Writing FreeSurfer neuroimaging data with freesurferformats

In this document, we show how to write brain imaging data to FreeSurfer binary files.

Writing FreeSurfer neuroimaging data with freesurferformats

Writing 1D morphometry data or other per-vertex information (in MGH, MGZ or curv format)

Morphometry data, or vertex-wise measures, are data that usually describe a measure like cortical thickness or surface area over the cortex. There is one scalar value per vertex of the brain surface mesh. Of course, you could write whatever you want (p-values, effect sizes at the vertex, …), as long as the data is scalar.

The package provides the write.fs.morph function to write any scalar data that does not require metadata like MR acquisition parameters or transforms. With this function, the format gets determined automatically from the file name. In the following example, we load the area and thickness values for a subject and write the product of area and thickness (which is not cortical volume, by the way) to new files in MGH, MGZ and curv format.

Let’s first load the data:

    library("freesurferformats");
    area = read.fs.morph(system.file("extdata", "lh.thickness", package = "freesurferformats", mustWork = TRUE));
    thickness = read.fs.morph(system.file("extdata", "lh.area.gz", package = "freesurferformats", mustWork = TRUE));
    mymorphdata = area * thickness;

Now we could write our derived data like this:

    format1 = write.fs.morph(tempfile(fileext = "mgz"), mymorphdata);
    format2 = write.fs.morph(tempfile(fileext = "mgh"), mymorphdata);
    format3 = write.fs.morph(tempfile(fileext = "curv"), mymorphdata);

Writing MGH and MGZ format files with full control over the header

When writing 3D or 4D images, the voxels alone are sometimes not enough: depending on what you intend to do with the data later on, you may want to save metadata like MR acquisition parameters or vox2ras transformation matrices in the file header. This is possible with the write.fs.mgh function, that gives you full control over the MGH header. The two relevant pieces of header data are:

  • the vox2ras_matrix: a 4x4 double matrix that encodes the affine transformation from voxel indices to x, y, z coordinates in space
  • the mr_params: a double vector of length 4 that contains the following values (in this order):
    • tr: repetition time in ms
    • flipangle: flip angle in radians
    • te: echo time in ms
    • ti: inversion time in ms

Here is an example that writes a file in MGH format including custom header data:

    mgh_outfile = "mystudy/subject1/mri/shifted_brain.mgz"
    data = array(data=rep(1L, 256*256*256), dim=c(256,256,256)); # not exactly a brain, but will do.
    mr_params = c(2300, 0.1, 2., 900.)
    vox2ras_matrix = matrix(c(-1,0,0,0,  0,0,-1,0,  0,1,0,0,  127.5,-98.6273,79.0953,1.000), nrow=4, byrow = FALSE)
    write.fs.mgh(mgh_outfile, data, vox2ras_matrix=vox2ras_matrix, mr_params=mr_params);

Note that if and only if you provide a ras2vox_matrix, the ras_good flag will be set to TRUE in the file header. If you do not provide mr_params, they default to all zero.

MGH data types

Data can be stored in the following data types in MGH files:

  • ‘MRI_UCHAR’ : 1 byte unsigned integer
  • ‘MRI_SHORT’ : 2 byte signed integer
  • ‘MRI_INT’ : 4 byte signed integer
  • ‘MRI_FLOAT’ : 4 byte signed floating point

One can control the data type that is used through the dtype parameter. The default value, ‘auto’, will determine the MRI data type from the R data type of the data parameter. The following rules are applied for R data types:

  • logical : the data gets coerced to integer, it is written as ‘MRI_UCHAR’
  • integer : written as ‘MRI_INT’ (no matter the range of values)
  • double : written as ‘MRI_FLOAT’

You can manually set the ‘dtype’ parameter to force a certain MRI data type. E.g., you could force ‘MRI_SHORT’ for your integer data if you are sure that the supported value range of that data type is enough for the values contained in your data. Be aware though that chosing an unsuitable MRI data type for your data will lead to an MGH file with incorrect data.

Hint: If you need to save data in a certain data type, cast your data before passing them to the write.fs.mgh function. E.g., if you have integers but want to store them as ‘MRI_FLOAT’, do the following:

    some_surface_mask = rep(1L, 163842);    
    some_surface_mask[30000:45000] = 0L;
    write.fs.mgh("regionmask_stored_as_MRI_FLOAT.mgh", as.double(some_surface_mask));

Notice: That was an example only. For a label, storing the indices as MRI_FLOAT makes little sense (you could use as.logical() instead to store them as as ‘MRI_UCHAR’).

Writing ‘curv’ format files (morphometry data)

You can use write.fs.curv to write arbitrary data in binary curv format. The result is identical to using write.fs.morph with any filename that does not end in mgh or mgz.)

    data = rnorm(120000, 2.0, 1.0);
    curvfile = "mystudy/subject1/surf/lh.random"
    write.fs.curv(curvfile, data);

It’s worth knowing the if your filename ends with .gz, the file will be written compressed.

Writing surface format files

You can use write.fs.surface to write triangular meshes in binary surface format (the format used for files like ‘surf/lh.white’ or ‘surf/rh.pial’). A mesh is defined by a list of vertices and a list of faces as indices into the latter, knows as an indexed face-set.

    vertices = matrix(rep(0.3, 15), nrow=3);     # 5 vertices
    faces = matrix(c(1L,2L,3L,2L,4L,3L,4L,5L,3L), nrow=3, byrow = TRUE);   # 3 faces

    write.fs.surface(tempfile(fileext="white"), vertices, faces);

The vertex indices used to define the faces should be 1-based, as used in R. They will be written 0-based to the file.

You can also use this function to write meshes in FreeSurfer ASCII format (.asc) or in VTK ASCII format (.vtk), see the parameter format for details.

Writing label files

Labels can be written with the write.fs.label function. A label is nothing but a list of vertex indices. These indices should be passed 1-based, as they are used in R.

output_file = tempfile();

# generate data
vertex_indices = seq(from = 10000, to=20000);

# write label to file
write.fs.label(output_file, vertex_indices);

You can attach a single floating point value to each vertex of the label, see the parameter ‘vertex_data’ for details.

Writing color lookup table (LUT) files in ASCII format

The following example uses the write.fs.colortable function to write a colortable:

colortable_df = data.frame("struct_index"=c(0, 1), "struct_name"=c("struct1", "struct2"),
                  "r"=c(80, 100), "g"=c(50, 40), "b"=c(250, 200), "a"=c(0, 0), stringsAsFactors = FALSE);

output_file = tempfile(fileext = ".txt");
write.fs.colortable(output_file, colortable_df);

A colortable is a table that contains one structure per row, and assigns each structure an RGBA color. Colortables are typically used to color the different brain regions of the segmented brain volume. The file ‘FreeSurferColorLUT.txt’ that comes with FreeSurfer is an example for a colortable.