The fsbrain software is designed to be used with the output of FreeSurfer and similar neuroimaging software packages. Running FreeSurfer’s recon-all on your T1w MRI scan results in a directory structure full of different files and file types for each subject. The fsbrain library uses knowledge on this directory layout to load the proper data.
However, while designed primarily with FreeSurfer in mind, fsbrain is not limited to FreeSurfer output, see below.
The fsbrain library uses freesurferformats to load a variety of neuroimaging file formats, including data exchange formats used by other brain imaging software. See the freesurferformats website for the full list.
You can use freesurferformats directly to load the data, then pass it to fsbrain. See the next question for an example.
Yes, the computational
anatomy toolbox (CAT12) for SPM writes surfaces in
GIFTI format and the morphometry data in curv format, both formats are
supported by fsbrain. After running CAT12 surface measure
computation on your subject
subject1, you should have the
following files in the surf/ subdir:
Try the following to visualize the gyrification data for the left hemisphere in fsbrain:
You should be able to load data from a number of different neuroimaging software packages with freesurferformats, as it supports the very common NIFTI and GIFTI file formats.
One should use the
vis.subject.pre function for that.
Here is a full example that loads both the surfaces and the per-vertex
data into R first and the creates a plot based on this pre-loaded data,
without accessing any files anymore:
library("fsbrain"); library("freesurferformats"); # Manually load some data from wherever you want. You can also compute this data in R of course, without loading anything. my_data_dir="~/data/study1/subject1"; # just an example. surf_lh <- read.fs.surface(file.path(my_data_dir, "lh_sphere.surf.gii")); surf_rh <- read.fs.surface(file.path(my_data_dir, "whatever/rh_sphere.obj")); pvd_lh = read.fs.morph(file.path(my_data_dir, "lh.curv")); # pvd is for 'per-vertex data'. pvd_rh = read.fs.morph(file.path(my_data_dir, "surf/converted/rh.mean_curvature.mgh")); # Call high-level API for live plot. surfaces = hemilist(surf_lh, surf_rh); pvd = hemilist(pvd_lh, pvd_rh); cm = vis.subject.pre(surfaces, pvd , draw_colorbar = T, rglactions = list('trans_fun'=limit_fun(-0.2, 0.2))); # Export if you feel like it export(cm, output_img = "~/out.png", grid_like = FALSE, colorbar_legend="Mean curvature [1/mm]");
Yes, you can install the BrainSuite
bssr R package to
read DFS files and then visualize them. In the following examples, the
~/data/brainsuite_subject1/ contains the output
of the cortical surface extraction sequence from BrainSuite (I used
version 19b), applied to the fsbrain demo subject ‘subject1’.
If the DFS surface file contains vertex colors, labels and/or morphometry data, you can display that as well:
# load and turn into fs.surface bd_thickness = bssr::readdfs('~/data/brainsuite_subject1/subject1_v1_sMRI.pvc-thickness_0-6mm.right.mid.cortex.dfs'); sfb_thickness = list('vertices'=t(bd_thickness$vertices), faces=t(bd_thickness$faces)+1L); class(sfb_thickness) = c(class(sfb_thickness), 'fs.surface'); # Display the vertex colors: fsbrain::vis.fs.surface(sfb_thickness, col=rgb(t(bd_thickness$vColor))); # Display labels (feel free to change the colormap). fsbrain::vis.fs.surface(sfb_thickness, per_vertex_data = bd_thickness$labels); # Display raw morphometry data (feel free to change the colormap). fsbrain::vis.fs.surface(sfb_thickness, per_vertex_data = bd_thickness$attributes);
bssr package is not on CRAN (as of September
2020), you will have to install it from the BrainSuite website.
To increase the output resolution, you need to increase the size of the rgl rendering device. To do this globally, before you call any fsbrain rendering function:
Alternatively, you can control the size when calling an
fsbrain visualization function by passing the same information
in the optional
rgloptions parameter, like this:
The 4 values in the ‘windowRect’ vector are the x and y screen positions (in pixels) where to position the upper left corner of the rendering window on your desktop, and the width and height of the plot/window.
Note that fsbrain renders images, which means the output is pixel-based (i.e., bitmap as opposed to vector graphics). To get high quality output, you need to increase the size of the rgl rendering device, as explained in the last question.
To save the plot to a file in PNG format, you can use an rglaction:
This opens the plot in a window as usual and also saves it in PNG format to the file subject1_thickness.png in your home directory.
While one could convert the PNGs to these formats, doing so makes little sense. The images produced by fsbrain are bitmap images, not vector graphics. I would recommend to export the figures in high resolution as PNG images instead.
The default settings used by the the export() function should be fine to reach the 300 DPI typically required for publications, unless you want the image to fill the entire page. In that case, you can further increase the output resolution.
If a journal requires you to submit vector graphics files (PDF, SVG, EPS, …), you should simply embed the bitmap in a vector format container and submit that. E.g., in the free vector graphics software Inkscape, you could do this with the following steps (takes less than a minute from the 2nd time you do it):
File --> Newif needed.
File --> Import...and select the PNG image. In the new
png bitmap image importwindow, the default settings should be fine (
Image Import Type: Embed,
Image DPI: From File,
Image Rendering Mode: None (Auto)), just click
File --> Document Properties, and on the
Custom Size, click on
Fit page to contentto expand it, then make sure the margins are all set to 0 (the default), and click the button
Resize page to drawing or selection.
You can now save the image as SVG, PDF, EPS or whatever by clicking
File --> Save as.... I would recommend EPS format if you
are using LateX, and PDF if you just need to upload the image in any
vector format to the publisher’s website. Make sure to check the result
in a standard vector graphics viewer like Adobe Acrobat before
This will of course not lead to a true vector graphics file that can
be scaled/zoomed in without losing quality, but that is not what they
are asking you to submit. They just want a vector container. I would not
recommend to try to vectorize (or
trace) the output of
fsbrain to generate a true vector image as the quality will usually be
bad, but you are of course free to try.
If you have data with very few but extreme outliers, almost all of
your plot will have a single color. This happens for example when
plotting curvature data. You can of course first load the data using
subject.morph.native, adjust it (transform it, remove
outliers), and then plot it using
In many cases, it is easier to use
rglactions to clip
the data to the 5th to 95th percentile, which can be done with an
Yes, this can be achieved in different ways:
The first option is to use
rglactions in combination
This will limit your data to the range 2 to 3.
If you want the data values outside the given range to be plotted as
background (in the color for
NA values), use
limit_fun_na instead of
See the answer to the next question for a second option.
This typically means that you want a colorbar that shows a larger data range than the real data range of your subject, for some of the subjects. In this case you should use the ‘range’ entry in ‘makecmap_options’:
This is useful when plotting group data to ensure that all subjects use the same colors for identical values, or when comparing statistical results.
Pass a colormap function to any visualization function that supports the makecmap_options parameter, as entry colFn like illustrated below:
In that example, we used the popular viridis colormap. In R, it is available from the viridis package. If you don’t have it, you can install it with:
Of course, you can use any colormap function you want, currently the only limitation is that it should accept an integer parameter: the requested number of colors.
The exact number of colors that will be requested depends on your data, and if the colormap you want only supports very few colors, you can use a wrapper function to interpolate. Here is an example for the very popular RColorBrewer package. Some of its colormaps have less than 10 colors, which is usually not enough for neuroimaging data. Here we wrap the ‘Blues’ palette, which has 9 colors:
Choosing a colormap can be a surprisingly complex question and there are many publications which discuss this topic. You may want to consider what kind of data you have and what property you want to highlight, how many colors you need, whether you want colorblind-friendly colors, how they look when printed in gray-scale, whether they look pleasing to you, and maybe many more dimensions.
The most important thing is to decide is whether you need a sequential, qualitative, or diverging palette for your data.
I am definitely not an expert, but here are some color functions I personally like and use with fsbrain:
They require the viridis and RColorBrewer packages to be installed. The qualitative map is fine if you do not have more than 11 different values.
If you have R >= 3.6, you may not need any extra packages: have a
look at the
grDevices::hcl.colors function. Here are some
If you want a heatmap-style colormap (single hue sequential, yellow/red), try:
Make sure to read the next entry as well if you are using a diverging colormap.
When using a diverging colormap, make sure to set the symm option to makecmap_options when using a visualization function, like this:
This ensures that the neutral color of the diverging colormap (usually white) is aligned with the zero mark in the colorbar/legend, by adapting the value range displayed on the colorbar.
The impression that the numbers of colors in the colorbar is lower than in the rendered image is a consequence of the rendering process: the lighting (shadows, highlights) and the material properties (glossyness, partial transparency) have an effect on the appearance of colors in the rendered image.
You can set the parameter n in the makecmap_options (see above) to request more colors, which will lead to a smooth colorbar.
This also means that more colors are used in the rendered image, but the effect will be less noticable.
By default, NA values are rendered in white. You can change this using col.na in makecmap_options:
This is also useful when plotting clusters, just set all values which are not part of any cluster to NA.
Yes, see the answer to the next question for details.
Yes, use the
vis.colortable.legend function. You can
pass an annotation or a color lookup table, and it will create a plot
that shows the colors and the structure (or region) names. The output
will be a separate plot, so you can use standard R methods to save it in
vector formats like PDF for best quality.
Hint: you can load a color lookup table with
While this is not possible in
rgl, fsbrain provides the
vislayout.from.coloredmeshes function to achieve this using
Image Magick. You need to have the suggested ‘magick’ package installed
for this to work. The function renders separate images, crops the output
figures to remove the background, then merges the seperate cropped
images into a final output image and saves it as a PNG file. Here is a
Note that your output resolution settings (see question above) now count for each of the single images. This means that you will get quite high resolution output in combination with the tight layout. This makes the function ideal for producing plots for publications.
You can adjust various settings, e.g., change the rendering style, select different views, and save it to a custom file name in your home directory:
It is also possible to plot a separate colorbar image and combine
that with the tight layout brainview image. Note that the settings for
the colorbar are stored in the coloredmeshes, and can be adjusted by
altering the initial call to
whatever visualization function you use) above.
output_cbar_img = "fsbrain_cbar.png"; output_final_img = "fsbrain_merged.png"; coloredmesh.plot.colorbar.separate(cm, image.plot_extra_options = list('horizontal' = TRUE), png_options = list('filename'=output_cbar_img, 'width'=1800)); combine.colorbar.with.brainview.image(output_brain_img, output_cbar_img, output_final_img);
You may have to play a bit with the resolution settings of your brain images and the colorbar to get this right (the background cropping makes it hard to compute the exact values in advance).
You should also have a look at the new
function, which produces an image with colorbar and provides sane
defaults for the resolution and suitable text sizes, etc.
To the best of my knowledge, RGL cannot produce transparent images (though you can of course render semi-transparent surfaces in the scene). However, since exporting PNGs with a transparent background is very useful for usage in presentations, we provide a hacked solution based on post-processing the output images with the magick package in R.
This required fsbrain version >= 0.4.3 and is implemented in the
export() function via the
Here is how it works: the
transparency_color is used as
a background color for rendering in RGL. During the fsbrain
post-processing, all pixels with this color in the image will be
replaced with transparency.
This means that: * The
transparency_color has to be an
RGB color, i.e., it cannot have an alpha value (if it is an RGBA color,
the alpha part is silently ignored by RGL it seems). * The
transparency_color can be almost any RGB color, but (1) it
should not occur in the brain surface visualization to prevent
transparent parts in the brain after post-processing. And (2) because of
anti-aliasing, the color may slightly affect the border pixels of the
rendered brain surfaces, so a neutral color like ‘#FFFFFF’ (for white)
or some gray value like ‘#DDDDDD’ is maybe more suitable than a hot
pink. At high resolution, this is hardly noticable though.
This happens due to the inflation. One can use the
rglactions parameter to push the hemis apart. By default,
it pushes them apart by the amount they overlap:
You can also set a distance manually:
If you need more customization options, have a look at the
shift.hemis.apart function. If you pass a named list in the
rglactions for ‘shift_hemis_apart’ (like in the previous example), the
entries are passed on to that function.
There is no ‘view’ setting for that yet (like ‘t4’ for 4 tiles), but you can set the ‘grid_like’ to false to get a horizontal strip of images. Here is an example:
subjects_dir = fsbrain::get_optional_data_filepath("subjects_dir"); view_angles = get.view.angle.names(angle_set = "t4"); coloredmeshes = vis.subject.morph.native(subjects_dir, "subject1", "thickness", views=NULL); vislayout.from.coloredmeshes(coloredmeshes, view_angles = view_angles, grid_like=FALSE, output_img="~/fsbrain_horizontal.png");
If you want a colorbar, you can use
export instead of
now recommended in any case.
Yes, see the example notebook files in the directory web of the fsbrain repository. The Rmd files are actually notebooks in R markdown format.
Yes, fsbrain is not limited to brain surface meshes, and a
wide array of mesh file formats are supported. Keep in mind though that
fsbrain works with triangular meshes. To visualize a mesh from a file,
the easiest way is to use
The full error most likely looks like this, or similar:
If you experience this error, it most likely happened when you tried to plot something with a colorbar, and did not increase the device size (image resolution). The error occurs if there is no space for the colorbar plot, and the solution is to increase the resolution as explained in the answer to the question ‘How can I set the output image resolution?’ above.
This can happen when visualizing morphometry data with function like
vis.subject.morph.native. The full error most likely looks
like this, or similar:
Most likely you have
Inf values in your data, most
likely in the medial wall, just try plotting without it:
If this does not help, load the data and inspect it manually.
The full message most likely looks like this:
This happens if you do not have X11, or no window can be opened. Possible reasons include that you are running R on a remote host using an SSH connection without X11 forwarding, or that you do not have XQuartz installed under MacOS.
If you do not want to plot (e.g., you only use fsbrain to load data), feel free to ignore this warning.
If you want to plot and cannot install X11 on a machine (e.g., a compute node of a high performance computing cluster), you can still plot on the box using xvfb or similar means, see the next question.
You can achieve this by using a virtual display server, e.g. by running R/fsbrain via xvfb under Linux. An example script that illustrates how to do this for fsbrain can be found in the fsbrain_xvfb_demo directory of the fsbrain repo on GitHub.
The output dimensions are limited by your screen resolution, so to be able to do something like this:
you would need to have a screen with a resolution larger than 3000x3000 pixels.
A technical note: This limitation exists due to the fact that rgl renders to an OpenGL window produced by your display server, and fsbrain basically takes a screenshot of that window. One can circumvent this limitation by using a virtual display server, e.g. by running R/fsbrain via xvfb under Linux. An example script that illustrates how to do this for fsbrain can be found in the fsbrain_xvfb_demo directory of the fsbrain repo on GitHub.
If you install the development version from github and try to load it without restarting R, you ma^y get this error:
To fix this, simply restart R.
So you get nothing at all after running a vis command, like this:
Then you can try to open the visualization in an rgl widget like this:
This should show the brain plot in rstudio.
Thanks to Clancy-wu for reporting this workaround.