Denoising for task-modulated functional connectivity: current practices, empirical benchmarking, and an SPM toolbox

Abstract

Denoising is routinely applied in resting-state functional connectivity (RSFC) analyses of fMRI data. However, in task-modulated functional connectivity (TMFC) studies, which assess dynamic changes in FC between task conditions, advanced denoising strategies are often overlooked. In a systematic review of psychophysiological interaction (PPI) and beta-series correlation (BSC) studies, nearly half relied only on standard head motion regression or did not report a denoising strategy. Here, we introduce TMFC_denoise, an SPM-based toolbox with a graphical interface for applying denoising procedures to task-based activation and TMFC analyses. The toolbox updates first-level general linear models with nuisance regressors, including head motion expansions, spike regressors, physiological and global signals, anatomical component-based noise correction (aCompCor), and weighted constant terms for robust weighted least-squares regression (rWLS). It also provides quality-control (QC) measures, including framewise displacement (FD), derivative of root mean square variance over voxels (DVARS), FD-DVARS correlations, and task-FD/task-DVARS correlations. Using an event-related motor task, we benchmarked 11 denoising pipelines across generalized PPI, BSC, and BSC with beta scrubbing. The results show that more aggressive denoising does not necessarily improve TMFC estimates; effective denoising requires balancing artifact reduction with preservation of TMFC effects. TMFC_denoise facilitates transparent implementation, evaluation, and reporting of denoising strategies in TMFC research.