Data inspections
We generated a variety of images and movies that provide a comprehensive look at the quality of the NSD data and pre-processing results.
In the various inspections, note that "sess00" corresponds to the prffloc scan session. Also, note that inspections are included even for the held-out data (now released). For example, for subj01, sess38-sess40 are the 3 held-out NSD scan sessions (now released), sess41 is the nsdsynthetic scan session, and sess42 is the nsdimagery scan session. As another example, for subj08, sess28-sess30 are the 3 held-out NSD scan sessions (now released).
Same as Supplementary Video 10 of the data paper. This shows the group-average b3 noise ceiling results on a rotating brain.
Same as Supplementary Video 1 of the data paper. These show the various modalities collected on the NSD subjects (T1, T2, EPI, TOF, SWI). The figures show the end-result of pre-processing and are all in the common anatomical space set by the T1 volume.
nsddata/inspections/flattenedsurfaces/
Screenshots showing where cuts were made to the fsaverage surface and each individual NSD subject’s surface in order to allow flattening of the cortical surfaces. Cuts were placed in approximately consistent locations across subjects.
Same as Supplementary Video 3 of the data paper. This movie shows the results of curvature-based fsaverage alignment for the NSD subjects.
nsddata/inspections/functionaltostructural/
These images show, for each NSD subject, the alignment achieved between the EPI data and the T2 anatomical volume. One result is shown for an affine transformation, and another result is shown for the nonlinear ANTS transformation. The ANTS transformation is used for the prepared NSD data.
nsddata/inspections/gradunwarp/
Sample figures illustrating the size of the gradient nonlinearity effect at the two different scanners used (3T and 7T).
Same as Supplementary Videos 6-9 in the data paper. These movies show the stability of the mean EPI and of BOLD signal strength across all scan sessions for all subjects.
nsddata/inspections/HRT2/
Figures illustrating the alignment achieved for the high-resolution T2 volume acquired for each NSD subject. The figures include the small box used for alignment (mask); the high-res T2 volume masked by this box (masked); the full high-res T2 volume (raw); manually defined MTL labels (rawlabels); resliced volume from the T2 anatomy masked by the box (T2matched_masked); and the full resliced volume from the T2 anatomy (T2matched).
Same as Supplementary Video 4. This shows the quality of the volume-based nonlinear MNI alignment.
nsddata/inspections/motioninspections*
At-a-glance inspection of all motion parameter estimates for all subjects in all sessions.
nsddata/inspections/physioinspections*
At-a-glance inspection of physiological data for all subjects in all sessions.
This movie shows an inspection of the overall stability of the pre-processed fMRI data. For each subject, we show 100 volumes randomly picked over time (all runs, all scan sessions).
nsddata/inspections/rawdatamovies/
Same as Supplementary Video 5. For one example run (NSD session 10, run 6), we show movies of both the raw fMRI volumes and the pre-processed fMRI volumes for each NSD subject.
nsddata/inspections/rois/
A variety of visualizations of the ROIs provided with the NSD dataset.
nsddata/inspections/sessionwise/
For each NSD subject, we inspect the mean EPI and ON-OFF R^2 results in each scan session (1.8-mm preparation).
nsddata/inspections/subjectmontages/
At-a-glance inspection of all NSD subjects' cortical surface reconstructions, including how well they align to the fsaverage template.
nsddata/inspections/surfaceinspections/
Same as Supplementary Video 2 from the data paper. These movies show 3 views of each subjects' T1 anatomy. The obtained FreeSurfer cortical surface reconstructions are indicated.
nsddata/inspections/surfacevisualizations/
A variety of different surface maps visualizing different aspects of the NSD dataset.
bNnc - These are noise ceilings computed for the different beta versions (b1, b2, b3). The noise ceilings reflect the case of 3 trials being averaged together. The color range is [0 75] with a jet colormap.
b3R2 - these are GLM R^2 values for the b3 GLM model. The values range from 0% to 100%. For display, the values are divided by 100, square-rooted, and then visualized using a range of [0 1] with a hot colormap.
[corticalsulc,HCP_MMP1,Kastner2015,nsdgeneral,streams] - These are visualizations of various ROI collections defined on fsaverage.
curvature - These are visualizations of binarized curvature values.
mean - These are visualizations of the mean fMRI signal using a range of [0 2000] with a gray colormap.
probmap - These are visualizations of the fraction of subjects that have a given ROI at each fsaverage vertex. The range is [0 1] with a copper colormap. Values that are equal to 0 are thresholded away.
R2 - These are R^2 values from the simple ON-OFF GLM model fitted to the NSD data. Values range from 0 to 100%. For display, values are divided by 100, square-rooted, and then visualized using a range of [0 1] with a hot colormap.
signaldropout - These are regions deemed to suffer from signal dropout (using methods described in the data paper). The maps are binary for each subject. The group result is the average of binary values across subjects. All are visualized using [0 1] with a winter colormap.
surfaceimperfections - These are regions deemed to suffer from cortical surface reconstruction errors (as described in the data paper). The maps are binary for each subject. The group result is the average of binary values across subjects. All are visualized using [0 1] with a winter colormap.
valid - These are indications of which vertices have valid data during the NSD experiment. The binary values are averaged across sessions conducted for a given subject. The color range is [0 1] with a jet colormap.
Note that "fsaverageflat" refers to a flattened version of the fsaverage surface, whereas "subjNNflat" refers to a flattened version of subject NN's native surface.