Fig 1.
Temporal interpolation reduces motion estimates.
All traces are for a single subject of the ME cohort, derived from the second-echo image (TE = 28 ms). At top, position estimates derived from AFNI’s 3dVolreg using heptic interpolation and an early-run reference volume. The corresponding motion (FD) trace is shown immediately below in red, and below that motion derived in the same data after 4 possible orderings of temporal interpolation. At bottom, a trace of the fraction of the brain despiked by 3dDespike (as it operated on raw data to produce the DS data). See S1 Fig for more parameter spaces in the same data (e.g., different TEs, reference volumes, software packages used to estimate motion, etc.). See S2 Fig for the same effects after using other software packages to perform temporal interpolation.
Fig 2.
Despiking alters brain position.
For a subject of the WU cohort, motion estimates and despike traces are shown at top. Slices of the raw and despiked image are shown spanning a large motion. There is a positive pitch (nose up) rotation at volumes 64 and 65 in the raw data. This motion is attenuated in the despiked data.
Fig 3.
Temporal interpolation tends to reduce estimated brain motion in fMRI scans.
Data for 5 cohorts are shown. At left, mean motion in a scan is shown for each subject estimated both before and after temporal interpolation. At right, each volume’s motion is shown in a similar manner with a blue smoothing curve over 2000 points. Raw FD forms the x-axis; the y-axis is the ratio of post-temporal-interpolation motion relative to the raw FD value. Inset p values are from paired t-tests of raw versus post-interpolation mean FD values.
Fig 4.
Volumes with motion are most easily identified in raw FD.
For the GSP and WU cohorts, scrubbing analyses in raw data are shown using several versions of FD traces at several thresholds. Note that the slope of the red points increases from left to right, indicating that greater amounts of distance-dependent artifact are being characterized at the thresholds approach the “floor values” of the FD trace. Similar data with the same effects are shown for all cohorts in S5 Fig and distributions of each cohort’s FD under various preprocessing schemes are shown in S4 Fig. The x-axis represents distance, spanning 0–180 mm.
Fig 5.
Temporal interpolation reduces distance dependent artifact.
For the NIH and ABIDE cohorts, scrubbing analyses in raw, DS, DS+TS, and TS data are shown using several versions of FD traces at a 0.2 mm threshold. From left to right, the volumes being censored are not changing, all that changes is the data undergoing censoring. Note that the slope of the red points decreases in DS and DS+TS data, indicating that smaller amounts of distance-dependent artifact are present in these volumes. Similar data with the same effects are shown for all cohorts in S6 Fig. The same result is seen regardless of threshold used (data not shown). The x-axis represents distance, spanning 0–180 mm.
Fig 6.
QC:RSFC correlations are not systematically changed across processing strategies.
For the GSP and NIH cohorts, QC:RSFC analyses using various FD traces in data from various stages of processing are shown. There is no systematic influence of processing in these cohorts or in the other cohorts (shown in S7 Fig). The x-axis represents distance, spanning 0–180 mm.
Fig 7.
The meaning of image “distortion”. Consider an image acquired in interleaved, ascending fashion.
When motion occurs (here, affecting slice acquisitions 7, 8, and 9), the shape of the brain is altered. This shape is often referred to as distorted, but distorted does not mean “warped”. Rather, distortion means that certain parts of the brain are never scanned during that timepoint, and other parts of the brain are scanned twice. There is no way to recover the missing information.