-
Notifications
You must be signed in to change notification settings - Fork 1
/
NOTES
58 lines (41 loc) · 3.17 KB
/
NOTES
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
NONE: No nuisance regression performed
S: Interpolation of time points where framewise displacement was greater than 0.5 mm
FIX: Non-aggressive regression of automatically identified (as artifacts) independent components
RP24: 24 motion parameter regression
WM1: Mean white-matter signal regression
CSF1: Mean CSF signal regression
CC: ACompCor (N PCA components) over combined CSF + WM mask. The number N of components is chosen as to explain 25% of the variance within the mask.
RP9: 6 motion parameters + WM1 + CSF1 + average whole-brain signal
FIX_CC: Aggressive removal or artifactual independent components + ACompCor [See Note 2 below]
_D: DiCER [Diffuse Cluster Estimation and Regression] procedure to remove widespread artifact. This is an alternative to global signal regression [See Note 1].
===========================================
[NOTE 1]
From Smith et al. NeuroImage 2013, HCP
"Preliminary analyses of measures of motion-related artefacts indicate that the ICA-FIX process greatly reduces but does not, in some datasets, totally eliminate motion artefacts
that are frame-specific and non-spatially-specific.
In coming months we will investigate further whether there is value in additional cleanup stages,
most likely to be added into the pipeline after the ICA+FIX denoising. One approach that is simple but effective is “motion scrubbing”, #
in which one identifies the timepoints that are “irreversibly” damaged by motion, and simply excises those from the timeseries analysis (Power et al., 2011).
We will evaluate this and other approaches, and where appropriate, make improvements to the temporal pre-processing pipeline."
[NOTE 2]
# From Parkes et al. 2018
Taken together, the above results allow us to draw six key conclusions.
First, Head Motion Parameters + Phys models without GSR are ineffective at
mitigating motion-related artefact from rs-fMRI,
regardless of the level of motion in the sample,
exclusionary criteria applied, or the use of expansion terms.
The application of GSR dramatically improves the performance of these pipelines.
Second, aCompCor pipelines may only be viable in low-motion datasets and perform poorly in high motion data (see Fig. 1).
Third, ICA-AROMA and censoring pipelines are superior to the other denoising strategies,
yielding the lowest QC-FC correlations (see Fig. 1, Fig. 4), lowest QC-FC distance-dependence (see Fig. 2, Fig. 4),
and minimal functional connectivity differences between high- and low-motion healthy controls (see Fig. 6).
Fourth, a major part of the benefit to motion-related artefact control in censoring pipelines comes
from the exclusion of participants with <4 min of uncensored data;
when this criterion is applied to all pipelines,
performance differences are marginal (except for the HMP pipelines without GSR).
Fifth, aCompCor and censoring pipelines yield high tDOF-loss,
marking them as relatively expensive methods for controlling
for motion-related artefact in rs-fMRI data.
Finally, methods that were more effective at denoising were associated
with reduced test-retest reliability,
suggesting that noise signals in BOLD data are reproducible.