Modular preprocessing pipelines can reintroduce artifacts into fMRI data

Abstract

The preprocessing pipelines typically used in both task and resting‐state functional magnetic resonance imaging (rs‐fMRI) analysis are modular in nature: They are composed of a number of separate filtering/regression steps, including removal of head motion covariates and band‐pass filtering, performed sequentially and in a flexible order. In this article, we illustrate the shortcomings of this approach, as we show how later preprocessing steps can reintroduce artifacts previously removed from the data in prior preprocessing steps. We show that each regression step is a geometric projection of data onto a subspace, and that performing a sequence of projections can move the data into subspaces no longer orthogonal to those previously removed, reintroducing signal related to nuisance covariates. Thus, linear filtering operations are not commutative, and the order in which the preprocessing steps are performed is critical. These issues can arise in practice when any combination of standard preprocessing steps including motion regression, scrubbing, component‐based correction, physiological correction, global signal regression, and temporal filtering are performed sequentially. In this work, we focus primarily on rs‐fMRI. We illustrate the problem both theoretically and empirically through application to a test–retest rs‐fMRI data set, and suggest remedies. These include (a) combining all steps into a single linear filter, or (b) sequential orthogonalization of covariates/linear filters performed in series.

Measuring shared responses across subjects using intersubject correlation

“…Nuisance variables and temporal filters should be combined into a single regression model to ensure that noise signals are not injected back into the data and to properly estimate degrees of freedom (Lindquist et al, 2019). If performing a volumetric analysis, ISCs should be computed within a brain or gray matter mask to reduce the number of subsequent tests (surface-based analyses preclude this step).…”

High-amplitude co-fluctuations in cortical activity drive functional connectivity

“…The mean BOLD signal for each cortical node data was linearly detrended, band-pass filtered (0.008-0.08 Hz) [48], confound regressed and standardized using Nilearn's signal.clean, which removes confounds orthogonally to the temporal filters [52]. The confound regression employed [53] included 6 motion estimates, time series of the mean CSF, mean WM, and mean global signal, the derivatives of these nine regressors, and the squares these 18 terms.…”

Edge-centric functional network representations of human cerebral cortex reveal overlapping system-level architecture

“…The mean BOLD signal for each cortical node data was linearly detrended, band-pass filtered (0.008-0.08 Hz) [105], confound regressed and standardized using Nilearn's signal.clean, which removes confounds orthogonally to the temporal filters [108]. The confound regression employed [109] included 6 motion estimates, time series of the mean CSF, mean WM, and mean global signal, the derivatives of these nine regressors, and the squares these 18 terms.…”