TABLE 3.

Overview of major findings and recommendations

Denoising framework
	Recommendations	Benefits
Filtering/regression/censoring	Use a linear regression model to perform all steps simultaneously	Better control for nuisance‐related variability (see Hallquist et al., 2013) Better control for residual nominal tDoF Provides an upper limit to the number of censored volumes
Treatment of multiple epochs within the same functional run	Avoid splitting the functional run in epochs	Denoising the whole run reduces the number of confounding variables, increasing network identifiability metrics
Nuisance mask creation	Extract masks from high‐resolution segmentation maps using conservative probability thresholds and multiple erosion cycles	Prevents contamination from gray matter voxels (see Power et al., 2017). Otherwise, the extracted signals might behave like GSR

Pipeline evaluation
Pipelines	Strengths	Weaknesses	Recommendations
GSR (e.g., 24RP + 8WM&CSF + GSR)	The most effective strategy for balancing motion‐related effects across functional conditions	GSR remains controversial
Censoring (24RP + 8WM&CSF + GSR + T/P‐cens)	The best approach for controlling distance dependent artifacts	Reduced network identifiability metrics, especially with P‐censoring T‐censoring is prone to introduce additional biases	If possible, exclude high‐moving subjects (see Parkes et al., 2018) Distance‐dependent artifacts can also be controlled with lenient thresholds (FDjenk >0.2; see figure S13)
aCompCor50% (24RP+ aCompCor50%)	Best non‐GSR based pipeline	It might overfit the data, depending on the number of observations.	Use the preorthogonalization procedure to increase the noise prediction power and to reduce the number of extracted components (see Figure S11)
aCompCor (24RP+ aCompCor)	Lower number of consumed tDoF compared to aCompCor50%		Use the preorthogonalization procedure to increase the noise prediction power (see Figure S11)
ICA‐AROMA	Good control of motion‐related artifacts No direct regression of motion parameters Nonaggressive denoising	Depending on the number of observations, it might require a considerable number of tDoF	In long multiple‐condition experiment, evaluate the possibility of performing ICA‐AROMA in each epoch separately in order to reduce the number of noise‐classified components (see Figure S12)
RP (RP12, RP24)	Effective in combination with other strategies	It might remove true signals covarying with head motion	Prefer 24RP over 12RP