ROIs
The NSD dataset comes with a variety of regions of interest (ROIs). Some ROIs are derived from atlases and are automatically determined, whereas other ROIs reflect manual definition based on data from each subject.
Surface-derived ROIs
Some ROIs are generated from surface-based representations of the data. These ROIs include:
HCP_MMP1 is the Glasser et al., Nature, 2016 atlas.
Kastner2015 is the Wang et al., Cerebral Cortex, 2015 atlas.
nsdgeneral is a general ROI that was manually drawn on fsaverage covering voxels responsive to the NSD experiment in the posterior aspect of cortex.
corticalsulc is a folding-based atlas defined based on the curvature of fsaverage (sulci, gyri). It labels major sulci and some gyri throughout the whole cortex.
streams is an anatomical atlas that labels various “streams” in visual cortex. It is largely based on fsaverage folding but also takes into account the b3 noise ceiling results to ensure that the regions generally cover where there are stimulus-related signals. More details are provided below.
prf-visualrois is a collection of manually drawn ROIs based on results of the prf experiment. These ROIs consist of V1v, V1d, V2v, V2d, V3v, V3d, and hV4. These ROIs extend from the fovea (0° eccentricity) to peripheral cortical regions that still exhibit sensible signals in the prf experiment given the limited stimulus size (this means up to about ~5-6° eccentricity).
prf-eccrois is a collection of manually drawn ROIs that cover the exact same cortical extent as the prf-visualrois ROIs. These ROIs consist of ecc0pt5, ecc1, ecc2, ecc4, and ecc4+, and indicate increasing “concentric” ROIs that cover up to 0.5°, 1°, 2°, 4°, and >4° eccentricity.
floc-faces is a collection of manually drawn ROIs based on results of the floc experiment. These ROIs consist of OFA, FFA-1, FFA-2, mTL-faces ("mid temporal lobe faces"), and aTL-faces ("anterior temporal lobe faces"). These ROIs were the result of (liberal) thresholding at t > 0 (flocfacestval).
floc-words is a collection of manually drawn ROIs based on results of the floc experiment. These ROIs consist of OWFA, VWFA-1, VWFA-2, mfs-words ("mid fusiform sulcus words"), and mTL-words ("mid temporal lobe words"). These ROIs were the result of (liberal) thresholding at t > 0 (flocwordtval).
floc-places is a collection of manually drawn ROIs based on results of the floc experiment. These ROIs consist of OPA, PPA, and RSC. These ROIs were the result of (liberal) thresholding at t > 0 (flocplacestval).
floc-bodies is a collection of manually drawn ROIs based on results of the floc experiment. These ROIs consist of EBA, FBA-1, FBA-2, and mTL-bodies ("mid temporal lobe bodies"). These ROIs were the result of (liberal) thresholding at t > 0 (flocbodiestval).
Note that for the floc-faces, floc-words, and floc-bodies ROIs, not all subjects have all of these ROIs in every hemisphere.
Please note that the floc-related ROIs are quite liberal (given the threshold of t > 0) and will look quite "large" relative to what one may be typically used to. It is a good idea to carefully visualize the ROIs; you can easily whittle down the ROIs using a more stringent threshold if you desire.
The ROIs listed above are initially defined in surface space. For convenience, we have also created volumetric versions of the ROIs. Values of -1 indicate non-cortical voxels in the case of ROIs in volume format. Values of 0 indicate non-labeled vertices/voxels. Positive integers indicate labelings for vertices/voxels.
When surface-based ROIs are converted to volume format, there is an implicit parameter that controls the spatial extent of the volume version. We attempted to create volume ROIs that are not too liberal and not too conservative.
Note that although surface-defined ROIs in floc-faces, floc-words, floc-places, and floc-bodies are guaranteed to be t > 0, after conversion to volume space, this constraint may not be entirely still true. If you use the volume versions, you may want to consider further shrinking down these ROIs.
Volume-derived ROIs
Some ROIs are generated from volume-based representations of the data. These ROIs include:
thalamus provides manual segmentation of thalamic regions: LGN, SC, and pulvinar (several subdivisions). Regions were defined in each hemisphere by an expert. Definition was based mostly on T1 anatomical data, but for the pulvinar, MNI-based results from other datasets were projected to each subject to aid ROI definition. Note that as a matter of definition, the ventral pulvinar is most correlated with early visual cortex; the dorsal lateral pulvinar is most correlated with the attention network; and the dorsal medial pulvinar is most correlated with the default-mode network. Additional information: LGN and SC were defined based on T1 and T2 image contrast. For the ventral pulvinar, the extent of the pulvinar was defined based on T1 and T2 contrast and then constrained to the ventral lateral portion based on the extent of the two ventral pulvinar maps reported in Arcaro et al., Journal of Neuroscience, 2015. The dorsolateral pulvinar was based on the average correlation with IPS maps; and the dorsomedial pulvinar was based on average correlation with precuneus (as reported in Arcaro et al. Nature Communications 2018).
MTL provides manual segmentation of various regions in the medial temporal lobe, including hippocampal subfields. A expert human annotator used the raw high-resolution T2 volumes and manually segmented regions according to Berron et al., NeuroImage Clinical, 2017 for each of the 8 NSD subjects. These ROI labelings were then co-registered to the official isotropic T2 volume space and processed.
Also, note that ROI labels are mutually exclusive across hemispheres (i.e. every voxel is either assigned to the left hemisphere, right hemisphere, or neither).
ROI files
For convenience, ROI files have been prepared in multiple spaces. ROI files are available in functional spaces (func1pt8mm, func1mm) as well as anatomical spaces (anat). For ROIs in anatomical space, we provide ROIs at 0.8-mm anatomical resolution. ROI files are also available in surface space (FreeSurfer .mgz).
nsddata/ppdata/subjAA/*/roi/[lh,rh].EEE.nii.gz
These are volumes providing integer labels for ROI EEE, generated separately for each hemisphere.
nsddata/ppdata/subjAA/*/roi/EEE.nii.gz
These are volumes providing integer labels for ROI EEE, combining across hemispheres.
nsddata/ppdata/subjAA/anat/roi/other/*.nii.gz
The thalamus and MTL segmentations are originally drawn at 0.5-mm; for completeness, we provide here the original anat0pt5 version of these segmentations, as well as an anat1pt0 version of these segmentations.
nsddata/freesurfer/*/label/[lh,rh].EEE.mgz
These are surface files providing integer labels for ROI EEE.
nsddata/freesurfer/*/label/EEE.mgz.{ctab,txt}
This is a text file specifying the meaning of the labels in surface-based ROI EEE. (The same labels apply to volume versions of the ROI.) In other words, this is the critical text file that tells you what each of the integer labels means in terms of ROI names.
prf-visualrois:
V1d,V1v,V2d,V2v,V3d,V3v - dorsal and ventral subdivisions of V1, V2, and V3
hV4 - human V4
prf-eccrois:
ecc0pt5,ecc1,ecc2,ecc4,ecc4+ - eccentricity-restricted regions within early visual areas V1, V2, and V3
floc-faces:
OFA - occipital face area
FFA-1 - posterior section of fusiform face area
FFA-2 - anterior section of fusiform face area
mTL-faces - face-selective region in middle portion of temporal lobe
aTL-faces - face-selective region in anterior portion of temporal lobe
floc-words:
OVWFA - occipital visual word form area
VWFA-1 - posterior section of visual word form area
VWFA-2 - anterior section of visual word form area
mfs-words - word-selective region located near the mid-fusiform sulcus
mTL-words - word-selective region in middle portion of temporal lobe
floc-places:
OPA - occipital place area
PPA - parahippocampal place area
RSC - retrospenial cortex (place-selective)
floc-bodies:
EBA - extrastriate body area (can also be referred to as LOTC-bodies (lateral occipitotemporal cortex))
FBA-1 - posterior section of fusiform body area (can also be referred to as VOTC-bodies-1 (ventral occipitotemporal cortex))
FBA-2 - anterior section of fusiform body area (can also be referred to as VOTC-bodies-2)
mTL-bodies - body-selective region in middle portion of temporal lobe
nsddata/templates/EEE.ctab
This is a text file specifying the meaning of the labels in volume-based ROI EEE. For example, EEE can be "thalamus" or "MTL".
Probmap files
For convenience, we also create "probmap" (probabilistic map) results. Specifically, we take each manually defined cortical ROI and map these via nearest-neighbor interpolation to fsaverage and then compute the fraction of subjects at each vertex that has each individual ROI present.
nsddata/freesurfer/fsaverage/label/[lh.rh].RRR.mgz
This file consists of fractions between 0 and 1. The value indicates the fraction of subjects that have ROI RRR present at a given fsaverage vertex.
Other files
The following files are intermediate files created in the process of the manual segmentation of the MTL ROI collection.
nsddata/ppdata/subjAA/anat/HRT2/HRT2_raw.nii.gz
The raw high-resolution T2 volume used for MTL segmentation.
nsddata/ppdata/subjAA/anat/HRT2/HRT2_mask.nii.gz
The binary mask within which an affine transformation was optimized to match the official 0.5-mm T2 volume.
nsddata/ppdata/subjAA/anat/HRT2/MTL_rawlabels.nii.gz
The manually defined MTL labels (same space as the HRT2_raw.nii.gz volume).
nsddata/ppdata/subjAA/anat/HRT2/T2matched.nii.gz
Given the affine transformation determined (within the HRT2_mask), this volume is the result of reslicing through the 0.5-mm T2 volume to match the HRT2_raw volume (cubic interpolation).
Additional information on the streams ROIs
Early visual cortex ROI:
The early visual cortex ROI was drawn as the union of the V1v, V1d, V2v, V2d, V3v and V3d ROIs from the Wang 2015 retinotopic atlas. Additionally, V2v and V2d were connected such that the part of the occipital pole typically containing foveal representations was also included. The same was repeated for V3v and V3d.
Intermediate ROIs:
Three intermediate ROIs were drawn corresponding to each of the three streams: ventral, lateral and parietal. All three ROIs border the early visual cortex ROI on the posterior side.
The intermediate ventral ROI was drawn to reflect the inferior boundary of hV4 from the Wang atlas and to include the inferior occipital gyrus (IOG), with the anterior border of the ROI drawn based on the anterior edge of the inferior occipital sulcus (IOS).
The intermediate lateral ROI was drawn directly superior to the intermediate ventral ROI, with the superior and anterior borders determined as the LO1 and LO2 boundaries from the Wang atlas.
The intermediate parietal ROI was drawn directly superior to that, reflecting exactly the borders of the union of V3A and V3B from the Wang atlas.
Higher-level ROIs:
Three higher-level ROIs were drawn for each of the ventral, lateral and parietal streams, bordering their respective intermediate ROIs on their posterior edges.
The ventral ROI was drawn to follow the anterior lingual sulcus (ALS), including the anterior lingual gyrus (ALG) on its inferior border and to follow the inferior lip of the inferior temporal sulcus (ITS) on its superior border. The anterior border was drawn based on the midpoint of the occipital temporal sulcus (OTS).
The lateral ROI was drawn such that the higher-level ventral ROI was its inferior border and the superior lip of the superior temporal sulcus (STS) was used to mark the anterior/superior boundary. The rest of the superior boundary traced the edge of angular gyrus, up to the tip of the posterior STS (pSTS).
The parietal ROI was drawn to reflect the boundary of the lateral ROI on its inferior edge and to otherwise trace the borders of and include the union of IPS0, IPS1, IPS2, IPS3, IPS4, IPS5 and SPL1 from the Wang atlas.