. Due to the sporadic nature of RS temporal dynamics, pre-specifying temporal regressors to reduce the confounding effects of RS activity on task activation detection is far from trivial. In this paper, we propose a novel approach that exploits the intrinsic task-rest relationships in brain activity for addressing this challenging problem. With an approximate task activation pattern serving as a seed, we first infer areas in the brain that are intrinsically connected to this seed from RS-fMRI data. We then apply principal component analysis to extract the RS component within the task fMRI time courses of the identified intrinsically-connected brain areas. Using the learned RS modulations as confound regressors, we re-estimate the task activation pattern, and repeat this process until convergence. On real data, we show that removal of the estimated RS modulations from task fMRI data significantly improves activation detection. Our results thus provide further support for the presence of continual RS activity superimposed on task fMRI response.Keywordsactivation detectionfMRIresting statetask-rest interactions