Project a volume onto the cortical surface of a subject using its own (native) space.

Samples a brain volume at the positions of the cortical surface vertices of a subject and returns the resulting per-vertex values. This is the equivalent of the FreeSurfer command line tool mri_vol2surf (or of the project_vol2surf function of the Python package yabplot), implemented in R. The volume and the surface must be defined in the same coordinate space, which is the case for the volumes in the mri directory of a subject (e.g., brain.mgz) and the surfaces in its surf directory (e.g., lh.white) – both are in the subject's native space.

The volume is sampled at the vertex positions using trilinear interpolation (default) or nearest neighbor interpolation. Vertices which fall outside the volume (e.g., medial wall vertices at the bottom of a brainmask volume, or vertices outside a truncated field of view) get the value NA by default, and a warning is emitted. Use clamp = TRUE to instead use the value of the closest voxel at the volume border, and check_fov = FALSE to suppress the check.

To project a group-level statistical map (e.g., in MNI space) onto a template like fsaverage, see template.vol2surf.

Usage

subject.vol2surf(
  subjects_dir,
  subject_id,
  volume,
  surface = "white",
  hemi = "both",
  surface_frac = NULL,
  frac_surface = "pial",
  interpolation = "trilinear",
  frame = 1L,
  cortex_only = FALSE,
  clamp = FALSE,
  check_fov = TRUE,
  affine = NULL,
  vox2ras = "auto"
)

Arguments

subjects_dir: character string. The FreeSurfer SUBJECTS_DIR, i.e., a directory containing the data for all your subjects, each in a subdir named after the subject identifier.
subject_id: character string. The subject identifier.
volume: character string or named list. A volume filepath, the name of a volume file in the mri directory of the subject without file extension (e.g., brain or aseg), or an in-memory volume given as a named list with entries 'data' (numeric 3D or 4D array) and 'affine' (4x4 numeric matrix mapping 0-based voxel indices to world coordinates). Examples for the latter are read.fs.mgh (use with_header = TRUE and pass list(data = vol$data, affine = vol$header$vox2ras_matrix)) and nibabel in Python.
surface: character string, the name of the surface to sample. Examples: 'white', 'pial', 'midthickness'. The corresponding surf/?h.<surface> files must exist for the subject.
hemi: character string, one of 'lh', 'rh' or 'both'. The hemisphere to project to.
surface_frac: numeric scalar in range 0..1, or NULL (the default). If given, the volume is not sampled at the vertices of surface, but at the vertices of the surface which lies at the given fraction on the line between the corresponding vertices of surface (fraction 0) and frac_surface (fraction 1). E.g., with surface = "white", frac_surface = "pial" and surface_frac = 0.5, the mid-cortical surface is used. This is what the --surf-frac option of mri_vol2surf does. Note that no surface normals are required, the interpolation is purely geometric.
frac_surface: character string, the second surface for surface_frac. Ignored if surface_frac is NULL. Defaults to 'pial'.
interpolation: character string, one of 'trilinear' (the default, an alias 'linear' is accepted) or 'nearest'. Trilinear interpolation is suitable for continuous data like thickness or t-statistics, while nearest neighbor is what you want for discrete data like atlas labels, segmentation indices or p-values (it never averages the values of neighboring voxels).
frame: positive integer scalar, the index of the volume ('frame') to use if volume contains more than one. Defaults to 1. Ignored for 3D volumes.
cortex_only: logical, whether to set the values of all vertices which are not part of the cortex (as defined by the label file label/?h.cortex.label) to NA. This masks the medial wall. Defaults to FALSE.
clamp: logical. How to handle surface vertices which are outside the volume: if FALSE (the default), their value is set to NA; if TRUE, the value of the closest voxel at the volume border is used instead (edge replication, like the mode = 'nearest' parameter of scipy.ndimage.map_coordinates). See also check_fov.
check_fov: logical, whether to check for surface vertices which are outside the volume and emit a warning listing them. Defaults to TRUE. The check is very cheap (a few comparisons per vertex), the parameter exists mainly so that you can silence the warning when you know that a large part of the surface is outside the volume on purpose.
affine: numeric 4x4 matrix or NULL (the default). The matrix that maps 0-based voxel indices to world coordinates, i.e., the coordinate space in which the surface vertices are defined. By default this is derived from the volume (see vox2ras). Use this parameter only if you want to override the transformation completely.
vox2ras: character string, how the transformation from voxel indices to the coordinate space of the surface is obtained. One of 'auto' (the default), 'tkr' or 'header'. For MGH/MGZ volumes, 'auto' uses the FreeSurfer tkregister convention, in which the brain surfaces are defined: the direction cosines and voxel sizes from the volume header, but with the origin moved to the center of the volume (see vox2ras_tkr and vol.tkreg.affine). This matches what the FreeSurfer tools do (it is the difference between the --vox2ras-tkr and the --vox2ras option of mri_vol2surf), and it matters for volumes whose header carries a non-zero center-of-RAS offset (cras) – such volumes are common, e.g., when the data was imported from DICOM. For NIfTI volumes, 'auto' uses the transformation matrix from the file header (sform/qform), as there is no tkregister convention for NIfTI files. Use 'header' or 'tkr' to force one of the two interpretations. Ignored if affine is given.

Value

a numerical vector of per-vertex values, one value per surface vertex, or a hemilist (named list with entries lh and rh) of such vectors if hemi is 'both'. Note that the values are in the order of the vertices of the sampled surface, which is the order used by the surface file, so they can be used directly with functions like vis.subject.morph.native.

Note

Trilinear interpolation is used by default, while the FreeSurfer tool mri_vol2surf defaults to nearest neighbor interpolation: pass interpolation = "nearest" to reproduce its output exactly. Note also that the affine matrix of a NIfTI file is converted into an MGH-style header by the reader, which stores the direction cosines with limited precision: for volumes with an oblique (rotated) orientation the transformation can differ from the value in the file by about 1e-07, and sheared transformations are not supported (the reader warns about this).

Examples