Sources and implications of whole-brain fMRI signals in humans
Jonathan Power, Mark Plitt, Tim Laumann, Alex Martin
Neuroimage 2016 (in press)
Figures and files (xlsx, pptx)
Below are movies associated with the article "Sources and implications of whole-brain fMRI signals in humans". All movies are 1080p and will look best at full-screen resolutions. The movies stream from YouTube and are organized as playlists. You can select cohort or subject from the playlist drop down menu at upper left of the movies (the three horizontal bars). Click on links next to the captions to download the movies. Files to download are hosted on Dropbox Pro; if the links don't work our traffic has exceeded its 200GB/day limit and the links will be re-enabled the following day.
Video 1 (.movs - 1.7GB):
This video shows FreeSurfer-derived tissue compartments for each subject of each cohort. The underlays shown on the first frame are the atlas-registered MP-RAGE (1 mm isotropic), the atlas-registered MP-RAGE sampled to EPI resolution (3 mm isotropic), and the atlas-registered EPI data (3 mm isotropic; mean intensity over time is shown). The overlays are masks derived from FreeSurfer's aparc+aseg.mgz image. The 3 blues represent gray matter: the cortical ribbon, subcortical structures, and the cerebellum. Green represents the white matter. Yellow represents the ventricles. In order to characterize signals specific to the white matter and ventricles (i.e., not due to gray matter), several erosions of these masks are used in the paper and are presented in several frames (0-4 erosions at 1 mm resolution, followed by nearest-neighbor resampling to EPI resolution). The next-to-final frame per subject show our preferred nuisance masks for 3 mm isotropic EPI data: white matter masks after 4 erosions, ventricle masks after 2 erosions, and no erosion of gray matter masks. The final frame shows the signal compartments illustrated in the paper: the spectrum of greens represents (light to dark) 0-1 erosions, 2-3 erosions, and 4 erosions and the two yellows (light to dark) represent 0-1 erosions and 2 erosions. Note that approximately half or more white matter voxels fall within 0-1 erosions of gray matter, as do most ventricle voxels. Note also that in the final frame the FreeSurfer-derived masks are themselves masked according to the mean EPI intensity (voxels with < 20% the modal EPI value are removed since they were either not scanned or are in locations of signal dropout).
Video 2 (.movs - 770MB)
This video shows the signals found in various parts of the brain for each subject of each cohort. At top, for reference, is a trace of motion (FD). At middle, all voxels within the brain are shown, organized by the tissue compartments shown in Video 1. Signals are always scaled -20 to 20 in mode 1000 units (i.e., -2% to 2%). All gray matter voxels are shown with a blue bar at the side, and all white matter and ventricle voxels have green or yellow bars at the side. A bright green line divides gray matter voxels from all other voxels (which are usually considered nuisance signals). The global signal is the average of all these signals. At bottom, the mean signal within each compartment is shown in the first frames of each subject. Colors correspond to the compartments above. The black trace is the global signal (the gray trace is an alternate computation of the global signal after standardizing all voxel signals). In the final frames for each subject the traces are replaced by slices of GSCORR (top) and the mean EPI intensity over time (bottom). GSCORR is the Pearson correlation of the global fMRI signal to each voxel signal. The GSCORR heat map is scaled from r = 0-1.
Video 3a (.mov - 195MB)
This video shows the physiological signals obtained in the NIH cohort at a fine temporal scale. Pulse oximeter traces are shown at top, along with detected (and vetted) peaks in the signal, from which the instantaneous heart rate and pulse amplitude are calculated. Traces from the respiratory belt strapped about the abdomen are at bottom. Pulse oximeter traces were examined in detail by the authors to identify signal abnormalities (unusual frequency content, unusual beat-to-beat intervals or heart rate changes) so that the heart rate is believed to consist of (only) accurate measurements. Pulse amplitude was calculated as the difference between signal at vetted peaks (heart beats) and the signal 0.14 seconds prior to the peaks (7 data points at 50 Hz).
This video is helpful 1) for checking that pulse-ox peaks were identified correctly and that calculated heart rate changes reflect the data; 2) for observing the modulation of heart rate and pulse pressure with respiratory cycles; 3) for observing the variability in signal quality and physiology across subjects and time.
Video 3b (.mov - 35MB)
This video shows the physiological signals obtained in the NIH cohort at a coarser, scan-length scale. In green and blue are the raw pulse oximeter and respiratory belt traces. In green points the instantaneous heart rate determined in Video 3a is shown, along with dimmed offset curves that downsample the heart rate to 1- and 2-TR intervals (TR = 3.5 sec in these data). In gray points is the peak amplitude, which often reflects arterial pulse pressure, derived from the difference between gray and green points in Video 3a. Dimmed offset curves reflect 1- and 2-TR samplings of the pulse pressure. The heart rate traces have been checked thoroughly by eye (Video 3a) and are believed to be accurate; the pulse pressure traces do contain some noise due to signal abnormalities, but the noise does not appear systematic and is averaged away in the downsampled traces. The dimmed traces below the respiratory trace reflect RVT (respiration volume per unit time, calculated by AFNI's retrots.m), and RV (calculated after Chang et al., 2009), both of which summarize the volume of air breathed per time. Because the effects of respiration on the BOLD signal include substantial delayed effects, a respiratory response function (RRF) was derived in these data from obvious large breaths in 4 subjects. The derived RRF matches well the RRF reported in Birn et al., 2008 and Chang et al., 2009 (see Supplemental Figure S4). The RVT and RV traces are replaced by those traces convolved with the RRF in the second frame of each subject. In black, the global fMRI signal is shown, and in red, head motion is shown in an FD trace. At bottom, all in-brain voxels are shown as in prior Figures and Videos.
Video 4 (.mov - 550MB for 4a-c NIH; .mov 4c for WU, ABIDE, GSP, and RP - 1.1GB)
Video 4a-c illustrates the changes produced in fMRI signals by various denoising strategies in the NIH cohort. Video 4a shows single strategies involving physiological records (which are only applicable to the NIH cohort since other cohorts lack physiological records). Video 4b shows single strategies that can be applied to any dataset, such as regressing white matter and ventricle signals and motion parameters. Video 4c combines strategies.
In all videos, the upper panel shows the 6 motion parameters and the FD trace of head motion for the scan. The middle panel shows the same physiological traces shown in Video 3b. The bottom panel shows all in-brain voxels as in Video 2. When a regression is performed, a post-regression global signal is plotted in gray over the original black global signal trace in the middle panel. And, if respiratory-related regressors are involved, these too are plotted. Only a single RVT regressor is shown, but 5 lagged versions of the regressor exist and are used. Similarly only a single RV regressor is shown, but 3 lagged regressors exist.
Video 5 (.mov - 20MB): Video 4c but for HCP data, before and after FIX ICA. The subjects of the "40 Unrelated Subjects" are shown.
Bonus (.mov - 50MB): GSCORR in 6.5 hours of RP data, analyzed at 3 mm isotropic resolution, paired with 1 mm isotropic MP-RAGE