The efficacy of resting-state fMRI denoising pipelines for motion correction and behavioural prediction

Resting-state functional magnetic resonance imaging (rs-fMRI) is a pivotal tool for mapping the functional organization of the brain and its relation to individual differences in behaviour. One challenge for the field is that rs-fMRI signals are contaminated by multiple sources of noise that can contaminate these rs-fMRI signals, affecting the reliability and validity of any derivative phenotypes and attenuating their correlations with behaviour. Here, we investigate the efficacy of different noise mitigation pipelines, including white matter and cerebrospinal fluid regression, independent component analysis (ICA)-based artefact removal, volume censoring, global signal regression (GSR), and diffuse cluster estimation and regression (DiCER), in simultaneously achieving two objectives: mitigating motion-related artefacts and augmenting brain-behaviour associations. Our analysis, which employed three distinct quality control metrics to evaluate motion influence and a kernel ridge regression for behavioural predictions of 81 different behavioural variables across two independent datasets, revealed that no single pipeline universally excels at achieving both objectives consistently across different cohorts. Pipelines combining ICA-FIX and GSR demonstrate a reasonable trade-off between motion reduction and behavioural prediction performance, but inter-pipeline variations in predictive performance are modest.